Datasets:

VietMedTeam
/

veri-render

	{"sample_id": "sample_00051", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1900\nsize = 44\ncolormap = \"magma\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.554388 * np.exp(\n -((X - 0.610856)*2 / (2 1.0371982)\n + (Y - 1.517410)2 / (2 * 0.6797812))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1900\nsize = 44\ncolormap = \"magma\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.554388 * np.exp(\n -((X - 0.610856)*2 / (2 1.0371982)\n + (Y - 1.517410)2 / (2 * 0.679781**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1900}
	{"sample_id": "sample_00052", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1901\nsize = 27\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.909467 * np.exp(\n -((X - 0.009287)*2 / (2 1.2776282)\n + (Y - 2.744406)2 / (2 * 0.629802*2))\n)\nZ += 0.714654 np.exp(\n -((X - -0.036967)*2 / (2 0.7523102)\n + (Y - 2.230718)2 / (2 * 1.069237*2))\n)\nZ += -1.857357 np.exp(\n -((X - -0.940575)*2 / (2 0.5041852)\n + (Y - 0.917902)2 / (2 * 0.5084482))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1901\nsize = 27\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.909467 * np.exp(\n -((X - 0.009287)*2 / (2 1.2776282)\n + (Y - 2.744406)2 / (2 * 0.629802*2))\n)\nZ += 0.714654 np.exp(\n -((X - -0.036967)*2 / (2 0.7523102)\n + (Y - 2.230718)2 / (2 * 1.069237*2))\n)\nZ += -1.857357 np.exp(\n -((X - -0.940575)*2 / (2 0.5041852)\n + (Y - 0.917902)2 / (2 * 0.508448**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1901}
	{"sample_id": "sample_00053", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1902\nsize = 46\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 1.398239 * np.exp(\n -((X - -0.751440)*2 / (2 0.8174552)\n + (Y - 0.277919)2 / (2 * 0.675317*2))\n)\nZ += -1.122808 np.exp(\n -((X - -0.091660)*2 / (2 0.5987332)\n + (Y - 1.789293)2 / (2 * 0.9083342))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1902\nsize = 46\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 1.398239 * np.exp(\n -((X - -0.751440)*2 / (2 0.8174552)\n + (Y - 0.277919)2 / (2 * 0.675317*2))\n)\nZ += -1.122808 np.exp(\n -((X - -0.091660)*2 / (2 0.5987332)\n + (Y - 1.789293)2 / (2 * 0.908334**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1902}
	{"sample_id": "sample_00054", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1903\nsize = 30\ncolormap = \"viridis\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.033551 * np.exp(\n -((X - -1.404482)*2 / (2 1.3392892)\n + (Y - 1.882034)2 / (2 * 1.391416*2))\n)\nZ += 1.047285 np.exp(\n -((X - 0.383779)*2 / (2 0.6351752)\n + (Y - 1.768766)2 / (2 * 1.2347612))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1903\nsize = 30\ncolormap = \"viridis\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.033551 * np.exp(\n -((X - -1.404482)*2 / (2 1.3392892)\n + (Y - 1.882034)2 / (2 * 1.391416*2))\n)\nZ += 1.047285 np.exp(\n -((X - 0.383779)*2 / (2 0.6351752)\n + (Y - 1.768766)2 / (2 * 1.234761**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1903}
	{"sample_id": "sample_00055", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1904\nsize = 25\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.254077 * np.exp(\n -((X - -1.025582)*2 / (2 0.9469832)\n + (Y - -0.990845)2 / (2 * 1.097465*2))\n)\nZ += 1.991914 np.exp(\n -((X - 0.660525)*2 / (2 1.0325352)\n + (Y - 1.696424)2 / (2 * 1.404641*2))\n)\nZ += -1.787233 np.exp(\n -((X - -1.538434)*2 / (2 0.9503882)\n + (Y - 2.198314)2 / (2 * 0.5733032))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1904\nsize = 25\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.254077 * np.exp(\n -((X - -1.025582)*2 / (2 0.9469832)\n + (Y - -0.990845)2 / (2 * 1.097465*2))\n)\nZ += 1.991914 np.exp(\n -((X - 0.660525)*2 / (2 1.0325352)\n + (Y - 1.696424)2 / (2 * 1.404641*2))\n)\nZ += -1.787233 np.exp(\n -((X - -1.538434)*2 / (2 0.9503882)\n + (Y - 2.198314)2 / (2 * 0.573303**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1904}
	{"sample_id": "sample_00056", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1905\nsize = 45\ncolormap = \"magma\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.252087 * np.exp(\n -((X - -1.716627)*2 / (2 1.4182812)\n + (Y - -0.529773)2 / (2 * 1.283903*2))\n)\nZ += 1.412459 np.exp(\n -((X - 1.529702)*2 / (2 1.4473622)\n + (Y - -0.880475)2 / (2 * 1.1678492))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1905\nsize = 45\ncolormap = \"magma\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.252087 * np.exp(\n -((X - -1.716627)*2 / (2 1.4182812)\n + (Y - -0.529773)2 / (2 * 1.283903*2))\n)\nZ += 1.412459 np.exp(\n -((X - 1.529702)*2 / (2 1.4473622)\n + (Y - -0.880475)2 / (2 * 1.167849**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1905}
	{"sample_id": "sample_00057", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1906\nsize = 35\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.478871 * np.exp(\n -((X - -0.752787)*2 / (2 1.1512012)\n + (Y - -0.425881)2 / (2 * 1.4443692))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1906\nsize = 35\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.478871 * np.exp(\n -((X - -0.752787)*2 / (2 1.1512012)\n + (Y - -0.425881)2 / (2 * 1.444369**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1906}
	{"sample_id": "sample_00058", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1907\nsize = 42\ncolormap = \"viridis\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 0.592669 * np.exp(\n -((X - 1.216654)*2 / (2 1.4172142)\n + (Y - -0.698115)2 / (2 * 1.244495*2))\n)\nZ += 1.383131 np.exp(\n -((X - 0.968027)*2 / (2 0.8641722)\n + (Y - 0.815769)2 / (2 * 1.395209*2))\n)\nZ += 1.402953 np.exp(\n -((X - 1.965273)*2 / (2 0.7792462)\n + (Y - 1.011683)2 / (2 * 1.0772742))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1907\nsize = 42\ncolormap = \"viridis\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 0.592669 * np.exp(\n -((X - 1.216654)*2 / (2 1.4172142)\n + (Y - -0.698115)2 / (2 * 1.244495*2))\n)\nZ += 1.383131 np.exp(\n -((X - 0.968027)*2 / (2 0.8641722)\n + (Y - 0.815769)2 / (2 * 1.395209*2))\n)\nZ += 1.402953 np.exp(\n -((X - 1.965273)*2 / (2 0.7792462)\n + (Y - 1.011683)2 / (2 * 1.077274**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1907}
	{"sample_id": "sample_00059", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1908\nsize = 26\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -0.672556 * np.exp(\n -((X - 0.812188)*2 / (2 0.5828402)\n + (Y - 1.030430)2 / (2 * 1.4148612))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1908\nsize = 26\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -0.672556 * np.exp(\n -((X - 0.812188)*2 / (2 0.5828402)\n + (Y - 1.030430)2 / (2 * 1.414861**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1908}
	{"sample_id": "sample_00060", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1909\nsize = 34\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.811477 * np.exp(\n -((X - 1.913968)*2 / (2 1.1239022)\n + (Y - 1.897390)2 / (2 * 1.075150*2))\n)\nZ += 1.836162 np.exp(\n -((X - 0.490576)*2 / (2 1.1390292)\n + (Y - 2.586806)2 / (2 * 0.8287082))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1909\nsize = 34\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.811477 * np.exp(\n -((X - 1.913968)*2 / (2 1.1239022)\n + (Y - 1.897390)2 / (2 * 1.075150*2))\n)\nZ += 1.836162 np.exp(\n -((X - 0.490576)*2 / (2 1.1390292)\n + (Y - 2.586806)2 / (2 * 0.828708**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1909}
	{"sample_id": "sample_00121", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2900\nn_points = 161\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.831950 * np.sin(1.118913 * x + 1.008837)\ny1 = 1.338358 * np.sin(0.932295 * x + 2.234393)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.83, f=1.12, φ=1.01\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.34, f=0.93, φ=2.23\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (2 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2900\nn_points = 161\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.831950 * np.sin(1.118913 * x + 1.008837)\ny1 = 1.338358 * np.sin(0.932295 * x + 2.234393)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.83, f=1.12, φ=1.01\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.34, f=0.93, φ=2.23\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (2 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2900}
	{"sample_id": "sample_00122", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2901\nn_points = 145\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.155640 * np.sin(1.302779 * x + 4.981734)\ny1 = 0.911540 * np.sin(2.947745 * x + 2.938028)\ny2 = 0.859045 * np.sin(1.454283 * x + 4.031508)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.16, f=1.30, φ=4.98\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.91, f=2.95, φ=2.94\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=0.86, f=1.45, φ=4.03\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2901\nn_points = 145\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.155640 * np.sin(1.302779 * x + 4.981734)\ny1 = 0.911540 * np.sin(2.947745 * x + 2.938028)\ny2 = 0.859045 * np.sin(1.454283 * x + 4.031508)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.16, f=1.30, φ=4.98\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.91, f=2.95, φ=2.94\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=0.86, f=1.45, φ=4.03\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2901}
	{"sample_id": "sample_00123", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2902\nn_points = 80\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.781911 * np.sin(1.543293 * x + 4.144060)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.78, f=1.54, φ=4.14\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2902\nn_points = 80\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.781911 * np.sin(1.543293 * x + 4.144060)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.78, f=1.54, φ=4.14\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2902}
	{"sample_id": "sample_00124", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2903\nn_points = 141\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.857967 * np.sin(0.942220 * x + 5.084951)\ny1 = 0.590013 * np.sin(1.142741 * x + 0.572998)\ny2 = 1.924395 * np.sin(1.682643 * x + 4.928356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.86, f=0.94, φ=5.08\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.59, f=1.14, φ=0.57\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.92, f=1.68, φ=4.93\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2903\nn_points = 141\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.857967 * np.sin(0.942220 * x + 5.084951)\ny1 = 0.590013 * np.sin(1.142741 * x + 0.572998)\ny2 = 1.924395 * np.sin(1.682643 * x + 4.928356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.86, f=0.94, φ=5.08\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.59, f=1.14, φ=0.57\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.92, f=1.68, φ=4.93\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2903}
	{"sample_id": "sample_00125", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2904\nn_points = 153\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.514431 * np.sin(1.513446 * x + 0.116435)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.51, f=1.51, φ=0.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2904\nn_points = 153\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.514431 * np.sin(1.513446 * x + 0.116435)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.51, f=1.51, φ=0.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2904}
	{"sample_id": "sample_00126", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2905\nn_points = 165\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.607619 * np.sin(1.667818 * x + 6.143726)\ny1 = 1.205378 * np.sin(1.973711 * x + 0.172039)\ny2 = 1.652390 * np.sin(2.122295 * x + 0.591268)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.61, f=1.67, φ=6.14\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.21, f=1.97, φ=0.17\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.65, f=2.12, φ=0.59\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2905\nn_points = 165\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.607619 * np.sin(1.667818 * x + 6.143726)\ny1 = 1.205378 * np.sin(1.973711 * x + 0.172039)\ny2 = 1.652390 * np.sin(2.122295 * x + 0.591268)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.61, f=1.67, φ=6.14\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.21, f=1.97, φ=0.17\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.65, f=2.12, φ=0.59\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2905}
	{"sample_id": "sample_00127", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2906\nn_points = 115\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.617971 * np.sin(1.020470 * x + 0.512607)\ny1 = 0.790376 * np.sin(0.861338 * x + 0.057956)\ny2 = 1.269849 * np.sin(1.617446 * x + 5.271811)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.62, f=1.02, φ=0.51\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.79, f=0.86, φ=0.06\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.27, f=1.62, φ=5.27\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2906\nn_points = 115\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.617971 * np.sin(1.020470 * x + 0.512607)\ny1 = 0.790376 * np.sin(0.861338 * x + 0.057956)\ny2 = 1.269849 * np.sin(1.617446 * x + 5.271811)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.62, f=1.02, φ=0.51\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.79, f=0.86, φ=0.06\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.27, f=1.62, φ=5.27\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2906}
	{"sample_id": "sample_00128", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2907\nn_points = 85\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.032634 * np.sin(2.206059 * x + 2.518333)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.03, f=2.21, φ=2.52\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2907\nn_points = 85\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.032634 * np.sin(2.206059 * x + 2.518333)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.03, f=2.21, φ=2.52\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2907}
	{"sample_id": "sample_00129", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2908\nn_points = 131\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.698043 * np.sin(2.606526 * x + 5.039795)\ny1 = 1.662470 * np.sin(1.724687 * x + 1.478118)\ny2 = 1.207486 * np.sin(2.091718 * x + 0.338356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.70, f=2.61, φ=5.04\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.66, f=1.72, φ=1.48\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.21, f=2.09, φ=0.34\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2908\nn_points = 131\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.698043 * np.sin(2.606526 * x + 5.039795)\ny1 = 1.662470 * np.sin(1.724687 * x + 1.478118)\ny2 = 1.207486 * np.sin(2.091718 * x + 0.338356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.70, f=2.61, φ=5.04\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.66, f=1.72, φ=1.48\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.21, f=2.09, φ=0.34\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2908}
	{"sample_id": "sample_00130", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2909\nn_points = 127\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.044005 * np.sin(1.288522 * x + 3.606051)\ny1 = 0.548443 * np.sin(2.583436 * x + 2.986435)\ny2 = 1.236664 * np.sin(1.559843 * x + 5.118033)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.04, f=1.29, φ=3.61\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.55, f=2.58, φ=2.99\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.24, f=1.56, φ=5.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2909\nn_points = 127\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.044005 * np.sin(1.288522 * x + 3.606051)\ny1 = 0.548443 * np.sin(2.583436 * x + 2.986435)\ny2 = 1.236664 * np.sin(1.559843 * x + 5.118033)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.04, f=1.29, φ=3.61\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.55, f=2.58, φ=2.99\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.24, f=1.56, φ=5.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2909}
	{"sample_id": "sample_00161", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3900\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.821660\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3900\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.821660\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3900}
	{"sample_id": "sample_00162", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3901\nn_petals = 4 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.977908\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3901\nn_petals = 4 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.977908\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3901}
	{"sample_id": "sample_00163", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3902\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.288979\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3902\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.288979\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3902}
	{"sample_id": "sample_00164", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3903\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.576224\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3903\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.576224\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3903}
	{"sample_id": "sample_00165", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3904\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.635304\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3904\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.635304\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3904}
	{"sample_id": "sample_00166", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3905\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.350061\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3905\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.350061\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3905}
	{"sample_id": "sample_00167", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3906\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.654892\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3906\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.654892\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3906}
	{"sample_id": "sample_00168", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3907\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.984689\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3907\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.984689\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3907}
	{"sample_id": "sample_00169", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3908\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.757216\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3908\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.757216\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3908}
	{"sample_id": "sample_00170", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3909\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.407149\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3909\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.407149\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3909}
	{"sample_id": "sample_00201", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4900\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.49, 1.40, 3.60, 1.51]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4900\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.49, 1.40, 3.60, 1.51]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4900}
	{"sample_id": "sample_00202", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4901\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [-1.89, -1.52, -2.96, -2.85, -1.38]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4901\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [-1.89, -1.52, -2.96, -2.85, -1.38]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4901}
	{"sample_id": "sample_00203", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4902\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [2.71, -2.88, 3.19, 2.36, 2.85]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4902\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [2.71, -2.88, 3.19, 2.36, 2.85]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4902}
	{"sample_id": "sample_00204", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4903\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [-2.93, 3.58, 4.65, 0.56]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4903\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [-2.93, 3.58, 4.65, 0.56]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4903}
	{"sample_id": "sample_00205", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4904\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [1.31, 3.65, 0.96, -1.43, 0.04]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4904\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [1.31, 3.65, 0.96, -1.43, 0.04]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4904}
	{"sample_id": "sample_00206", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4905\ncolor = \"steelblue\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [4.79, 4.99, -2.15, -2.90]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4905\ncolor = \"steelblue\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [4.79, 4.99, -2.15, -2.90]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4905}
	{"sample_id": "sample_00207", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4906\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [2.27, 3.88, -0.89, 0.48]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4906\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [2.27, 3.88, -0.89, 0.48]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4906}
	{"sample_id": "sample_00208", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4907\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E', 'F']\nheights = [0.40, -0.06, 0.04, -2.27, 2.11, -2.08]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (6 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4907\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E', 'F']\nheights = [0.40, -0.06, 0.04, -2.27, 2.11, -2.08]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (6 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4907}
	{"sample_id": "sample_00209", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4908\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [3.93, 3.38, -1.72, -2.13]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4908\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [3.93, 3.38, -1.72, -2.13]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4908}
	{"sample_id": "sample_00210", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4909\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.29, -0.41, 4.48, 4.96]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4909\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.29, -0.41, 4.48, 4.96]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4909}
	{"sample_id": "sample_00231", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5900\nn_points = 59\nslope = 1.149370\nintercept = 1.299066\nnoise_std = 0.539668\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5900\nn_points = 59\nslope = 1.149370\nintercept = 1.299066\nnoise_std = 0.539668\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5900}
	{"sample_id": "sample_00232", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5901\nn_points = 37\nslope = -1.855496\nintercept = -1.322886\nnoise_std = 0.441639\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5901\nn_points = 37\nslope = -1.855496\nintercept = -1.322886\nnoise_std = 0.441639\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5901}
	{"sample_id": "sample_00233", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5902\nn_points = 37\nslope = -1.529994\nintercept = -0.004273\nnoise_std = 0.730252\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5902\nn_points = 37\nslope = -1.529994\nintercept = -0.004273\nnoise_std = 0.730252\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5902}
	{"sample_id": "sample_00234", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5903\nn_points = 51\nslope = 1.440904\nintercept = 0.738086\nnoise_std = 0.352827\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5903\nn_points = 51\nslope = 1.440904\nintercept = 0.738086\nnoise_std = 0.352827\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5903}
	{"sample_id": "sample_00235", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5904\nn_points = 59\nslope = 0.890379\nintercept = 1.055611\nnoise_std = 0.559377\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5904\nn_points = 59\nslope = 0.890379\nintercept = 1.055611\nnoise_std = 0.559377\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5904}
	{"sample_id": "sample_00236", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5905\nn_points = 49\nslope = -1.929874\nintercept = -0.664004\nnoise_std = 0.672751\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5905\nn_points = 49\nslope = -1.929874\nintercept = -0.664004\nnoise_std = 0.672751\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5905}
	{"sample_id": "sample_00237", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5906\nn_points = 40\nslope = 1.016841\nintercept = 1.151654\nnoise_std = 0.678537\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5906\nn_points = 40\nslope = 1.016841\nintercept = 1.151654\nnoise_std = 0.678537\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5906}
	{"sample_id": "sample_00238", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5907\nn_points = 52\nslope = -0.891248\nintercept = 0.612783\nnoise_std = 0.538393\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5907\nn_points = 52\nslope = -0.891248\nintercept = 0.612783\nnoise_std = 0.538393\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5907}
	{"sample_id": "sample_00239", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5908\nn_points = 53\nslope = 1.044824\nintercept = 0.686445\nnoise_std = 0.641637\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5908\nn_points = 53\nslope = 1.044824\nintercept = 0.686445\nnoise_std = 0.641637\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5908}
	{"sample_id": "sample_00240", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5909\nn_points = 38\nslope = -0.918191\nintercept = -1.427642\nnoise_std = 0.533250\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5909\nn_points = 38\nslope = -0.918191\nintercept = -1.427642\nnoise_std = 0.533250\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5909}
	{"sample_id": "sample_00291", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.803\\,cos(1.790x + 1.502)\n\nx \\in [0, 2\\pi]\nn_points = 229\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.803\\,cos(1.790x + 1.502)\n\nx \\in [0, 2\\pi]\nn_points = 229", "generator_type": "latex", "seed": 6900}
	{"sample_id": "sample_00292", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.723\\,cos(1.810x + 0.773)\n\nx \\in [0, 2\\pi]\nn_points = 102\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.723\\,cos(1.810x + 0.773)\n\nx \\in [0, 2\\pi]\nn_points = 102", "generator_type": "latex", "seed": 6901}
	{"sample_id": "sample_00293", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.814\\,cos(1.098x + 1.407)\n\nx \\in [0, 2\\pi]\nn_points = 213\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.814\\,cos(1.098x + 1.407)\n\nx \\in [0, 2\\pi]\nn_points = 213", "generator_type": "latex", "seed": 6902}
	{"sample_id": "sample_00294", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.745\\,cos(2.896x + 1.774)\n\nx \\in [0, 2\\pi]\nn_points = 211\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.745\\,cos(2.896x + 1.774)\n\nx \\in [0, 2\\pi]\nn_points = 211", "generator_type": "latex", "seed": 6903}
	{"sample_id": "sample_00295", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.537\\,sin(1.486x + 0.274) + 0.899\\,cos(1.249x + 0.255)\n\nx \\in [0, 2\\pi]\nn_points = 219\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.537\\,sin(1.486x + 0.274) + 0.899\\,cos(1.249x + 0.255)\n\nx \\in [0, 2\\pi]\nn_points = 219", "generator_type": "latex", "seed": 6904}
	{"sample_id": "sample_00296", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.933\\,sin(1.359x + 3.084)\n\nx \\in [0, 2\\pi]\nn_points = 168\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.933\\,sin(1.359x + 3.084)\n\nx \\in [0, 2\\pi]\nn_points = 168", "generator_type": "latex", "seed": 6905}
	{"sample_id": "sample_00297", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.152\\,cos(2.279x + 1.723) + 1.637\\,sin(2.199x + 1.175)\n\nx \\in [0, 2\\pi]\nn_points = 189\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.152\\,cos(2.279x + 1.723) + 1.637\\,sin(2.199x + 1.175)\n\nx \\in [0, 2\\pi]\nn_points = 189", "generator_type": "latex", "seed": 6906}
	{"sample_id": "sample_00298", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.143\\,sin(1.289x + 1.056)\n\nx \\in [0, 2\\pi]\nn_points = 229\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.143\\,sin(1.289x + 1.056)\n\nx \\in [0, 2\\pi]\nn_points = 229", "generator_type": "latex", "seed": 6907}
	{"sample_id": "sample_00299", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.080\\,sin(1.208x + 0.887) + 1.355\\,cos(2.425x + 1.128)\n\nx \\in [0, 2\\pi]\nn_points = 169\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.080\\,sin(1.208x + 0.887) + 1.355\\,cos(2.425x + 1.128)\n\nx \\in [0, 2\\pi]\nn_points = 169", "generator_type": "latex", "seed": 6908}
	{"sample_id": "sample_00300", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.961\\,sin(2.271x + 1.308)\n\nx \\in [0, 2\\pi]\nn_points = 147\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.961\\,sin(2.271x + 1.308)\n\nx \\in [0, 2\\pi]\nn_points = 147", "generator_type": "latex", "seed": 6909}
	{"sample_id": "sample_00331", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.273x^3 - 0.686x^2 - 1.162x + 0.540\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.273x^3 - 0.686x^2 - 1.162x + 0.540\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7900}
	{"sample_id": "sample_00332", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.348x^2 - 0.103x + 1.950\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.348x^2 - 0.103x + 1.950\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7901}
	{"sample_id": "sample_00333", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.483x^2 + 1.711x - 0.297\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.483x^2 + 1.711x - 0.297\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7902}
	{"sample_id": "sample_00334", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.866x^3 - 1.746x^2 - 1.098x + 0.551\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.866x^3 - 1.746x^2 - 1.098x + 0.551\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7903}
	{"sample_id": "sample_00335", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.989x^2 - 0.364x - 1.721\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.989x^2 - 0.364x - 1.721\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7904}
	{"sample_id": "sample_00336", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.359x^4 - 1.104x^3 + 1.774x^2 + 1.846x + 0.056\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.359x^4 - 1.104x^3 + 1.774x^2 + 1.846x + 0.056\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7905}
	{"sample_id": "sample_00337", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.290x^4 - 0.794x^3 - 1.678x^2 + 1.818x - 1.313\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.290x^4 - 0.794x^3 - 1.678x^2 + 1.818x - 1.313\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7906}
	{"sample_id": "sample_00338", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.969x^3 + 1.223x^2 - 0.856x + 0.131\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.969x^3 + 1.223x^2 - 0.856x + 0.131\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7907}
	{"sample_id": "sample_00339", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.790x^2 + 0.281x - 0.190\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.790x^2 + 0.281x - 0.190\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7908}
	{"sample_id": "sample_00340", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.931x^2 - 1.875x - 0.749\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.931x^2 - 1.875x - 0.749\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7909}
	{"sample_id": "sample_00371", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.051 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.051 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8900}
	{"sample_id": "sample_00372", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.252 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.252 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8901}
	{"sample_id": "sample_00373", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.425 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.425 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8902}
	{"sample_id": "sample_00374", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.501 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.501 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8903}
	{"sample_id": "sample_00375", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.474 \\left\|\\cos\\left(\\frac{3\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.474 \\left\|\\cos\\left(\\frac{3\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8904}
	{"sample_id": "sample_00376", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 0.922 \\left\|\\cos\\left(\\frac{4\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 0.922 \\left\|\\cos\\left(\\frac{4\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8905}
	{"sample_id": "sample_00377", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.956 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.956 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8906}
	{"sample_id": "sample_00378", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.363 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.363 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8907}
	{"sample_id": "sample_00379", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.059 \\left\|\\cos\\left(\\frac{6\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.059 \\left\|\\cos\\left(\\frac{6\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8908}
	{"sample_id": "sample_00380", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 0.806 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 0.806 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8909}
	{"sample_id": "sample_00411", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 12.688 m/s\n θ = 41.74°\n g = 9.800 m/s²\n t ∈ [0, 1.689 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 12.688 m/s\n θ = 41.74°\n g = 9.800 m/s²\n t ∈ [0, 1.689 s]", "generator_type": "physics_spec", "seed": 9900}
	{"sample_id": "sample_00412", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 8.125 m/s\n θ = 59.16°\n g = 9.800 m/s²\n t ∈ [0, 1.313 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 8.125 m/s\n θ = 59.16°\n g = 9.800 m/s²\n t ∈ [0, 1.313 s]", "generator_type": "physics_spec", "seed": 9901}
	{"sample_id": "sample_00413", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 14.264 m/s\n θ = 63.54°\n g = 9.800 m/s²\n t ∈ [0, 2.324 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 14.264 m/s\n θ = 63.54°\n g = 9.800 m/s²\n t ∈ [0, 2.324 s]", "generator_type": "physics_spec", "seed": 9902}
	{"sample_id": "sample_00414", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.164 m/s\n θ = 57.42°\n g = 9.800 m/s²\n t ∈ [0, 1.996 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.164 m/s\n θ = 57.42°\n g = 9.800 m/s²\n t ∈ [0, 1.996 s]", "generator_type": "physics_spec", "seed": 9903}
	{"sample_id": "sample_00415", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 16.119 m/s\n θ = 60.54°\n g = 9.800 m/s²\n t ∈ [0, 2.680 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 16.119 m/s\n θ = 60.54°\n g = 9.800 m/s²\n t ∈ [0, 2.680 s]", "generator_type": "physics_spec", "seed": 9904}
	{"sample_id": "sample_00416", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.521 m/s\n θ = 44.11°\n g = 9.800 m/s²\n t ∈ [0, 1.871 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.521 m/s\n θ = 44.11°\n g = 9.800 m/s²\n t ∈ [0, 1.871 s]", "generator_type": "physics_spec", "seed": 9905}
	{"sample_id": "sample_00417", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 10.590 m/s\n θ = 61.69°\n g = 9.800 m/s²\n t ∈ [0, 1.739 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 10.590 m/s\n θ = 61.69°\n g = 9.800 m/s²\n t ∈ [0, 1.739 s]", "generator_type": "physics_spec", "seed": 9906}
	{"sample_id": "sample_00418", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 11.954 m/s\n θ = 35.80°\n g = 9.800 m/s²\n t ∈ [0, 1.424 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 11.954 m/s\n θ = 35.80°\n g = 9.800 m/s²\n t ∈ [0, 1.424 s]", "generator_type": "physics_spec", "seed": 9907}
	{"sample_id": "sample_00419", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.968 m/s\n θ = 38.95°\n g = 9.800 m/s²\n t ∈ [0, 1.526 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.968 m/s\n θ = 38.95°\n g = 9.800 m/s²\n t ∈ [0, 1.526 s]", "generator_type": "physics_spec", "seed": 9908}
	{"sample_id": "sample_00420", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 9.063 m/s\n θ = 33.45°\n g = 9.800 m/s²\n t ∈ [0, 1.018 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 9.063 m/s\n θ = 33.45°\n g = 9.800 m/s²\n t ∈ [0, 1.018 s]", "generator_type": "physics_spec", "seed": 9909}
	{"sample_id": "sample_00461", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.915\n ω = 2.096 rad/s\n φ = 2.739 rad\n t ∈ [0, 5.994 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.915\n ω = 2.096 rad/s\n φ = 2.739 rad\n t ∈ [0, 5.994 s]", "generator_type": "physics_spec", "seed": 10900}
	{"sample_id": "sample_00462", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.978\n ω = 0.941 rad/s\n φ = 2.212 rad\n t ∈ [0, 13.351 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.978\n ω = 0.941 rad/s\n φ = 2.212 rad\n t ∈ [0, 13.351 s]", "generator_type": "physics_spec", "seed": 10901}
	{"sample_id": "sample_00463", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.835\n ω = 1.141 rad/s\n φ = 3.038 rad\n t ∈ [0, 11.017 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.835\n ω = 1.141 rad/s\n φ = 3.038 rad\n t ∈ [0, 11.017 s]", "generator_type": "physics_spec", "seed": 10902}
	{"sample_id": "sample_00464", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.416\n ω = 1.601 rad/s\n φ = 0.407 rad\n t ∈ [0, 7.851 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.416\n ω = 1.601 rad/s\n φ = 0.407 rad\n t ∈ [0, 7.851 s]", "generator_type": "physics_spec", "seed": 10903}
	{"sample_id": "sample_00465", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.937\n ω = 1.406 rad/s\n φ = 0.178 rad\n t ∈ [0, 8.938 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.937\n ω = 1.406 rad/s\n φ = 0.178 rad\n t ∈ [0, 8.938 s]", "generator_type": "physics_spec", "seed": 10904}
	{"sample_id": "sample_00466", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.816\n ω = 2.098 rad/s\n φ = 2.949 rad\n t ∈ [0, 5.990 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.816\n ω = 2.098 rad/s\n φ = 2.949 rad\n t ∈ [0, 5.990 s]", "generator_type": "physics_spec", "seed": 10905}
	{"sample_id": "sample_00467", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.252\n ω = 1.044 rad/s\n φ = 0.144 rad\n t ∈ [0, 12.037 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.252\n ω = 1.044 rad/s\n φ = 0.144 rad\n t ∈ [0, 12.037 s]", "generator_type": "physics_spec", "seed": 10906}
	{"sample_id": "sample_00468", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.006\n ω = 1.625 rad/s\n φ = 1.087 rad\n t ∈ [0, 7.733 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.006\n ω = 1.625 rad/s\n φ = 1.087 rad\n t ∈ [0, 7.733 s]", "generator_type": "physics_spec", "seed": 10907}
	{"sample_id": "sample_00469", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.876\n ω = 1.291 rad/s\n φ = 1.004 rad\n t ∈ [0, 9.735 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.876\n ω = 1.291 rad/s\n φ = 1.004 rad\n t ∈ [0, 9.735 s]", "generator_type": "physics_spec", "seed": 10908}
	{"sample_id": "sample_00470", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.767\n ω = 2.339 rad/s\n φ = 0.969 rad\n t ∈ [0, 5.373 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.767\n ω = 2.339 rad/s\n φ = 0.969 rad\n t ∈ [0, 5.373 s]", "generator_type": "physics_spec", "seed": 10909}
	{"sample_id": "sample_00491", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.157\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.157", "generator_type": "l_system", "seed": 11900}
	{"sample_id": "sample_00492", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 1.147\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 1.147", "generator_type": "l_system", "seed": 11901}
	{"sample_id": "sample_00493", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.100\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.100", "generator_type": "l_system", "seed": 11902}
	{"sample_id": "sample_00494", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.123\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.123", "generator_type": "l_system", "seed": 11903}
	{"sample_id": "sample_00495", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.863\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.863", "generator_type": "l_system", "seed": 11904}
	{"sample_id": "sample_00496", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.951\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.951", "generator_type": "l_system", "seed": 11905}
	{"sample_id": "sample_00497", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.189\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.189", "generator_type": "l_system", "seed": 11906}
	{"sample_id": "sample_00498", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.163\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.163", "generator_type": "l_system", "seed": 11907}
	{"sample_id": "sample_00499", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.966\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.966", "generator_type": "l_system", "seed": 11908}
	{"sample_id": "sample_00500", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.825\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.825", "generator_type": "l_system", "seed": 11909}

	{"sample_id": "sample_00051", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1900\nsize = 44\ncolormap = \"magma\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.554388 * np.exp(\n -((X - 0.610856)*2 / (2 1.0371982)\n + (Y - 1.517410)2 / (2 * 0.6797812))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1900\nsize = 44\ncolormap = \"magma\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.554388 * np.exp(\n -((X - 0.610856)*2 / (2 1.0371982)\n + (Y - 1.517410)2 / (2 * 0.679781**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1900}
	{"sample_id": "sample_00052", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1901\nsize = 27\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.909467 * np.exp(\n -((X - 0.009287)*2 / (2 1.2776282)\n + (Y - 2.744406)2 / (2 * 0.629802*2))\n)\nZ += 0.714654 np.exp(\n -((X - -0.036967)*2 / (2 0.7523102)\n + (Y - 2.230718)2 / (2 * 1.069237*2))\n)\nZ += -1.857357 np.exp(\n -((X - -0.940575)*2 / (2 0.5041852)\n + (Y - 0.917902)2 / (2 * 0.5084482))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1901\nsize = 27\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.909467 * np.exp(\n -((X - 0.009287)*2 / (2 1.2776282)\n + (Y - 2.744406)2 / (2 * 0.629802*2))\n)\nZ += 0.714654 np.exp(\n -((X - -0.036967)*2 / (2 0.7523102)\n + (Y - 2.230718)2 / (2 * 1.069237*2))\n)\nZ += -1.857357 np.exp(\n -((X - -0.940575)*2 / (2 0.5041852)\n + (Y - 0.917902)2 / (2 * 0.508448**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1901}
	{"sample_id": "sample_00053", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1902\nsize = 46\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 1.398239 * np.exp(\n -((X - -0.751440)*2 / (2 0.8174552)\n + (Y - 0.277919)2 / (2 * 0.675317*2))\n)\nZ += -1.122808 np.exp(\n -((X - -0.091660)*2 / (2 0.5987332)\n + (Y - 1.789293)2 / (2 * 0.9083342))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1902\nsize = 46\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 1.398239 * np.exp(\n -((X - -0.751440)*2 / (2 0.8174552)\n + (Y - 0.277919)2 / (2 * 0.675317*2))\n)\nZ += -1.122808 np.exp(\n -((X - -0.091660)*2 / (2 0.5987332)\n + (Y - 1.789293)2 / (2 * 0.908334**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1902}
	{"sample_id": "sample_00054", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1903\nsize = 30\ncolormap = \"viridis\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.033551 * np.exp(\n -((X - -1.404482)*2 / (2 1.3392892)\n + (Y - 1.882034)2 / (2 * 1.391416*2))\n)\nZ += 1.047285 np.exp(\n -((X - 0.383779)*2 / (2 0.6351752)\n + (Y - 1.768766)2 / (2 * 1.2347612))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1903\nsize = 30\ncolormap = \"viridis\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.033551 * np.exp(\n -((X - -1.404482)*2 / (2 1.3392892)\n + (Y - 1.882034)2 / (2 * 1.391416*2))\n)\nZ += 1.047285 np.exp(\n -((X - 0.383779)*2 / (2 0.6351752)\n + (Y - 1.768766)2 / (2 * 1.234761**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1903}
	{"sample_id": "sample_00055", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1904\nsize = 25\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.254077 * np.exp(\n -((X - -1.025582)*2 / (2 0.9469832)\n + (Y - -0.990845)2 / (2 * 1.097465*2))\n)\nZ += 1.991914 np.exp(\n -((X - 0.660525)*2 / (2 1.0325352)\n + (Y - 1.696424)2 / (2 * 1.404641*2))\n)\nZ += -1.787233 np.exp(\n -((X - -1.538434)*2 / (2 0.9503882)\n + (Y - 2.198314)2 / (2 * 0.5733032))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1904\nsize = 25\ncolormap = \"plasma\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.254077 * np.exp(\n -((X - -1.025582)*2 / (2 0.9469832)\n + (Y - -0.990845)2 / (2 * 1.097465*2))\n)\nZ += 1.991914 np.exp(\n -((X - 0.660525)*2 / (2 1.0325352)\n + (Y - 1.696424)2 / (2 * 1.404641*2))\n)\nZ += -1.787233 np.exp(\n -((X - -1.538434)*2 / (2 0.9503882)\n + (Y - 2.198314)2 / (2 * 0.573303**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1904}
	{"sample_id": "sample_00056", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1905\nsize = 45\ncolormap = \"magma\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.252087 * np.exp(\n -((X - -1.716627)*2 / (2 1.4182812)\n + (Y - -0.529773)2 / (2 * 1.283903*2))\n)\nZ += 1.412459 np.exp(\n -((X - 1.529702)*2 / (2 1.4473622)\n + (Y - -0.880475)2 / (2 * 1.1678492))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1905\nsize = 45\ncolormap = \"magma\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.252087 * np.exp(\n -((X - -1.716627)*2 / (2 1.4182812)\n + (Y - -0.529773)2 / (2 * 1.283903*2))\n)\nZ += 1.412459 np.exp(\n -((X - 1.529702)*2 / (2 1.4473622)\n + (Y - -0.880475)2 / (2 * 1.167849**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1905}
	{"sample_id": "sample_00057", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1906\nsize = 35\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.478871 * np.exp(\n -((X - -0.752787)*2 / (2 1.1512012)\n + (Y - -0.425881)2 / (2 * 1.4443692))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1906\nsize = 35\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.478871 * np.exp(\n -((X - -0.752787)*2 / (2 1.1512012)\n + (Y - -0.425881)2 / (2 * 1.444369**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1906}
	{"sample_id": "sample_00058", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1907\nsize = 42\ncolormap = \"viridis\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 0.592669 * np.exp(\n -((X - 1.216654)*2 / (2 1.4172142)\n + (Y - -0.698115)2 / (2 * 1.244495*2))\n)\nZ += 1.383131 np.exp(\n -((X - 0.968027)*2 / (2 0.8641722)\n + (Y - 0.815769)2 / (2 * 1.395209*2))\n)\nZ += 1.402953 np.exp(\n -((X - 1.965273)*2 / (2 0.7792462)\n + (Y - 1.011683)2 / (2 * 1.0772742))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1907\nsize = 42\ncolormap = \"viridis\"\nn_blobs = 3\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += 0.592669 * np.exp(\n -((X - 1.216654)*2 / (2 1.4172142)\n + (Y - -0.698115)2 / (2 * 1.244495*2))\n)\nZ += 1.383131 np.exp(\n -((X - 0.968027)*2 / (2 0.8641722)\n + (Y - 0.815769)2 / (2 * 1.395209*2))\n)\nZ += 1.402953 np.exp(\n -((X - 1.965273)*2 / (2 0.7792462)\n + (Y - 1.011683)2 / (2 * 1.077274**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1907}
	{"sample_id": "sample_00059", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1908\nsize = 26\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -0.672556 * np.exp(\n -((X - 0.812188)*2 / (2 0.5828402)\n + (Y - 1.030430)2 / (2 * 1.4148612))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1908\nsize = 26\ncolormap = \"viridis\"\nn_blobs = 1\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -0.672556 * np.exp(\n -((X - 0.812188)*2 / (2 0.5828402)\n + (Y - 1.030430)2 / (2 * 1.414861**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1908}
	{"sample_id": "sample_00060", "split": "consistent", "domain": "data_visualization", "family": "heatmap", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1909\nsize = 34\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.811477 * np.exp(\n -((X - 1.913968)*2 / (2 1.1239022)\n + (Y - 1.897390)2 / (2 * 1.075150*2))\n)\nZ += 1.836162 np.exp(\n -((X - 0.490576)*2 / (2 1.1390292)\n + (Y - 2.586806)2 / (2 * 0.8287082))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only** this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 1909\nsize = 34\ncolormap = \"inferno\"\nn_blobs = 2\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(-3.0, 3.0, size)\ny = np.linspace(-2.0, 4.0, size)\nX, Y = np.meshgrid(x, y)\nZ = np.zeros((size, size))\n\nZ += -1.811477 * np.exp(\n -((X - 1.913968)*2 / (2 1.1239022)\n + (Y - 1.897390)2 / (2 * 1.075150*2))\n)\nZ += 1.836162 np.exp(\n -((X - 0.490576)*2 / (2 1.1390292)\n + (Y - 2.586806)2 / (2 * 0.828708**2))\n)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nim = ax.pcolormesh(X, Y, Z, cmap=colormap, shading=\"auto\")\nfig.colorbar(im, ax=ax, label=\"Intensity\")\nax.set_title(f\"Gaussian Heatmap ({n_blobs} blob(s))\")\nax.set_xlabel(\"X\")\nax.set_ylabel(\"Y\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 1909}
	{"sample_id": "sample_00121", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2900\nn_points = 161\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.831950 * np.sin(1.118913 * x + 1.008837)\ny1 = 1.338358 * np.sin(0.932295 * x + 2.234393)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.83, f=1.12, φ=1.01\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.34, f=0.93, φ=2.23\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (2 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2900\nn_points = 161\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.831950 * np.sin(1.118913 * x + 1.008837)\ny1 = 1.338358 * np.sin(0.932295 * x + 2.234393)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.83, f=1.12, φ=1.01\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.34, f=0.93, φ=2.23\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (2 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2900}
	{"sample_id": "sample_00122", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2901\nn_points = 145\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.155640 * np.sin(1.302779 * x + 4.981734)\ny1 = 0.911540 * np.sin(2.947745 * x + 2.938028)\ny2 = 0.859045 * np.sin(1.454283 * x + 4.031508)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.16, f=1.30, φ=4.98\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.91, f=2.95, φ=2.94\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=0.86, f=1.45, φ=4.03\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2901\nn_points = 145\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.155640 * np.sin(1.302779 * x + 4.981734)\ny1 = 0.911540 * np.sin(2.947745 * x + 2.938028)\ny2 = 0.859045 * np.sin(1.454283 * x + 4.031508)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.16, f=1.30, φ=4.98\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.91, f=2.95, φ=2.94\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=0.86, f=1.45, φ=4.03\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2901}
	{"sample_id": "sample_00123", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2902\nn_points = 80\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.781911 * np.sin(1.543293 * x + 4.144060)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.78, f=1.54, φ=4.14\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2902\nn_points = 80\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.781911 * np.sin(1.543293 * x + 4.144060)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.78, f=1.54, φ=4.14\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2902}
	{"sample_id": "sample_00124", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2903\nn_points = 141\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.857967 * np.sin(0.942220 * x + 5.084951)\ny1 = 0.590013 * np.sin(1.142741 * x + 0.572998)\ny2 = 1.924395 * np.sin(1.682643 * x + 4.928356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.86, f=0.94, φ=5.08\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.59, f=1.14, φ=0.57\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.92, f=1.68, φ=4.93\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2903\nn_points = 141\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.857967 * np.sin(0.942220 * x + 5.084951)\ny1 = 0.590013 * np.sin(1.142741 * x + 0.572998)\ny2 = 1.924395 * np.sin(1.682643 * x + 4.928356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.86, f=0.94, φ=5.08\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.59, f=1.14, φ=0.57\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.92, f=1.68, φ=4.93\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2903}
	{"sample_id": "sample_00125", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2904\nn_points = 153\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.514431 * np.sin(1.513446 * x + 0.116435)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.51, f=1.51, φ=0.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2904\nn_points = 153\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.514431 * np.sin(1.513446 * x + 0.116435)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.51, f=1.51, φ=0.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2904}
	{"sample_id": "sample_00126", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2905\nn_points = 165\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.607619 * np.sin(1.667818 * x + 6.143726)\ny1 = 1.205378 * np.sin(1.973711 * x + 0.172039)\ny2 = 1.652390 * np.sin(2.122295 * x + 0.591268)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.61, f=1.67, φ=6.14\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.21, f=1.97, φ=0.17\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.65, f=2.12, φ=0.59\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2905\nn_points = 165\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.607619 * np.sin(1.667818 * x + 6.143726)\ny1 = 1.205378 * np.sin(1.973711 * x + 0.172039)\ny2 = 1.652390 * np.sin(2.122295 * x + 0.591268)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.61, f=1.67, φ=6.14\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.21, f=1.97, φ=0.17\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.65, f=2.12, φ=0.59\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2905}
	{"sample_id": "sample_00127", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2906\nn_points = 115\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.617971 * np.sin(1.020470 * x + 0.512607)\ny1 = 0.790376 * np.sin(0.861338 * x + 0.057956)\ny2 = 1.269849 * np.sin(1.617446 * x + 5.271811)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.62, f=1.02, φ=0.51\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.79, f=0.86, φ=0.06\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.27, f=1.62, φ=5.27\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2906\nn_points = 115\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.617971 * np.sin(1.020470 * x + 0.512607)\ny1 = 0.790376 * np.sin(0.861338 * x + 0.057956)\ny2 = 1.269849 * np.sin(1.617446 * x + 5.271811)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.62, f=1.02, φ=0.51\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.79, f=0.86, φ=0.06\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.27, f=1.62, φ=5.27\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2906}
	{"sample_id": "sample_00128", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2907\nn_points = 85\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.032634 * np.sin(2.206059 * x + 2.518333)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.03, f=2.21, φ=2.52\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2907\nn_points = 85\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.032634 * np.sin(2.206059 * x + 2.518333)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.03, f=2.21, φ=2.52\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (1 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2907}
	{"sample_id": "sample_00129", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2908\nn_points = 131\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.698043 * np.sin(2.606526 * x + 5.039795)\ny1 = 1.662470 * np.sin(1.724687 * x + 1.478118)\ny2 = 1.207486 * np.sin(2.091718 * x + 0.338356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.70, f=2.61, φ=5.04\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.66, f=1.72, φ=1.48\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.21, f=2.09, φ=0.34\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2908\nn_points = 131\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 0.698043 * np.sin(2.606526 * x + 5.039795)\ny1 = 1.662470 * np.sin(1.724687 * x + 1.478118)\ny2 = 1.207486 * np.sin(2.091718 * x + 0.338356)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=0.70, f=2.61, φ=5.04\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=1.66, f=1.72, φ=1.48\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.21, f=2.09, φ=0.34\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2908}
	{"sample_id": "sample_00130", "split": "consistent", "domain": "data_visualization", "family": "line_plot", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2909\nn_points = 127\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.044005 * np.sin(1.288522 * x + 3.606051)\ny1 = 0.548443 * np.sin(2.583436 * x + 2.986435)\ny2 = 1.236664 * np.sin(1.559843 * x + 5.118033)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.04, f=1.29, φ=3.61\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.55, f=2.58, φ=2.99\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.24, f=1.56, φ=5.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 2909\nn_points = 127\n\n# ── Data ────────────────────────────────────────────────────────────────────\nx = np.linspace(0.0, 2.0 * np.pi, n_points)\n\ny0 = 1.044005 * np.sin(1.288522 * x + 3.606051)\ny1 = 0.548443 * np.sin(2.583436 * x + 2.986435)\ny2 = 1.236664 * np.sin(1.559843 * x + 5.118033)\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nax.plot(x, y0, color=\"tab:blue\", label=\"A=1.04, f=1.29, φ=3.61\")\nax.plot(x, y1, color=\"tab:orange\", label=\"A=0.55, f=2.58, φ=2.99\")\nax.plot(x, y2, color=\"tab:green\", label=\"A=1.24, f=1.56, φ=5.12\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(\"Sinusoidal Line Plot (3 curve(s))\")\nax.legend(fontsize=8)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 2909}
	{"sample_id": "sample_00161", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3900\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.821660\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3900\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.821660\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3900}
	{"sample_id": "sample_00162", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3901\nn_petals = 4 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.977908\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3901\nn_petals = 4 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.977908\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3901}
	{"sample_id": "sample_00163", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3902\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.288979\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3902\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.288979\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3902}
	{"sample_id": "sample_00164", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3903\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.576224\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3903\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.576224\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3903}
	{"sample_id": "sample_00165", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3904\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.635304\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3904\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.635304\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3904}
	{"sample_id": "sample_00166", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3905\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.350061\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3905\nn_petals = 5 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.350061\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3905}
	{"sample_id": "sample_00167", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3906\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.654892\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3906\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.654892\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3906}
	{"sample_id": "sample_00168", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3907\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.984689\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3907\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.984689\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3907}
	{"sample_id": "sample_00169", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3908\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.757216\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3908\nn_petals = 6 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.757216\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3908}
	{"sample_id": "sample_00170", "split": "consistent", "domain": "data_visualization", "family": "polar", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3909\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.407149\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 3909\nn_petals = 7 # k in r = A\|cos(k·θ/2)\|\namplitude = 1.407149\nn_points = 1000\n\n# ── Data ────────────────────────────────────────────────────────────────────\ntheta = np.linspace(0.0, 2.0 * np.pi, n_points)\nr = amplitude * np.abs(np.cos(n_petals * theta / 2.0))\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(subplot_kw={\"projection\": \"polar\"}, figsize=(5, 5))\nax.plot(theta, r, linewidth=1.5)\nax.set_title(\n f\"Rose Curve (k={n_petals}, A={amplitude:.2f})\", pad=15\n)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 3909}
	{"sample_id": "sample_00201", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4900\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.49, 1.40, 3.60, 1.51]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4900\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.49, 1.40, 3.60, 1.51]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4900}
	{"sample_id": "sample_00202", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4901\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [-1.89, -1.52, -2.96, -2.85, -1.38]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4901\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [-1.89, -1.52, -2.96, -2.85, -1.38]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4901}
	{"sample_id": "sample_00203", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4902\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [2.71, -2.88, 3.19, 2.36, 2.85]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4902\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [2.71, -2.88, 3.19, 2.36, 2.85]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4902}
	{"sample_id": "sample_00204", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4903\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [-2.93, 3.58, 4.65, 0.56]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4903\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [-2.93, 3.58, 4.65, 0.56]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4903}
	{"sample_id": "sample_00205", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4904\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [1.31, 3.65, 0.96, -1.43, 0.04]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4904\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E']\nheights = [1.31, 3.65, 0.96, -1.43, 0.04]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (5 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4904}
	{"sample_id": "sample_00206", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4905\ncolor = \"steelblue\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [4.79, 4.99, -2.15, -2.90]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4905\ncolor = \"steelblue\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [4.79, 4.99, -2.15, -2.90]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4905}
	{"sample_id": "sample_00207", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4906\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [2.27, 3.88, -0.89, 0.48]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4906\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [2.27, 3.88, -0.89, 0.48]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4906}
	{"sample_id": "sample_00208", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4907\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E', 'F']\nheights = [0.40, -0.06, 0.04, -2.27, 2.11, -2.08]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (6 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4907\ncolor = \"mediumseagreen\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D', 'E', 'F']\nheights = [0.40, -0.06, 0.04, -2.27, 2.11, -2.08]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (6 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4907}
	{"sample_id": "sample_00209", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4908\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [3.93, 3.38, -1.72, -2.13]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4908\ncolor = \"coral\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [3.93, 3.38, -1.72, -2.13]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4908}
	{"sample_id": "sample_00210", "split": "consistent", "domain": "data_visualization", "family": "bar_chart", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4909\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.29, -0.41, 4.48, 4.96]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ─────────────────────────────────────────────────────────────\nseed = 4909\ncolor = \"mediumpurple\"\n\n# ── Data ────────────────────────────────────────────────────────────────────\nlabels = ['A', 'B', 'C', 'D']\nheights = [1.29, -0.41, 4.48, 4.96]\n\n# ── Plot ────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(7, 4))\nbars = ax.bar(labels, heights, color=color, edgecolor='white', linewidth=0.5)\nax.axhline(0, color='black', linewidth=0.8)\nfor bar, h in zip(bars, heights):\n va = 'bottom' if h >= 0 else 'top'\n offset = 0.05 if h >= 0 else -0.05\n ax.text(bar.get_x() + bar.get_width()/2, h + offset,\n f'{h:.2f}', ha='center', va=va, fontsize=8, fontweight='bold')\nax.set_title(\"Bar Chart (4 categories)\")\nax.set_xlabel(\"Category\")\nax.set_ylabel(\"Value\")\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 4909}
	{"sample_id": "sample_00231", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5900\nn_points = 59\nslope = 1.149370\nintercept = 1.299066\nnoise_std = 0.539668\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5900\nn_points = 59\nslope = 1.149370\nintercept = 1.299066\nnoise_std = 0.539668\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5900}
	{"sample_id": "sample_00232", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5901\nn_points = 37\nslope = -1.855496\nintercept = -1.322886\nnoise_std = 0.441639\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5901\nn_points = 37\nslope = -1.855496\nintercept = -1.322886\nnoise_std = 0.441639\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5901}
	{"sample_id": "sample_00233", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5902\nn_points = 37\nslope = -1.529994\nintercept = -0.004273\nnoise_std = 0.730252\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5902\nn_points = 37\nslope = -1.529994\nintercept = -0.004273\nnoise_std = 0.730252\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5902}
	{"sample_id": "sample_00234", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5903\nn_points = 51\nslope = 1.440904\nintercept = 0.738086\nnoise_std = 0.352827\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5903\nn_points = 51\nslope = 1.440904\nintercept = 0.738086\nnoise_std = 0.352827\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5903}
	{"sample_id": "sample_00235", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5904\nn_points = 59\nslope = 0.890379\nintercept = 1.055611\nnoise_std = 0.559377\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5904\nn_points = 59\nslope = 0.890379\nintercept = 1.055611\nnoise_std = 0.559377\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5904}
	{"sample_id": "sample_00236", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5905\nn_points = 49\nslope = -1.929874\nintercept = -0.664004\nnoise_std = 0.672751\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5905\nn_points = 49\nslope = -1.929874\nintercept = -0.664004\nnoise_std = 0.672751\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5905}
	{"sample_id": "sample_00237", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5906\nn_points = 40\nslope = 1.016841\nintercept = 1.151654\nnoise_std = 0.678537\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5906\nn_points = 40\nslope = 1.016841\nintercept = 1.151654\nnoise_std = 0.678537\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5906}
	{"sample_id": "sample_00238", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5907\nn_points = 52\nslope = -0.891248\nintercept = 0.612783\nnoise_std = 0.538393\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5907\nn_points = 52\nslope = -0.891248\nintercept = 0.612783\nnoise_std = 0.538393\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5907}
	{"sample_id": "sample_00239", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5908\nn_points = 53\nslope = 1.044824\nintercept = 0.686445\nnoise_std = 0.641637\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5908\nn_points = 53\nslope = 1.044824\nintercept = 0.686445\nnoise_std = 0.641637\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5908}
	{"sample_id": "sample_00240", "split": "consistent", "domain": "data_visualization", "family": "scatter", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```python\nimport numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5909\nn_points = 38\nslope = -0.918191\nintercept = -1.427642\nnoise_std = 0.533250\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "import numpy as np\nimport matplotlib.pyplot as plt\n\n# ── Parameters ──────────────────────────────────────────────────────────────\nseed = 5909\nn_points = 38\nslope = -0.918191\nintercept = -1.427642\nnoise_std = 0.533250\n\n# ── Data ────────────────────────────────────────────────────────────────────\nrng = np.random.default_rng(seed)\nx = np.sort(rng.uniform(-3.0, 3.0, n_points))\ny = slope * x + intercept + rng.normal(0.0, noise_std, n_points)\n\nx_line = np.array([-3.0, 3.0])\ny_line = slope * x_line + intercept\n\n# ── Plot ─────────────────────────────────────────────────────────────────────\nfig, ax = plt.subplots(figsize=(6, 5))\nax.scatter(x, y, alpha=0.6, s=30, color=\"steelblue\", zorder=2)\nax.plot(x_line, y_line, color=\"crimson\", linewidth=2, zorder=3,\n label=f\"y = {slope:.3f}x + {intercept:.3f}\")\nax.set_xlabel(\"x\")\nax.set_ylabel(\"y\")\nax.set_title(f\"Linear Scatter (n={n_points})\")\nax.legend(fontsize=9)\nax.grid(True, alpha=0.3)\nfig.tight_layout()\nplt.show()\n", "generator_type": "python", "seed": 5909}
	{"sample_id": "sample_00291", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.803\\,cos(1.790x + 1.502)\n\nx \\in [0, 2\\pi]\nn_points = 229\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.803\\,cos(1.790x + 1.502)\n\nx \\in [0, 2\\pi]\nn_points = 229", "generator_type": "latex", "seed": 6900}
	{"sample_id": "sample_00292", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.723\\,cos(1.810x + 0.773)\n\nx \\in [0, 2\\pi]\nn_points = 102\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.723\\,cos(1.810x + 0.773)\n\nx \\in [0, 2\\pi]\nn_points = 102", "generator_type": "latex", "seed": 6901}
	{"sample_id": "sample_00293", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.814\\,cos(1.098x + 1.407)\n\nx \\in [0, 2\\pi]\nn_points = 213\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.814\\,cos(1.098x + 1.407)\n\nx \\in [0, 2\\pi]\nn_points = 213", "generator_type": "latex", "seed": 6902}
	{"sample_id": "sample_00294", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.745\\,cos(2.896x + 1.774)\n\nx \\in [0, 2\\pi]\nn_points = 211\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.745\\,cos(2.896x + 1.774)\n\nx \\in [0, 2\\pi]\nn_points = 211", "generator_type": "latex", "seed": 6903}
	{"sample_id": "sample_00295", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.537\\,sin(1.486x + 0.274) + 0.899\\,cos(1.249x + 0.255)\n\nx \\in [0, 2\\pi]\nn_points = 219\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.537\\,sin(1.486x + 0.274) + 0.899\\,cos(1.249x + 0.255)\n\nx \\in [0, 2\\pi]\nn_points = 219", "generator_type": "latex", "seed": 6904}
	{"sample_id": "sample_00296", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.933\\,sin(1.359x + 3.084)\n\nx \\in [0, 2\\pi]\nn_points = 168\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.933\\,sin(1.359x + 3.084)\n\nx \\in [0, 2\\pi]\nn_points = 168", "generator_type": "latex", "seed": 6905}
	{"sample_id": "sample_00297", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.152\\,cos(2.279x + 1.723) + 1.637\\,sin(2.199x + 1.175)\n\nx \\in [0, 2\\pi]\nn_points = 189\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.152\\,cos(2.279x + 1.723) + 1.637\\,sin(2.199x + 1.175)\n\nx \\in [0, 2\\pi]\nn_points = 189", "generator_type": "latex", "seed": 6906}
	{"sample_id": "sample_00298", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.143\\,sin(1.289x + 1.056)\n\nx \\in [0, 2\\pi]\nn_points = 229\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.143\\,sin(1.289x + 1.056)\n\nx \\in [0, 2\\pi]\nn_points = 229", "generator_type": "latex", "seed": 6907}
	{"sample_id": "sample_00299", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.080\\,sin(1.208x + 0.887) + 1.355\\,cos(2.425x + 1.128)\n\nx \\in [0, 2\\pi]\nn_points = 169\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.080\\,sin(1.208x + 0.887) + 1.355\\,cos(2.425x + 1.128)\n\nx \\in [0, 2\\pi]\nn_points = 169", "generator_type": "latex", "seed": 6908}
	{"sample_id": "sample_00300", "split": "consistent", "domain": "mathematical_plots", "family": "sinusoid", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.961\\,sin(2.271x + 1.308)\n\nx \\in [0, 2\\pi]\nn_points = 147\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.961\\,sin(2.271x + 1.308)\n\nx \\in [0, 2\\pi]\nn_points = 147", "generator_type": "latex", "seed": 6909}
	{"sample_id": "sample_00331", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.273x^3 - 0.686x^2 - 1.162x + 0.540\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.273x^3 - 0.686x^2 - 1.162x + 0.540\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7900}
	{"sample_id": "sample_00332", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.348x^2 - 0.103x + 1.950\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.348x^2 - 0.103x + 1.950\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7901}
	{"sample_id": "sample_00333", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 1.483x^2 + 1.711x - 0.297\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 1.483x^2 + 1.711x - 0.297\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7902}
	{"sample_id": "sample_00334", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.866x^3 - 1.746x^2 - 1.098x + 0.551\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.866x^3 - 1.746x^2 - 1.098x + 0.551\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7903}
	{"sample_id": "sample_00335", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.989x^2 - 0.364x - 1.721\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.989x^2 - 0.364x - 1.721\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7904}
	{"sample_id": "sample_00336", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.359x^4 - 1.104x^3 + 1.774x^2 + 1.846x + 0.056\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.359x^4 - 1.104x^3 + 1.774x^2 + 1.846x + 0.056\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7905}
	{"sample_id": "sample_00337", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.290x^4 - 0.794x^3 - 1.678x^2 + 1.818x - 1.313\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.290x^4 - 0.794x^3 - 1.678x^2 + 1.818x - 1.313\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7906}
	{"sample_id": "sample_00338", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = 0.969x^3 + 1.223x^2 - 0.856x + 0.131\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = 0.969x^3 + 1.223x^2 - 0.856x + 0.131\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7907}
	{"sample_id": "sample_00339", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.790x^2 + 0.281x - 0.190\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.790x^2 + 0.281x - 0.190\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7908}
	{"sample_id": "sample_00340", "split": "consistent", "domain": "mathematical_plots", "family": "polynomial", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nf(x) = -1.931x^2 - 1.875x - 0.749\n\nx \\in [-3.0, 3.0]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "f(x) = -1.931x^2 - 1.875x - 0.749\n\nx \\in [-3.0, 3.0]", "generator_type": "latex", "seed": 7909}
	{"sample_id": "sample_00371", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.051 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.051 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8900}
	{"sample_id": "sample_00372", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.252 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.252 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8901}
	{"sample_id": "sample_00373", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.425 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.425 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8902}
	{"sample_id": "sample_00374", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.501 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.501 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8903}
	{"sample_id": "sample_00375", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.474 \\left\|\\cos\\left(\\frac{3\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.474 \\left\|\\cos\\left(\\frac{3\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8904}
	{"sample_id": "sample_00376", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 0.922 \\left\|\\cos\\left(\\frac{4\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 0.922 \\left\|\\cos\\left(\\frac{4\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8905}
	{"sample_id": "sample_00377", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.956 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.956 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8906}
	{"sample_id": "sample_00378", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.363 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.363 \\left\|\\cos\\left(\\frac{5\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8907}
	{"sample_id": "sample_00379", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 1.059 \\left\|\\cos\\left(\\frac{6\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 1.059 \\left\|\\cos\\left(\\frac{6\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8908}
	{"sample_id": "sample_00380", "split": "consistent", "domain": "mathematical_plots", "family": "rose_formula", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```latex\nr(\\theta) = 0.806 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "r(\\theta) = 0.806 \\left\|\\cos\\left(\\frac{7\\theta}{2}\\right)\\right\|\n\n\\theta \\in [0, 2\\pi]", "generator_type": "latex", "seed": 8909}
	{"sample_id": "sample_00411", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 12.688 m/s\n θ = 41.74°\n g = 9.800 m/s²\n t ∈ [0, 1.689 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 12.688 m/s\n θ = 41.74°\n g = 9.800 m/s²\n t ∈ [0, 1.689 s]", "generator_type": "physics_spec", "seed": 9900}
	{"sample_id": "sample_00412", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 8.125 m/s\n θ = 59.16°\n g = 9.800 m/s²\n t ∈ [0, 1.313 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 8.125 m/s\n θ = 59.16°\n g = 9.800 m/s²\n t ∈ [0, 1.313 s]", "generator_type": "physics_spec", "seed": 9901}
	{"sample_id": "sample_00413", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 14.264 m/s\n θ = 63.54°\n g = 9.800 m/s²\n t ∈ [0, 2.324 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 14.264 m/s\n θ = 63.54°\n g = 9.800 m/s²\n t ∈ [0, 2.324 s]", "generator_type": "physics_spec", "seed": 9902}
	{"sample_id": "sample_00414", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.164 m/s\n θ = 57.42°\n g = 9.800 m/s²\n t ∈ [0, 1.996 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.164 m/s\n θ = 57.42°\n g = 9.800 m/s²\n t ∈ [0, 1.996 s]", "generator_type": "physics_spec", "seed": 9903}
	{"sample_id": "sample_00415", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 16.119 m/s\n θ = 60.54°\n g = 9.800 m/s²\n t ∈ [0, 2.680 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 16.119 m/s\n θ = 60.54°\n g = 9.800 m/s²\n t ∈ [0, 2.680 s]", "generator_type": "physics_spec", "seed": 9904}
	{"sample_id": "sample_00416", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.521 m/s\n θ = 44.11°\n g = 9.800 m/s²\n t ∈ [0, 1.871 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.521 m/s\n θ = 44.11°\n g = 9.800 m/s²\n t ∈ [0, 1.871 s]", "generator_type": "physics_spec", "seed": 9905}
	{"sample_id": "sample_00417", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 10.590 m/s\n θ = 61.69°\n g = 9.800 m/s²\n t ∈ [0, 1.739 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 10.590 m/s\n θ = 61.69°\n g = 9.800 m/s²\n t ∈ [0, 1.739 s]", "generator_type": "physics_spec", "seed": 9906}
	{"sample_id": "sample_00418", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 11.954 m/s\n θ = 35.80°\n g = 9.800 m/s²\n t ∈ [0, 1.424 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 11.954 m/s\n θ = 35.80°\n g = 9.800 m/s²\n t ∈ [0, 1.424 s]", "generator_type": "physics_spec", "seed": 9907}
	{"sample_id": "sample_00419", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.968 m/s\n θ = 38.95°\n g = 9.800 m/s²\n t ∈ [0, 1.526 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 13.968 m/s\n θ = 38.95°\n g = 9.800 m/s²\n t ∈ [0, 1.526 s]", "generator_type": "physics_spec", "seed": 9908}
	{"sample_id": "sample_00420", "split": "consistent", "domain": "geometry_physics", "family": "projectile", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nProjectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 9.063 m/s\n θ = 33.45°\n g = 9.800 m/s²\n t ∈ [0, 1.018 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Projectile motion:\n x(t) = v₀ cos(θ) · t\n y(t) = v₀ sin(θ) · t − ½ g t²\n\nParameters:\n v₀ = 9.063 m/s\n θ = 33.45°\n g = 9.800 m/s²\n t ∈ [0, 1.018 s]", "generator_type": "physics_spec", "seed": 9909}
	{"sample_id": "sample_00461", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.915\n ω = 2.096 rad/s\n φ = 2.739 rad\n t ∈ [0, 5.994 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.915\n ω = 2.096 rad/s\n φ = 2.739 rad\n t ∈ [0, 5.994 s]", "generator_type": "physics_spec", "seed": 10900}
	{"sample_id": "sample_00462", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.978\n ω = 0.941 rad/s\n φ = 2.212 rad\n t ∈ [0, 13.351 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.978\n ω = 0.941 rad/s\n φ = 2.212 rad\n t ∈ [0, 13.351 s]", "generator_type": "physics_spec", "seed": 10901}
	{"sample_id": "sample_00463", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.835\n ω = 1.141 rad/s\n φ = 3.038 rad\n t ∈ [0, 11.017 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.835\n ω = 1.141 rad/s\n φ = 3.038 rad\n t ∈ [0, 11.017 s]", "generator_type": "physics_spec", "seed": 10902}
	{"sample_id": "sample_00464", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.416\n ω = 1.601 rad/s\n φ = 0.407 rad\n t ∈ [0, 7.851 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.416\n ω = 1.601 rad/s\n φ = 0.407 rad\n t ∈ [0, 7.851 s]", "generator_type": "physics_spec", "seed": 10903}
	{"sample_id": "sample_00465", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.937\n ω = 1.406 rad/s\n φ = 0.178 rad\n t ∈ [0, 8.938 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.937\n ω = 1.406 rad/s\n φ = 0.178 rad\n t ∈ [0, 8.938 s]", "generator_type": "physics_spec", "seed": 10904}
	{"sample_id": "sample_00466", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.816\n ω = 2.098 rad/s\n φ = 2.949 rad\n t ∈ [0, 5.990 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.816\n ω = 2.098 rad/s\n φ = 2.949 rad\n t ∈ [0, 5.990 s]", "generator_type": "physics_spec", "seed": 10905}
	{"sample_id": "sample_00467", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.252\n ω = 1.044 rad/s\n φ = 0.144 rad\n t ∈ [0, 12.037 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.252\n ω = 1.044 rad/s\n φ = 0.144 rad\n t ∈ [0, 12.037 s]", "generator_type": "physics_spec", "seed": 10906}
	{"sample_id": "sample_00468", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.006\n ω = 1.625 rad/s\n φ = 1.087 rad\n t ∈ [0, 7.733 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.006\n ω = 1.625 rad/s\n φ = 1.087 rad\n t ∈ [0, 7.733 s]", "generator_type": "physics_spec", "seed": 10907}
	{"sample_id": "sample_00469", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.876\n ω = 1.291 rad/s\n φ = 1.004 rad\n t ∈ [0, 9.735 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 1.876\n ω = 1.291 rad/s\n φ = 1.004 rad\n t ∈ [0, 9.735 s]", "generator_type": "physics_spec", "seed": 10908}
	{"sample_id": "sample_00470", "split": "consistent", "domain": "geometry_physics", "family": "harmonic_oscillator", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nHarmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.767\n ω = 2.339 rad/s\n φ = 0.969 rad\n t ∈ [0, 5.373 s]\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "Harmonic oscillator:\n x(t) = A cos(ωt + φ)\n\nParameters:\n A = 0.767\n ω = 2.339 rad/s\n φ = 0.969 rad\n t ∈ [0, 5.373 s]", "generator_type": "physics_spec", "seed": 10909}
	{"sample_id": "sample_00491", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.157\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.157", "generator_type": "l_system", "seed": 11900}
	{"sample_id": "sample_00492", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 1.147\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 1.147", "generator_type": "l_system", "seed": 11901}
	{"sample_id": "sample_00493", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.100\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.100", "generator_type": "l_system", "seed": 11902}
	{"sample_id": "sample_00494", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.123\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.123", "generator_type": "l_system", "seed": 11903}
	{"sample_id": "sample_00495", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.863\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.863", "generator_type": "l_system", "seed": 11904}
	{"sample_id": "sample_00496", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.951\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 4\n Step length: 0.951", "generator_type": "l_system", "seed": 11905}
	{"sample_id": "sample_00497", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.189\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.189", "generator_type": "l_system", "seed": 11906}
	{"sample_id": "sample_00498", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.163\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 1.163", "generator_type": "l_system", "seed": 11907}
	{"sample_id": "sample_00499", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.966\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F-F+F+FF-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.966", "generator_type": "l_system", "seed": 11908}
	{"sample_id": "sample_00500", "split": "consistent", "domain": "fractals_procedural_geometry", "family": "l_system", "perturbation_type": "none", "perturbation_description": "", "prompt": "You are evaluating a scientific visualization for causal consistency.\n\nThe following specification is the symbolic generator — it fully specifies\nwhat the output plot should look like:\n\n```text\nL-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.825\n```\n\nI am showing you an image that claims to be the output of this generator.\n\n---\n\n## Your Task\n\n1. Read the specification carefully. Reason about what the plot should look like\n (shape, orientation, color mapping, symmetry, value signs, etc.).\n2. Examine the attached image.\n3. Decide whether the image is causally consistent with the generator.\n\nIf the image is not consistent, classify the inconsistency using exactly one\nof these labels:\n\n\| Label \| Meaning \|\n\|---\|---\|\n\| `colormap_inversion` \| The colormap used is different from what the code specifies \|\n\| `axis_swap` \| Axes or data dimensions are transposed or mirrored \|\n\| `sign_inversion` \| Values are negated — peaks and troughs (or bar directions) are swapped \|\n\| `amplitude_scale` \| The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes \|\n\| `phase_shift` \| The pattern is shifted from its correct position \|\n\| `frequency_doubling` \| The number of oscillations or cycles is wrong \|\n\| `dc_offset` \| The curves or point cloud are shifted away from their correct baseline \|\n\| `wrong_petal_count` \| The number of petals/lobes differs from what the formula produces \|\n\| `symmetry_mismatch` \| The image contains asymmetry that the code cannot produce \|\n\| `bar_order_swap` \| Two bars have had their heights swapped — a bar's value label contradicts the code's heights list \|\n\| `coefficient_scale` \| Polynomial coefficients are scaled but the formula in the spec is unchanged \|\n\| `wrong_gravity` \| Trajectory uses a different gravitational constant than the spec \|\n\| `wrong_launch_angle` \| Trajectory uses a different launch angle than the spec \|\n\| `wrong_iteration_depth` \| L-system rendered with a different iteration count than the spec \|\n\| `wrong_angle` \| L-system rendered with a different turn angle than the spec \|\n\n---\n\n## Response Format\n\nRespond with only this JSON object and nothing else:\n\n```json\n{\n \"consistent\": true \| false,\n \"bug_type\": \"<one label from the table above, or null if consistent>\",\n \"reasoning\": \"<1–3 sentences: what you expected vs. what you see>\"\n}\n```\n", "generator_content": "L-system:\n Axiom: F\n Rules:\n F → F+F-F-F+F\n Angle: 90°\n Iterations: 3\n Step length: 0.825", "generator_type": "l_system", "seed": 11909}