Dataset Viewer
Auto-converted to Parquet Duplicate
row_id
stringlengths
38
99
prompt
stringlengths
421
4.36k
trace
stringlengths
159
57k
hop
int64
0
5
tool_name
stringclasses
5 values
answer_format
stringclasses
5 values
gold
stringlengths
1
96
compacted
bool
2 classes
row:graph:graph:src_761577fa6538e2ed:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -41)) wp2 = wp1.circle(100) wp3 = wp2.extrude(81) wp4 = wp3.edges() wp5 = wp4.fillet(2) result = wp5 ``` Identify the target workplane that satisfies all...
We are given a CadQuery script and asked to identify the target workplane that satisfies: "Among the remaining candidates, choose the one with the smallest workplane origin height." The script defines wp1 through wp5. We need to output the variable name of the workplane that meets the constraint. The script: wp1 = cq....
1
submit_text
text
wp1
false
row:graph:src_f8d598a7cae260b8:synth:v2:parameterized_feature_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -67)) wp2 = wp1.circle(27) wp3 = wp2.workplane(offset=100) wp4 = wp3.circle(57) wp5 = wp4.loft(combine=True) wp6 = cq.Workplane('XY', origin=(0, 0, -67)) ...
The solid features are wp5 (frustum, 3 faces), wp10 (frustum, 3 faces), wp11 (hollow frustum, 4 faces), wp16 (sweep of rectangle, 6 faces), wp17 (union, 10 faces). Ordering by face count ascending: wp5 (3), wp10 (3), wp11 (4), wp16 (6), wp17 (10). The 3rd candidate is wp11. {"name": "submit_text", "arguments": {"text"...
2
submit_text
text
wp11
true
row:graph:src_290c642dc6e502d6:synth:v3:parameterized_face_spatial_relation_chain_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY') wp2 = wp1.box(200, 54, 27) wp3 = wp2.edges() wp4 = wp3.fillet(2) wp5 = wp4.faces('>Z') wp6 = wp5.workplane() wp7 = wp6.pushPoints([[-80, 0], [-48, 0], [-16, 0], [16, 0]...
Let’s break down the CadQuery construction step by step. 1. **wp1 = `Workplane('XY')`**: Start on XY plane. 2. **wp2 = `.box(200, 54, 27)`**: Creates a box centered at (0,0,0) that is 200 mm in X, 54 mm in Y, 27 mm in Z. So: X from -100 to 100, Y from -27 to 27, Z from -13.5 to 13.5. 3. **wp3 = `.edges()`**: Selects...
5
submit_label
candidate_label
"F"
false
row:graph:src_f8d598a7cae260b8:synth:v2:parameterized_feature_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -67)) wp2 = wp1.circle(27) wp3 = wp2.workplane(offset=100) wp4 = wp3.circle(57) wp5 = wp4.loft(combine=True) wp6 = cq.Workplane('XY', origin=(0, 0, -67)) ...
We need to analyze the script to determine face counts for the various workplane objects and find the one that meets the constraint. The problem requires: - Identify target feature with a specific **face count** property. - Among candidates, choose the 3rd from smallest to greatest face count. Let's trace each workpl...
2
submit_text
text
wp11
false
row:graph:graph:src_761577fa6538e2ed:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -41)) wp2 = wp1.circle(100) wp3 = wp2.extrude(81) wp4 = wp3.edges() wp5 = wp4.fillet(2) result = wp5 ``` Identify the target workplane that satisfies all...
We need to understand the CadQuery script. It creates a Workplane at origin (0,0,-41) with XY plane. Then adds a circle of radius 100? Actually circle(100) creates a circle of radius 100? In CadQuery, circle(radius) draws a circle. So wp2 is a circle on that workplane. Then extrude 81 units. So wp3 is a solid cylinder ...
1
submit_text
text
wp1
false
row:graph:graph:src_5a76c86945d5f757:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('YZ', origin=(-100, 0, 0)) wp2 = wp1.circle(89) wp3 = wp2.extrude(200) wp4 = wp3.edges() wp5 = wp4.chamfer(5) result = wp5 ``` Identify the target workplane that satisfies a...
wp1 origin (-100,0,0) has z=0. All subsequent workplanes (wp2-wp5) inherit the same origin from wp1 via circle, extrude, edges, chamfer. All have z=0, so wp1 (first) qualifies as smallest. {"name": "submit_text", "arguments": {"text": "wp1"}} The script defines five workplane variables (wp1 through wp5), all derived ...
1
submit_text
text
wp1
true
row:graph:src_8c2c76f928d78b2f:synth:v0:parameterized_path_point_extreme
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY') wp2 = wp1.cylinder(44, 100) wp3 = cq.Workplane('XY') wp4 = wp3.cylinder(44, 64) wp5 = wp2.cut(wp4) wp6 = cq.Workplane('XY') wp7 = wp6.splineApprox([[82, 0, 0], [81, 16,...
We are given a CadQuery script. The question asks: "Identify the target path point that satisfies all of these constraints: - The target point is one of the points on a path. - The target point can be described by its Y coordinate. - Within the narrowed set, select the candidate with the maximum Y coordinate. - For any...
3
submit_coordinate
coordinate
[-9, 82, 36]
false
row:graph:src_290c642dc6e502d6:synth:v3:parameterized_face_spatial_relation_chain_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY') wp2 = wp1.box(200, 54, 27) wp3 = wp2.edges() wp4 = wp3.fillet(2) wp5 = wp4.faces('>Z') wp6 = wp5.workplane() wp7 = wp6.pushPoints([[-80, 0], [-48, 0], [-16, 0], [16, 0]...
We are given a CadQuery script and need to identify a target face that satisfies constraints: adjacent to another face, parallel to another face, and then among remaining candidates, choose the 6th item after sorting by descending face center Y coordinate. The candidate labels are A through F with descriptions. First,...
5
submit_label
candidate_label
"F"
false
cadqa_point_membership_73d613a36eb7ba7d
Given the CadQuery script below, classify point `[146.1622778101684, 0.0, -0.08248006601032642]` as one of inside, outside, on_boundary, or ambiguous at the target state. Answer with exactly one allowed token. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -72)) wp2 = wp1.splineApprox([[88, 0, ...
The script creates a complex shape consisting of a central swept tube (from the spline path) and many small extruded cylinders positioned at various locations around it. The query point is `[146.1622778101684, 0.0, -0.08248006601032642]`. Reasoning: - The main body is formed by sweeping a circle of radius 8 along a sp...
0
submit_text
text
outside
false
row:graph:graph:src_445c19cac5db338d:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY') wp2 = wp1.box(197, 200, 10) wp3 = cq.Workplane('XY') wp4 = wp3.box(197, 13, 3) wp5 = wp4.translate([0, -36, 3]) wp6 = cq.Workplane('XY') wp7 = wp6.box(197, 13, 3) wp8 =...
We need to identify the target workplane that satisfies constraints: "Among the remaining candidates, choose the one with the smallest workplane origin height." The problem says: "Identify the target workplane that satisfies all of these constraints: - Among the remaining candidates, choose the one with the smallest wo...
1
submit_text
text
wp11
false
row:graph:src_a7788080e2ac9568:synth:v0:parameterized_path_point_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -87)) wp2 = wp1.splineApprox([[23, 0, 0], [23, 4, 2], [21, 8, 4], [19, 12, 6], [17, 16, 8], [14, 19, 10], [10, 21, 12], [6, 22, 14], [1, 23, 16], [-3, 23,...
1) The path in the sketch is the spline wire created by splineApprox on wp1 (Workplane('XY', origin=(0,0,-87))) using the given list of 3D world coordinate points. The points are on the spline. 2) The list contains 101 points with Z coordinates from 0 to 200 inclusive in increments of 2 (e.g., (23,0,0), (23,4,2), ..., ...
3
submit_coordinate
coordinate
[21, -8, 196]
true
cadqa_feature_provenance_choice_dc0c6ee2745d0b29
Given the CadQuery script below and the described target, choose the feature ID that best matches the recorded provenance, or answer ambiguous_or_merged. Target: material region changed by wp26. Choices: wp1, wp2, wp3, wp4, wp5, wp6, wp7, wp8, wp9, wp10, wp11, wp12, wp13, wp14, wp15, wp16, wp17, wp18, wp19, wp20, wp21...
The target is "material region changed by wp26". Looking at the script, `wp26` is defined as `wp22.union(wp25)`, which adds the extruded cylinder from `wp25` to the shape. The material region added or changed by `wp26` is the result of that union operation. The feature directly created by `wp26` is the union itself...
0
submit_label
candidate_label
wp26
false
row:graph:graph:src_c39a861dbf652d38:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -2)) wp2 = wp1.box(200, 125, 18) wp3 = wp2.faces('>Z') wp4 = wp3.workplane() wp5 = wp4.rect(200, 125) wp6 = wp5.extrude(5) wp7 = wp6.faces('>Z') wp8 = wp7...
wp1 origin=(0,0,-2) height=-2; all others have heights 7 or 12. Thus smallest is -2; among candidates (all workplanes), wp1 is the one with explicitly set negative origin. {"name": "submit_text", "arguments": {"text": "wp1"}}
1
submit_text
text
wp1
true
row:graph:src_cac1d4a39cbe829e:synth:v1:parameterized_feature_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -85)) wp2 = wp1.splineApprox([[93, 0, 0], [89, 24, 3], [81, 46, 7], [66, 65, 10], [48, 79, 14], [26, 89, 17], [2, 93, 20], [-21, 90, 24], [-44, 82, 27], [...
Volume candidates: wp5 (cylinder, π*85^2*171≈3.88e6), wp9 (swept solid, area 15*15=225 times path length ~1202 → 270,450), wp10 (union > wp5). Sorted smallest→largest: wp9, wp5, wp10. 2nd smallest = wp5. {"name": "submit_text", "arguments": {"text": "wp5"}}. But before finalizing, ensure that the reasoning is correct...
2
submit_text
text
wp5
true
row:graph:src_a564b36fe441dc21:synth:v1:parameterized_face_spatial_relation_chain_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XZ') wp2 = wp1.moveTo(75, 0) wp3 = wp2.circle(25) wp4 = wp3.revolve(360, [0, 0, 0], [0, 2, 0]) wp5 = cq.Workplane('XZ') wp6 = wp5.polyline([[50, 16], [95, 30], [89, 46], [46...
We need to identify the target face that satisfies two constraints: 1. Has another face parallel to it. 2. Has another face opposite to it. Each candidate here seems to be a face from one of the extruded polyline shapes (before cutting them from the revolved base). Since all extrusions are in the Y direction (b...
5
submit_label
candidate_label
"D"
false
row:graph:graph:src_7fe7c16131f2b4ef:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -1)) wp2 = wp1.box(200, 103, 19) wp3 = wp2.faces('>Z') wp4 = wp3.workplane() wp5 = wp4.rect(200, 103) wp6 = wp5.extrude(2) wp7 = wp6.faces('>Z') wp8 = wp7...
Workplane origins: wp1 at (0,0,-1) height -1; wp4 at (0,0,8.5); wp8 at (0,0,10.5); wp12 at (0,0,10.5); wp17 at (0,0,8.5); wp21 at (0,0,8.5). The smallest is wp1 with height -1. {"name": "submit_text", "arguments": {"text": "wp1"}}
1
submit_text
text
wp1
true
row:graph:src_e0877e953bfb1632:synth:v0:parameterized_void_region_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('YZ', origin=(-100, 0, 0)) wp2 = wp1.circle(42) wp3 = wp2.extrude(131) wp4 = cq.Workplane(cq.Plane(origin=cq.Vector(-34, 0, 42), xDir=cq.Vector(0, 3, 0), normal=cq.Vector(3, ...
We need to find the cut operations that remove material and rank them by the volume of material removed. First, identify all cut operations in the script: 1. `wp24 = wp12.cut(wp23)` — wp12 is a cylinder of radius 57 extruded -21 (thickness 21), and wp23 is the union of two cylinders (radius 42, extruded -21) placed a...
3
submit_text
text
"wp40"
false
row:graph:src_32040cbd5b685437:synth:v0:parameterized_face_spatial_relation_chain_rank
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY') wp2 = wp1.box(200, 13, 8) wp3 = wp2.edges() wp4 = wp3.fillet(1) wp5 = cq.Workplane('XY') wp6 = wp5.transformed(offset=[-88, 14, 0], rotate=[0, 0, 269]) wp7 = wp6.polygo...
We need to understand the script. The script creates a box and then cuts many extruded hexagons (polygon(6,6) likely a regular hexagon with side length? Actually polygon(6,6) means 6 sides and radius 6? Possibly a hexagon. Then it extrudes both ways (both=True) so it cuts through the box. The cuts are at various positi...
5
submit_label
candidate_label
"E"
false
row:graph:src_a9645b50d0f33d97:synth:v0:parameterized_feature_extreme
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -85)) wp2 = wp1.splineApprox([[92, 0, 0], [80, 45, 7], [47, 79, 14], [2, 92, 20], [-43, 81, 27], [-78, 49, 34], [-92, 5, 41], [-82, -41, 48], [-51, -77, 5...
Only solid features have meaningful face counts: wp5 is the extruded cylinder (radius 85, height 170) with 3 faces; wp9 is the rectangular-sweep solid with more faces (6); wp10 is the union of wp5 and wp9, so its face count is at least that of wp5. Wires (wp2, wp4, wp8) have 0 faces. Among remaining solid candidates, t...
2
submit_text
text
wp5
true
row:graph:graph:src_b966c1e81b88b84b:q1:argmax_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -61)) wp2 = wp1.splineApprox([[88, 0, 0], [77, 42, 5], [47, 74, 10], [6, 88, 15], [-37, 79, 20], [-71, 52, 25], [-87, 11, 30], [-82, -32, 34], [-56, -68, ...
Let’s go step by step. We are asked: among the candidate workplanes, pick the one with the **greatest workplane origin height** (the highest Z-coordinate of the origin). List the workplane definitions and their origins: 1. **wp1** – Workplane('XY', origin=(0, 0, -61)) Origin: (0, 0, -61) → height = -61 2. **...
1
submit_text
text
wp9
false
cadqa_feature_provenance_choice_f1aab3733ed744c0
Given the CadQuery script below and the described target, choose the feature ID that best matches the recorded provenance, or answer ambiguous_or_merged. Target: material region changed by wp10. Choices: wp1, wp2, wp3, wp4, wp5, wp6, wp7, wp8, wp9, wp10, wp11, wp12, wp13, wp14, wp15, wp16, wp17, wp18, wp19, wp20, wp21...
The target is "material region changed by wp10". Looking at the script: - `wp10 = wp6.union(wp9)` creates a union of the main swept body (`wp6`) and the first extruded cylinder (`wp9`). - There is no further modification of `wp10` itself after that line; the next operations use `wp10` as the base for further unio...
0
submit_label
candidate_label
wp10
false
row:graph:graph:src_52502a9d163c189b:q1:argmin_origin_z_workplane
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XZ', origin=(-1, -14, 0)) wp2 = wp1.moveTo(80, 0) wp3 = wp2.circle(21) wp4 = wp3.revolve(16, [0, 0, 0], [0, 2, 0]) wp5 = wp4.rotate([-1, -14, 0], [-1, -12, 0], 28) wp6 = wp4...
We need to identify which workplane among `wp1` to `wp28` has the smallest workplane origin height. Looking at the script, all workplanes after `wp1` are derived via operations like `moveTo`, `circle`, `revolve`, `rotate`, and `union`. The origin height refers to the Y-coordinate (or Z in some orientations) of the w...
1
submit_text
text
wp1
false
row:graph:src_4f329454f966c395:synth:v4:parameterized_feature_extreme
Given the CadQuery script below, answer the question by reasoning about the CAD construction. ```python import cadquery as cq wp1 = cq.Workplane('XY', origin=(0, 0, -78)) wp2 = wp1.splineApprox([[89, 0, 0], [65, 61, 6], [6, 89, 12], [-57, 68, 19], [-88, 11, 25], [-72, -52, 31], [-17, -87, 37], [48, -75, 44], [86, -22,...
Let's walk through the script step-by-step. 1. `wp1` is a Workplane in XY with origin at (0,0,-78), but nothing is built on it yet — just defines the plane. 2. `wp2` is a splineApprox wire generated from many 3D points on `wp1`'s plane, with `makeWire=True`. 3. `wp3` switches to XZ plane at (0,0,-78). 4. `wp4` centers...
2
submit_text
text
wp9
false
End of preview. Expand in Data Studio

CAD-QA RL Training Set (2000 samples)

Verifiable-reward RL training rows for CAD geometry reasoning, derived from the CAD-QA benchmark (release_10k train split, cad_browsecomp difficulty). Each row is a closed-book question: a CadQuery script + a question about the resulting geometry. The gold answer is exact-match verifiable.

Files

  • data/train/ — 2000 rows
  • data/validation/ — 98 rows (the benchmark's eval-slice rows; excluded from training — do not train on this file)

Schema

field type meaning
row_id string unique row id
prompt string the closed-book prompt (CadQuery script + question)
answer_format string text | candidate_label | coordinate
gold_answer string exact-match target, JSON-encoded ("wp15", "A",
"[1.0, 2.0, 3.0]"); decode with json.loads before comparing
difficulty_bucket string cad_browsecomp
logical_hop_count int reasoning hops
geometry_hop_count int geometry hops
relation_families list[str] relation families exercised

Usage

from datasets import load_dataset
ds = load_dataset("MRiabov/cadqa-rl-2000")  # splits: train (2000), validation (98)
# reward: exact match on json.loads(gold_answer) after stripping the completion
# (see docs/rl_runbook.md in the CAD-QA repo)

Reward contract

Score = 1.0 iff parse_answer(completion, answer_format) == gold_answer. text: raw string; candidate_label: raw token (A/B/C/…); coordinate: 3-tuple of floats rounded to 6 decimals. Unparseable output = 0.0.

Provenance

Synthetic, generated by the CAD-QA pipeline (AST/dataflow graph + CadQuery execution) from the public CADEvolve source corpus. No human or personal data.

Downloads last month
72