Michael Rabinovich Cursor commited on
Commit ·
e4e88e3
1
Parent(s): 18e0930
remove stale README (scripts-holding-folder note)
Browse filesThe README described a defunct scripts-holding-folder state and no
longer reflects the dataset. The real dataset card (HF front-matter +
fixture/input schema) is the Stage D task in space-setup/post-gt-swap.md;
drop the misleading note until that lands.
Co-authored-by: Cursor <cursoragent@cursor.com>
README.md
DELETED
|
@@ -1,32 +0,0 @@
|
|
| 1 |
-
# Dataset-curation scripts (holding folder)
|
| 2 |
-
|
| 3 |
-
These scripts validate that fixture ground-truth STEP files are well-formed.
|
| 4 |
-
They are **dataset-authoring tools**, not part of the public benchmark code.
|
| 5 |
-
|
| 6 |
-
They live here temporarily until the `cadgenbench-data` HF dataset repo is
|
| 7 |
-
created, at which point they will be moved there with `git mv`. Until then,
|
| 8 |
-
they remain in this holding folder so they are preserved in git history but
|
| 9 |
-
explicitly excluded from the `cadgenbench` package build (see
|
| 10 |
-
`pyproject.toml`'s `tool.setuptools.packages.find` exclusion).
|
| 11 |
-
|
| 12 |
-
## Contents
|
| 13 |
-
|
| 14 |
-
- `sanity_check_gt.py` — validates that each fixture's `ground_truth.step`
|
| 15 |
-
is canonical-posed, watertight, and BREP-valid.
|
| 16 |
-
- `sanity_check_jig_metric.py` — runs the interface-match metric on
|
| 17 |
-
each fixture's jig sub-volumes against its GT and asserts saturation.
|
| 18 |
-
- `sanity_check_submission.py` — validates a candidate submission directory
|
| 19 |
-
conforms to the submission spec (`docs/benchmark/submission.md`).
|
| 20 |
-
- `sanity_check_topo.py` — validates Betti-number agreement between the
|
| 21 |
-
GT BREP and its tessellation.
|
| 22 |
-
|
| 23 |
-
## Why they were moved out
|
| 24 |
-
|
| 25 |
-
These tools require the GT STEPs to run, which we may eventually keep
|
| 26 |
-
private. Coupling the public benchmark code to internal authoring
|
| 27 |
-
workflows would make a future split painful. Keeping them in this
|
| 28 |
-
holding folder makes the lift a one-command `git mv`.
|
| 29 |
-
|
| 30 |
-
Any metric-exercising logic that is genuinely useful as a regression
|
| 31 |
-
test for the metric code itself should be lifted into `tests/eval/`
|
| 32 |
-
rather than living here.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|