Upload VeriRender benchmark dataset

README.md

@@ -0,0 +1,69 @@
+---
+license: mit
+task_categories:
+  - visual-question-answering
+  - image-to-text
+language:
+  - en
+tags:
+  - vision-language-model
+  - benchmark
+  - causal-reasoning
+  - scientific-visualization
+  - multimodal
+pretty_name: VeriRender
+size_categories:
+  - n<1K
+configs:
+  - config_name: default
+    data_files:
+      - split: inconsistent
+        path: "inconsistent/**"
+      - split: consistent
+        path: "consistent/**"
+---
+
+# VeriRender Benchmark Dataset
+
+Causal consistency verification samples for Vision-Language Models.
+
+## Layout
+
+```text
+manifest.jsonl          ← canonical index (one row per sample)
+benchmark.yaml          ← config used to generate this release
+inconsistent/{domain}/{sample_id}/   ← corrupted evaluation samples
+consistent/{domain}/{sample_id}/     ← negative controls (clean images)
+```
+
+## Splits
+
+| Split | Description | Eval image |
+|---|---|---|
+| `inconsistent` | Symbolic spec is correct; image has a perturbation | `corrupted.png` |
+| `consistent` | Symbolic spec matches the clean image | `clean.png` |
+
+## Sample folder
+
+Each sample contains:
+- `spec.py` / `spec.tex` / `spec.txt` — symbolic generator (unchanged for inconsistent samples)
+- `clean.png` — faithful rendering
+- `corrupted.png` — perturbed rendering (inconsistent only)
+- `prompt.md` — VLM evaluation prompt
+- `metadata.json` — full provenance
+
+## Loading
+
+```python
+import json
+from pathlib import Path
+
+root = Path(".")
+rows = [json.loads(line) for line in (root / "manifest.jsonl").open()]
+```
+
+Or rebuild the manifest after edits:
+
+```bash
+python scripts/build_manifest.py
+```

benchmark.yaml

@@ -0,0 +1,92 @@
+benchmark:
+  n_per_spec: 10
+  consistent_mode: per_family   # none | per_family | per_spec
+
+domains:
+  data_visualization:
+    representation_type: python
+    families:
+      heatmap:
+        base_seed: 1000
+        perturbations:
+          - colormap_inversion
+          - axis_swap
+          - sign_inversion
+          - amplitude_scale
+          - symmetry_mismatch
+      line_plot:
+        base_seed: 2000
+        perturbations:
+          - sign_inversion
+          - phase_shift
+          - axis_swap
+          - amplitude_scale
+          - frequency_doubling
+          - dc_offset
+      polar:
+        base_seed: 3000
+        perturbations:
+          - sign_inversion
+          - wrong_petal_count
+          - symmetry_mismatch
+      bar_chart:
+        base_seed: 4000
+        perturbations:
+          - sign_inversion
+          - amplitude_scale
+          - bar_order_swap
+      scatter:
+        base_seed: 5000
+        perturbations:
+          - sign_inversion
+          - dc_offset
+
+  mathematical_plots:
+    representation_type: latex
+    families:
+      sinusoid:
+        base_seed: 6000
+        perturbations:
+          - sign_inversion
+          - amplitude_scale
+          - frequency_doubling
+          - phase_shift
+          - dc_offset
+      polynomial:
+        base_seed: 7000
+        perturbations:
+          - sign_inversion
+          - dc_offset
+          - coefficient_scale
+      rose_formula:
+        base_seed: 8000
+        perturbations:
+          - sign_inversion
+          - wrong_petal_count
+          - symmetry_mismatch
+
+  geometry_physics:
+    representation_type: physics_spec
+    families:
+      projectile:
+        base_seed: 9000
+        perturbations:
+          - sign_inversion
+          - wrong_gravity
+          - wrong_launch_angle
+      harmonic_oscillator:
+        base_seed: 10000
+        perturbations:
+          - sign_inversion
+          - amplitude_scale
+          - frequency_doubling
+          - phase_shift
+
+  fractals_procedural_geometry:
+    representation_type: l_system
+    families:
+      l_system:
+        base_seed: 11000
+        perturbations:
+          - wrong_iteration_depth
+          - wrong_angle

consistent/data_visualization/sample_00051/metadata.json

@@ -0,0 +1,37 @@
+{
+  "sample_id": "sample_00051",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00051",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1900,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1900,
+      "size": 44,
+      "n_blobs": 1,
+      "blobs": [
+        {
+          "cx": 0.61085577881573,
+          "cy": 1.5174101799883126,
+          "sx": 1.0371975800132418,
+          "sy": 0.6797807725608512,
+          "amp": -1.554387638420929
+        }
+      ],
+      "colormap": "magma"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00051/prompt.md

@@ -0,0 +1,94 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00051`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1900
+size      = 44
+colormap  = "magma"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.554388 * np.exp(
+    -((X - 0.610856)**2 / (2 * 1.037198**2)
+      + (Y - 1.517410)**2 / (2 * 0.679781**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00051/spec.py

@@ -0,0 +1,29 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1900
+size      = 44
+colormap  = "magma"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.554388 * np.exp(
+    -((X - 0.610856)**2 / (2 * 1.037198**2)
+      + (Y - 1.517410)**2 / (2 * 0.679781**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00052/metadata.json

@@ -0,0 +1,51 @@
+{
+  "sample_id": "sample_00052",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00052",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1901,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1901,
+      "size": 27,
+      "n_blobs": 3,
+      "blobs": [
+        {
+          "cx": 0.00928677522485355,
+          "cy": 2.7444061610545893,
+          "sx": 1.2776281321771092,
+          "sy": 0.6298019532754228,
+          "amp": -1.9094671824014697
+        },
+        {
+          "cx": -0.03696748795995486,
+          "cy": 2.2307181299603127,
+          "sx": 0.7523102453039284,
+          "sy": 1.06923728631266,
+          "amp": 0.7146535881166725
+        },
+        {
+          "cx": -0.9405754630883605,
+          "cy": 0.9179024703123635,
+          "sx": 0.5041846480212553,
+          "sy": 0.5084477268719658,
+          "amp": -1.8573569202593092
+        }
+      ],
+      "colormap": "plasma"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00052/prompt.md

@@ -0,0 +1,102 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00052`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1901
+size      = 27
+colormap  = "plasma"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.909467 * np.exp(
+    -((X - 0.009287)**2 / (2 * 1.277628**2)
+      + (Y - 2.744406)**2 / (2 * 0.629802**2))
+)
+Z += 0.714654 * np.exp(
+    -((X - -0.036967)**2 / (2 * 0.752310**2)
+      + (Y - 2.230718)**2 / (2 * 1.069237**2))
+)
+Z += -1.857357 * np.exp(
+    -((X - -0.940575)**2 / (2 * 0.504185**2)
+      + (Y - 0.917902)**2 / (2 * 0.508448**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00052/spec.py

@@ -0,0 +1,37 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1901
+size      = 27
+colormap  = "plasma"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.909467 * np.exp(
+    -((X - 0.009287)**2 / (2 * 1.277628**2)
+      + (Y - 2.744406)**2 / (2 * 0.629802**2))
+)
+Z += 0.714654 * np.exp(
+    -((X - -0.036967)**2 / (2 * 0.752310**2)
+      + (Y - 2.230718)**2 / (2 * 1.069237**2))
+)
+Z += -1.857357 * np.exp(
+    -((X - -0.940575)**2 / (2 * 0.504185**2)
+      + (Y - 0.917902)**2 / (2 * 0.508448**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00053/metadata.json

@@ -0,0 +1,44 @@
+{
+  "sample_id": "sample_00053",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00053",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1902,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1902,
+      "size": 46,
+      "n_blobs": 2,
+      "blobs": [
+        {
+          "cx": -0.7514402191317222,
+          "cy": 0.27791857789447594,
+          "sx": 0.8174546655531164,
+          "sy": 0.6753174765491572,
+          "amp": 1.3982386738655987
+        },
+        {
+          "cx": -0.09165976977256385,
+          "cy": 1.7892934338324253,
+          "sx": 0.5987332538964991,
+          "sy": 0.9083335335441174,
+          "amp": -1.1228078573248417
+        }
+      ],
+      "colormap": "inferno"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00053/prompt.md

@@ -0,0 +1,98 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00053`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1902
+size      = 46
+colormap  = "inferno"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += 1.398239 * np.exp(
+    -((X - -0.751440)**2 / (2 * 0.817455**2)
+      + (Y - 0.277919)**2 / (2 * 0.675317**2))
+)
+Z += -1.122808 * np.exp(
+    -((X - -0.091660)**2 / (2 * 0.598733**2)
+      + (Y - 1.789293)**2 / (2 * 0.908334**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00053/spec.py

@@ -0,0 +1,33 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1902
+size      = 46
+colormap  = "inferno"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += 1.398239 * np.exp(
+    -((X - -0.751440)**2 / (2 * 0.817455**2)
+      + (Y - 0.277919)**2 / (2 * 0.675317**2))
+)
+Z += -1.122808 * np.exp(
+    -((X - -0.091660)**2 / (2 * 0.598733**2)
+      + (Y - 1.789293)**2 / (2 * 0.908334**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00054/metadata.json

@@ -0,0 +1,44 @@
+{
+  "sample_id": "sample_00054",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00054",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1903,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1903,
+      "size": 30,
+      "n_blobs": 2,
+      "blobs": [
+        {
+          "cx": -1.404481930462251,
+          "cy": 1.8820338796892844,
+          "sx": 1.3392889176301213,
+          "sy": 1.3914159508053348,
+          "amp": -1.0335510656026674
+        },
+        {
+          "cx": 0.38377916053763617,
+          "cy": 1.7687655210580098,
+          "sx": 0.6351745477838537,
+          "sy": 1.2347606930609019,
+          "amp": 1.0472853749795503
+        }
+      ],
+      "colormap": "viridis"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00054/prompt.md

@@ -0,0 +1,98 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00054`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1903
+size      = 30
+colormap  = "viridis"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.033551 * np.exp(
+    -((X - -1.404482)**2 / (2 * 1.339289**2)
+      + (Y - 1.882034)**2 / (2 * 1.391416**2))
+)
+Z += 1.047285 * np.exp(
+    -((X - 0.383779)**2 / (2 * 0.635175**2)
+      + (Y - 1.768766)**2 / (2 * 1.234761**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00054/spec.py

@@ -0,0 +1,33 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1903
+size      = 30
+colormap  = "viridis"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.033551 * np.exp(
+    -((X - -1.404482)**2 / (2 * 1.339289**2)
+      + (Y - 1.882034)**2 / (2 * 1.391416**2))
+)
+Z += 1.047285 * np.exp(
+    -((X - 0.383779)**2 / (2 * 0.635175**2)
+      + (Y - 1.768766)**2 / (2 * 1.234761**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00055/metadata.json

@@ -0,0 +1,51 @@
+{
+  "sample_id": "sample_00055",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00055",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1904,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1904,
+      "size": 25,
+      "n_blobs": 3,
+      "blobs": [
+        {
+          "cx": -1.025582172236188,
+          "cy": -0.9908447430283012,
+          "sx": 0.9469829365771695,
+          "sy": 1.0974648499459585,
+          "amp": -1.2540773739724984
+        },
+        {
+          "cx": 0.6605254885914489,
+          "cy": 1.6964244630893819,
+          "sx": 1.0325346925641545,
+          "sy": 1.4046411544279211,
+          "amp": 1.9919144409047662
+        },
+        {
+          "cx": -1.5384335870949366,
+          "cy": 2.1983143560336385,
+          "sx": 0.9503880706817222,
+          "sy": 0.5733027464209298,
+          "amp": -1.7872325254775374
+        }
+      ],
+      "colormap": "plasma"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00055/prompt.md

@@ -0,0 +1,102 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00055`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1904
+size      = 25
+colormap  = "plasma"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.254077 * np.exp(
+    -((X - -1.025582)**2 / (2 * 0.946983**2)
+      + (Y - -0.990845)**2 / (2 * 1.097465**2))
+)
+Z += 1.991914 * np.exp(
+    -((X - 0.660525)**2 / (2 * 1.032535**2)
+      + (Y - 1.696424)**2 / (2 * 1.404641**2))
+)
+Z += -1.787233 * np.exp(
+    -((X - -1.538434)**2 / (2 * 0.950388**2)
+      + (Y - 2.198314)**2 / (2 * 0.573303**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00055/spec.py

@@ -0,0 +1,37 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1904
+size      = 25
+colormap  = "plasma"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.254077 * np.exp(
+    -((X - -1.025582)**2 / (2 * 0.946983**2)
+      + (Y - -0.990845)**2 / (2 * 1.097465**2))
+)
+Z += 1.991914 * np.exp(
+    -((X - 0.660525)**2 / (2 * 1.032535**2)
+      + (Y - 1.696424)**2 / (2 * 1.404641**2))
+)
+Z += -1.787233 * np.exp(
+    -((X - -1.538434)**2 / (2 * 0.950388**2)
+      + (Y - 2.198314)**2 / (2 * 0.573303**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00056/metadata.json

@@ -0,0 +1,44 @@
+{
+  "sample_id": "sample_00056",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00056",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1905,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1905,
+      "size": 45,
+      "n_blobs": 2,
+      "blobs": [
+        {
+          "cx": -1.716627103651596,
+          "cy": -0.5297733553676873,
+          "sx": 1.4182813576359248,
+          "sy": 1.2839030656303951,
+          "amp": -1.2520868682872761
+        },
+        {
+          "cx": 1.529701828758752,
+          "cy": -0.8804752259492181,
+          "sx": 1.4473616320018907,
+          "sy": 1.167848982598736,
+          "amp": 1.4124590602487401
+        }
+      ],
+      "colormap": "magma"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00056/prompt.md

@@ -0,0 +1,98 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00056`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1905
+size      = 45
+colormap  = "magma"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.252087 * np.exp(
+    -((X - -1.716627)**2 / (2 * 1.418281**2)
+      + (Y - -0.529773)**2 / (2 * 1.283903**2))
+)
+Z += 1.412459 * np.exp(
+    -((X - 1.529702)**2 / (2 * 1.447362**2)
+      + (Y - -0.880475)**2 / (2 * 1.167849**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00056/spec.py

@@ -0,0 +1,33 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1905
+size      = 45
+colormap  = "magma"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.252087 * np.exp(
+    -((X - -1.716627)**2 / (2 * 1.418281**2)
+      + (Y - -0.529773)**2 / (2 * 1.283903**2))
+)
+Z += 1.412459 * np.exp(
+    -((X - 1.529702)**2 / (2 * 1.447362**2)
+      + (Y - -0.880475)**2 / (2 * 1.167849**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00057/metadata.json

@@ -0,0 +1,37 @@
+{
+  "sample_id": "sample_00057",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00057",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1906,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1906,
+      "size": 35,
+      "n_blobs": 1,
+      "blobs": [
+        {
+          "cx": -0.7527872580354669,
+          "cy": -0.425881148330737,
+          "sx": 1.1512009975958548,
+          "sy": 1.44436939419273,
+          "amp": -1.4788705729600022
+        }
+      ],
+      "colormap": "viridis"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00057/prompt.md

@@ -0,0 +1,94 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00057`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1906
+size      = 35
+colormap  = "viridis"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.478871 * np.exp(
+    -((X - -0.752787)**2 / (2 * 1.151201**2)
+      + (Y - -0.425881)**2 / (2 * 1.444369**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00057/spec.py

@@ -0,0 +1,29 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1906
+size      = 35
+colormap  = "viridis"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.478871 * np.exp(
+    -((X - -0.752787)**2 / (2 * 1.151201**2)
+      + (Y - -0.425881)**2 / (2 * 1.444369**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00058/metadata.json

@@ -0,0 +1,51 @@
+{
+  "sample_id": "sample_00058",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00058",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1907,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1907,
+      "size": 42,
+      "n_blobs": 3,
+      "blobs": [
+        {
+          "cx": 1.2166542142247603,
+          "cy": -0.6981153273636793,
+          "sx": 1.4172144773618949,
+          "sy": 1.244495237722108,
+          "amp": 0.5926686450972745
+        },
+        {
+          "cx": 0.9680270489054204,
+          "cy": 0.8157688873253512,
+          "sx": 0.8641721068360724,
+          "sy": 1.3952087300689957,
+          "amp": 1.3831311830034514
+        },
+        {
+          "cx": 1.9652727205924059,
+          "cy": 1.0116830955054525,
+          "sx": 0.7792459388799935,
+          "sy": 1.0772741898826017,
+          "amp": 1.4029525572804764
+        }
+      ],
+      "colormap": "viridis"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00058/prompt.md

@@ -0,0 +1,102 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00058`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1907
+size      = 42
+colormap  = "viridis"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += 0.592669 * np.exp(
+    -((X - 1.216654)**2 / (2 * 1.417214**2)
+      + (Y - -0.698115)**2 / (2 * 1.244495**2))
+)
+Z += 1.383131 * np.exp(
+    -((X - 0.968027)**2 / (2 * 0.864172**2)
+      + (Y - 0.815769)**2 / (2 * 1.395209**2))
+)
+Z += 1.402953 * np.exp(
+    -((X - 1.965273)**2 / (2 * 0.779246**2)
+      + (Y - 1.011683)**2 / (2 * 1.077274**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00058/spec.py

@@ -0,0 +1,37 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1907
+size      = 42
+colormap  = "viridis"
+n_blobs   = 3
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += 0.592669 * np.exp(
+    -((X - 1.216654)**2 / (2 * 1.417214**2)
+      + (Y - -0.698115)**2 / (2 * 1.244495**2))
+)
+Z += 1.383131 * np.exp(
+    -((X - 0.968027)**2 / (2 * 0.864172**2)
+      + (Y - 0.815769)**2 / (2 * 1.395209**2))
+)
+Z += 1.402953 * np.exp(
+    -((X - 1.965273)**2 / (2 * 0.779246**2)
+      + (Y - 1.011683)**2 / (2 * 1.077274**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00059/metadata.json

@@ -0,0 +1,37 @@
+{
+  "sample_id": "sample_00059",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00059",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1908,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1908,
+      "size": 26,
+      "n_blobs": 1,
+      "blobs": [
+        {
+          "cx": 0.8121880777843526,
+          "cy": 1.030430488963911,
+          "sx": 0.5828399092978374,
+          "sy": 1.414860638338725,
+          "amp": -0.6725555530137711
+        }
+      ],
+      "colormap": "viridis"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00059/prompt.md

@@ -0,0 +1,94 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00059`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1908
+size      = 26
+colormap  = "viridis"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -0.672556 * np.exp(
+    -((X - 0.812188)**2 / (2 * 0.582840**2)
+      + (Y - 1.030430)**2 / (2 * 1.414861**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00059/spec.py

@@ -0,0 +1,29 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1908
+size      = 26
+colormap  = "viridis"
+n_blobs   = 1
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -0.672556 * np.exp(
+    -((X - 0.812188)**2 / (2 * 0.582840**2)
+      + (Y - 1.030430)**2 / (2 * 1.414861**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00060/metadata.json

@@ -0,0 +1,44 @@
+{
+  "sample_id": "sample_00060",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00060",
+  "domain": "data_visualization",
+  "family": "heatmap",
+  "seed": 1909,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 1909,
+      "size": 34,
+      "n_blobs": 2,
+      "blobs": [
+        {
+          "cx": 1.9139680809502884,
+          "cy": 1.8973902075583284,
+          "sx": 1.1239015619274189,
+          "sy": 1.0751496321070007,
+          "amp": -1.8114773319036859
+        },
+        {
+          "cx": 0.49057581098433056,
+          "cy": 2.5868064164145474,
+          "sx": 1.139029393598444,
+          "sy": 0.8287075768724488,
+          "amp": 1.8361618507470983
+        }
+      ],
+      "colormap": "inferno"
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00060/prompt.md

@@ -0,0 +1,98 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00060`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1909
+size      = 34
+colormap  = "inferno"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.811477 * np.exp(
+    -((X - 1.913968)**2 / (2 * 1.123902**2)
+      + (Y - 1.897390)**2 / (2 * 1.075150**2))
+)
+Z += 1.836162 * np.exp(
+    -((X - 0.490576)**2 / (2 * 1.139029**2)
+      + (Y - 2.586806)**2 / (2 * 0.828708**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** 2D heatmap (matplotlib `pcolormesh`)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00060/spec.py

@@ -0,0 +1,33 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed      = 1909
+size      = 34
+colormap  = "inferno"
+n_blobs   = 2
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(-3.0, 3.0, size)
+y = np.linspace(-2.0, 4.0, size)
+X, Y = np.meshgrid(x, y)
+Z = np.zeros((size, size))
+
+Z += -1.811477 * np.exp(
+    -((X - 1.913968)**2 / (2 * 1.123902**2)
+      + (Y - 1.897390)**2 / (2 * 1.075150**2))
+)
+Z += 1.836162 * np.exp(
+    -((X - 0.490576)**2 / (2 * 1.139029**2)
+      + (Y - 2.586806)**2 / (2 * 0.828708**2))
+)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(6, 5))
+im = ax.pcolormesh(X, Y, Z, cmap=colormap, shading="auto")
+fig.colorbar(im, ax=ax, label="Intensity")
+ax.set_title(f"Gaussian Heatmap  ({n_blobs} blob(s))")
+ax.set_xlabel("X")
+ax.set_ylabel("Y")
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00121/metadata.json

@@ -0,0 +1,39 @@
+{
+  "sample_id": "sample_00121",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00121",
+  "domain": "data_visualization",
+  "family": "line_plot",
+  "seed": 2900,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 2900,
+      "n_lines": 2,
+      "n_points": 161,
+      "lines": [
+        {
+          "freq": 1.1189131191591275,
+          "phase": 1.008836567919947,
+          "amp": 0.8319501032766907
+        },
+        {
+          "freq": 0.9322946479429992,
+          "phase": 2.2343933631344832,
+          "amp": 1.3383584330015925
+        }
+      ]
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00121/prompt.md

@@ -0,0 +1,89 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00121`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed     = 2900
+n_points = 161
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(0.0, 2.0 * np.pi, n_points)
+
+y0 = 0.831950 * np.sin(1.118913 * x + 1.008837)
+y1 = 1.338358 * np.sin(0.932295 * x + 2.234393)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(7, 4))
+ax.plot(x, y0, color="tab:blue", label="A=0.83, f=1.12, φ=1.01")
+ax.plot(x, y1, color="tab:orange", label="A=1.34, f=0.93, φ=2.23")
+ax.set_xlabel("x")
+ax.set_ylabel("y")
+ax.set_title("Sinusoidal Line Plot  (2 curve(s))")
+ax.legend(fontsize=8)
+ax.grid(True, alpha=0.3)
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** Line plot (sinusoidal curves)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00121/spec.py

@@ -0,0 +1,24 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed     = 2900
+n_points = 161
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(0.0, 2.0 * np.pi, n_points)
+
+y0 = 0.831950 * np.sin(1.118913 * x + 1.008837)
+y1 = 1.338358 * np.sin(0.932295 * x + 2.234393)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(7, 4))
+ax.plot(x, y0, color="tab:blue", label="A=0.83, f=1.12, φ=1.01")
+ax.plot(x, y1, color="tab:orange", label="A=1.34, f=0.93, φ=2.23")
+ax.set_xlabel("x")
+ax.set_ylabel("y")
+ax.set_title("Sinusoidal Line Plot  (2 curve(s))")
+ax.legend(fontsize=8)
+ax.grid(True, alpha=0.3)
+fig.tight_layout()
+plt.show()

consistent/data_visualization/sample_00122/metadata.json

@@ -0,0 +1,44 @@
+{
+  "sample_id": "sample_00122",
+  "split": "consistent",
+  "path": "consistent/data_visualization/sample_00122",
+  "domain": "data_visualization",
+  "family": "line_plot",
+  "seed": 2901,
+  "consistent": true,
+  "perturbation": {
+    "type": "none",
+    "description": ""
+  },
+  "symbolic_spec": {
+    "representation_type": "python",
+    "filename": "spec.py",
+    "params": {
+      "seed": 2901,
+      "n_lines": 3,
+      "n_points": 145,
+      "lines": [
+        {
+          "freq": 1.3027794766211365,
+          "phase": 4.981734333879934,
+          "amp": 1.1556399482473287
+        },
+        {
+          "freq": 2.9477452643605653,
+          "phase": 2.938028445668197,
+          "amp": 0.9115399721335747
+        },
+        {
+          "freq": 1.4542827508589877,
+          "phase": 4.031507938481674,
+          "amp": 0.8590449442353691
+        }
+      ]
+    }
+  },
+  "files": {
+    "symbolic_spec": "spec.py",
+    "clean_image": "clean.png",
+    "prompt": "prompt.md"
+  }
+}

consistent/data_visualization/sample_00122/prompt.md

@@ -0,0 +1,91 @@
+# VeriRender — Causal Consistency Evaluation
+**Sample:** `sample_00122`
+
+> **Before sending:** attach `clean.png` from this folder as the image,
+> then paste everything below the horizontal rule into the chat.
+
+---
+
+You are evaluating a scientific visualization for **causal consistency**.
+
+The following specification is the **symbolic generator** — it fully specifies
+what the output plot should look like:
+
+```python
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed     = 2901
+n_points = 145
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(0.0, 2.0 * np.pi, n_points)
+
+y0 = 1.155640 * np.sin(1.302779 * x + 4.981734)
+y1 = 0.911540 * np.sin(2.947745 * x + 2.938028)
+y2 = 0.859045 * np.sin(1.454283 * x + 4.031508)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(7, 4))
+ax.plot(x, y0, color="tab:blue", label="A=1.16, f=1.30, φ=4.98")
+ax.plot(x, y1, color="tab:orange", label="A=0.91, f=2.95, φ=2.94")
+ax.plot(x, y2, color="tab:green", label="A=0.86, f=1.45, φ=4.03")
+ax.set_xlabel("x")
+ax.set_ylabel("y")
+ax.set_title("Sinusoidal Line Plot  (3 curve(s))")
+ax.legend(fontsize=8)
+ax.grid(True, alpha=0.3)
+fig.tight_layout()
+plt.show()
+```
+
+**Domain:** Data visualization  
+**Plot family:** Line plot (sinusoidal curves)
+
+I am showing you an image that claims to be the output of this generator.
+
+---
+
+## Your Task
+
+1. Read the specification carefully. Reason about what the plot should look like
+   (shape, orientation, color mapping, symmetry, value signs, etc.).
+2. Examine the attached image.
+3. Decide whether the image is **causally consistent** with the generator.
+
+If the image is **not** consistent, classify the inconsistency using exactly one
+of these labels:
+
+| Label | Meaning |
+|---|---|
+| `colormap_inversion` | The colormap used is different from what the code specifies |
+| `axis_swap` | Axes or data dimensions are transposed or mirrored |
+| `sign_inversion` | Values are negated — peaks and troughs (or bar directions) are swapped |
+| `amplitude_scale` | The value scale is wrong — the colorbar, y-axis range, or bar value labels do not match the formula's amplitudes |
+| `phase_shift` | The pattern is shifted from its correct position |
+| `frequency_doubling` | The number of oscillations or cycles is wrong |
+| `dc_offset` | The curves or point cloud are shifted away from their correct baseline |
+| `wrong_petal_count` | The number of petals/lobes differs from what the formula produces |
+| `symmetry_mismatch` | The image contains asymmetry that the code cannot produce |
+| `bar_order_swap` | Two bars have had their heights swapped — a bar's value label contradicts the code's heights list |
+| `coefficient_scale` | Polynomial coefficients are scaled but the formula in the spec is unchanged |
+| `wrong_gravity` | Trajectory uses a different gravitational constant than the spec |
+| `wrong_launch_angle` | Trajectory uses a different launch angle than the spec |
+| `wrong_iteration_depth` | L-system rendered with a different iteration count than the spec |
+| `wrong_angle` | L-system rendered with a different turn angle than the spec |
+
+---
+
+## Response Format
+
+Respond with **only** this JSON object and nothing else:
+
+```json
+{
+  "consistent": true | false,
+  "bug_type": "<one label from the table above, or null if consistent>",
+  "confidence": "low | medium | high",
+  "reasoning": "<1–3 sentences: what you expected vs. what you see>"
+}
+```

consistent/data_visualization/sample_00122/spec.py

@@ -0,0 +1,26 @@
+import numpy as np
+import matplotlib.pyplot as plt
+
+# ── Parameters ─────────────────────────────────────────────────────────────
+seed     = 2901
+n_points = 145
+
+# ── Data ────────────────────────────────────────────────────────────────────
+x = np.linspace(0.0, 2.0 * np.pi, n_points)
+
+y0 = 1.155640 * np.sin(1.302779 * x + 4.981734)
+y1 = 0.911540 * np.sin(2.947745 * x + 2.938028)
+y2 = 0.859045 * np.sin(1.454283 * x + 4.031508)
+
+# ── Plot ────────────────────────────────────────────────────────────────────
+fig, ax = plt.subplots(figsize=(7, 4))
+ax.plot(x, y0, color="tab:blue", label="A=1.16, f=1.30, φ=4.98")
+ax.plot(x, y1, color="tab:orange", label="A=0.91, f=2.95, φ=2.94")
+ax.plot(x, y2, color="tab:green", label="A=0.86, f=1.45, φ=4.03")
+ax.set_xlabel("x")
+ax.set_ylabel("y")
+ax.set_title("Sinusoidal Line Plot  (3 curve(s))")
+ax.legend(fontsize=8)
+ax.grid(True, alpha=0.3)
+fig.tight_layout()
+plt.show()