[Bot] Kilo-gsd second Verify gate iteration on issue #53 still failing — composition reconstruction not producing valid output #55

New issue

Closed

opened 2026-07-24 06:38:28 +00:00 by Mark · 1 comment

Mark commented

2026-07-24 06:38:28 +00:00

Owner

Summary

The kilo-gsd fix pipeline for issue #53 (compose_body_with_features boolean cut) was dispatched a second time after the initial attempt (documented in #54) was incomplete. This second iteration is also failing to pass the verification gates.

Kanban card: t_b5d173a3 (QA probe → reporter)

Current Status

Issue #53 remains OPEN on Forgejo — no fix has been deployed
kilo-gsd IS actively running (PID 3510954, second iteration) at the time of this report
No commit references #53 anywhere in git history
Unstaged WIP changes in pc2cad_reconstruct.py and pc2cad_cli.py — code has been reworked but is incomplete

Verification Gate Failures (First Iteration)

The first kilo-gsd run on issue #53 failed the Verify gate with three distinct issues:

L-bracket coverage: 0.021 (needs ≥0.8) — The body reconstruction is not capturing the L-bracket shape; reconstruction produces only ~7 faces (= a bare box with no through-hole)
OCP circular-edge checker crash — TopExp_Explorer not iterable; the verification code for circular edges raised an unhandled exception
BdPolygon import error — The code uses BdPolygon but current build123d exposes it as Polygon; outdated import

Second Iteration WIP (Unstaged)

The second run produced uncommitted changes in:

pc2cad_reconstruct.py: Reworked reconstruct_to_build123d to use Plane with proper orientation, moved body_bbox extraction before body construction, tightened _select_body extras guard
pc2cad_cli.py: Added composed reconstruction path in --mode auto to call compose_body_with_features + export_reconstruction

These changes are incomplete and untested — no commit has been made.

Reproduction Steps

Attempt to fix issue #53 (compose_body_with_features boolean cut) via the kilo-gsd pipeline
Run the Verify gate on the produced output
Observe: L-bracket coverage check fails (0.021 vs ≥0.8 threshold)
Observe: OCP circular-edge checker crashes with TopExp_Explorer not iterable
Observe: BdPolygon import error in the generated code

Expected Behavior

The kilo-gsd pipeline should produce working code that:
- Passes the L-bracket coverage check (≥0.8)
- Correctly performs boolean subtraction between body and feature geometry
- Uses correct build123d API (Polygon not BdPolygon)
- Is committed and pushed to a branch
- Results in issue #53 being closed

Actual Behavior

Pipeline completes its dispatch phase but the code it generates:
- Fails coverage checks
- Uses broken API imports
- Crashes the verification tooling
- Is never committed
Fixit card completes on dispatch, not on actual fix landing

Why This Is a Process Bug

This reveals two systemic issues:

Premature fixit completion — The fixit card completes when kilo-gsd is dispatched, not when the fix is committed and verified. This means the monitoring systems see "done" while the actual work is still in progress or failing.
Verify gate fragility — The verification gates have their own bugs (crashing checkers, wrong API imports) that block valid fixes. The gates should be hardened before being used as blocking criteria.

Issue #53: Original bug report
Issue #54: Previous "fix never landed" report for the same issue
Kanban card t_b5d173a3

## Summary The kilo-gsd fix pipeline for issue #53 (compose_body_with_features boolean cut) was dispatched a second time after the initial attempt (documented in #54) was incomplete. This second iteration is also failing to pass the verification gates. **Kanban card:** t_b5d173a3 (QA probe → reporter) ## Current Status - **Issue #53** remains **OPEN** on Forgejo — no fix has been deployed - **kilo-gsd IS actively running** (PID 3510954, second iteration) at the time of this report - **No commit references #53** anywhere in git history - **Unstaged WIP changes** in `pc2cad_reconstruct.py` and `pc2cad_cli.py` — code has been reworked but is incomplete ## Verification Gate Failures (First Iteration) The first kilo-gsd run on issue #53 failed the Verify gate with three distinct issues: 1. **L-bracket coverage: 0.021 (needs ≥0.8)** — The body reconstruction is not capturing the L-bracket shape; reconstruction produces only ~7 faces (= a bare box with no through-hole) 2. **OCP circular-edge checker crash** — `TopExp_Explorer` not iterable; the verification code for circular edges raised an unhandled exception 3. **`BdPolygon` import error** — The code uses `BdPolygon` but current build123d exposes it as `Polygon`; outdated import ## Second Iteration WIP (Unstaged) The second run produced uncommitted changes in: - `pc2cad_reconstruct.py`: Reworked `reconstruct_to_build123d` to use `Plane` with proper orientation, moved `body_bbox` extraction before body construction, tightened `_select_body` extras guard - `pc2cad_cli.py`: Added composed reconstruction path in `--mode auto` to call `compose_body_with_features` + `export_reconstruction` These changes are incomplete and untested — no commit has been made. ## Reproduction Steps 1. Attempt to fix issue #53 (compose_body_with_features boolean cut) via the kilo-gsd pipeline 2. Run the Verify gate on the produced output 3. Observe: L-bracket coverage check fails (0.021 vs ≥0.8 threshold) 4. Observe: OCP circular-edge checker crashes with `TopExp_Explorer` not iterable 5. Observe: `BdPolygon` import error in the generated code ## Expected Behavior - The kilo-gsd pipeline should produce working code that: - Passes the L-bracket coverage check (≥0.8) - Correctly performs boolean subtraction between body and feature geometry - Uses correct build123d API (Polygon not BdPolygon) - Is committed and pushed to a branch - Results in issue #53 being closed ## Actual Behavior - Pipeline completes its dispatch phase but the code it generates: - Fails coverage checks - Uses broken API imports - Crashes the verification tooling - Is never committed - Fixit card completes on dispatch, not on actual fix landing ## Why This Is a Process Bug This reveals two systemic issues: 1. **Premature fixit completion** — The fixit card completes when kilo-gsd is *dispatched*, not when the fix is committed and verified. This means the monitoring systems see "done" while the actual work is still in progress or failing. 2. **Verify gate fragility** — The verification gates have their own bugs (crashing checkers, wrong API imports) that block valid fixes. The gates should be hardened before being used as blocking criteria. ## Related - Issue #53: Original bug report - Issue #54: Previous "fix never landed" report for the same issue - Kanban card t_b5d173a3