ahnyongjooandClaude Sonnet 5.5
964a841de1
- Code\rebar_pipeline, review_pipeline, pier-program을 정션에서 실제 폴더로 전환 - Code\rhino_mcp 및 CLAUDE_rhino배경.md 추가, .mcp.json은 PYTHONPATH로 새 위치 사용 - pier-program 내부 .git 제거(이력은 backup\pier-program_20261028.bundle에 보관) - worktree 가이드 반영(CLAUDE.local.md, HANDOFF.md) - D:\Rhino-MCP 경로 참조를 새 위치로 치환 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
4.5 KiB
R3/R4 and parapet report audit — 2026-09-15
This is the earlier, pre-zigzag snapshot. The subsequent six-model local-only results and remaining wing warnings are recorded in R34_SIX_MODEL_AUDIT_20260915.md.
Applied results
All changes use shared source-reading, placement and validation code. No job ID, drawing hash, fixed model quantity or copied geometry selects a pose. Drawing-count distribution remains an explicitly authorized review policy, not literal reconstruction of authored stations or engineering approval.
| Saved model | R3/R4 audit and publication |
|---|---|
| f37b4289c97298fed715a88d | Added B1 R3/R4 62 each, B2 R3/R4 58 each; all four groups pass |
| 2120ce4675c6c0e6340aaac8 | Added B1 R3/R4 64 each, B2 R3/R4 62 each; all four groups pass |
| 03dd556bb61ab4adec0b670d | Replaced B2 R3 78; preserved unrelated rows. B1/B3 R3 fail raw/rounded containment; R4 section axis remains ambiguous |
| dc0a31e7e7d7fda5afb703bd | No publication: R3 section axis ambiguous; R4 only half of proposed paths pass enclosure check |
| c37ca61ba994edf2547dcd78 | No publication: R3 rounded contact fails on some instances; R4 section axis ambiguous |
| 23d1a02df0d24b7bb0368d6f | Restored model has R3/R4 rows; fresh source seed inventory is empty. Existing rows preserved, not marked newly verified |
| 4ea7ba277676614b64ce3089 | Restored model has R3/R4 rows; fresh source seed inventory is empty. Existing rows preserved, not marked newly verified |
| 10d3dc5c7463012bd0dac27f | Failed saved job: B2 R3 candidate passes; other rows fail enclosure/contact/axis. No publication into failed job |
| d0701131f1f87d687322a4f4 | Existing R3/R4 retained: B1/B2 78 each, B3 18 each. Previously checked enclosure, not hook/engineering approval |
Nine groups / 570 bars were added or replaced in the three successful target models. Publication rechecks current hosts, actual target bars, raw and rounded tube containment, contact, physical enclosure, lengths and counts. Original unrelated concrete and reinforcement are preserved; each save has a backup. An incidental pair of surface contacts is not sufficient enclosure evidence.
Parapet report defect
The actual report worker reproduced START_PARAPET NA1 as 20 exposed bars even though the sampled backend tube test reported zero. The stored parapet bottom and roof top used differently rounded elevations: maximum contact difference 0.00000039190000045 m. Continuous interval inspection correctly detected that numerical gap. No exposure-length threshold or mark-specific exemption was added.
Shared contact normalization now closes only uniquely matched planar contacts within 0.000001 m, checking corners and interior/edge probes. It keeps the roof, parapet top and every reinforcement path unchanged. Actual 1 mm gaps and ambiguous hosts remain unchanged. Mesh-based parapets now use this same common contact rule, alongside the existing ruled-surface contact rule.
d070 start/end contact meshes were updated. All 32 parapet groups in the eight saved models with parapet bars now pass the actual report worker with zero exposed bars. The live 8097 page also returned all four d070 parapet rows as pass. An old report window is an immutable snapshot: reload the model and reopen the report to obtain the corrected geometry.
Parapet anchorage and longitudinal generation now independently check the actual viewer-rounded paths against their explicit parapet/roof hosts.
Wing workflow retained in this release
The common current-source wing stage publishes strict validated groups plus explicitly unapproved review groups using the existing wing workflow. d070 has 52 wing groups across four occurrences (13 each), with all 79 existing body groups preserved. WR2/WR7/WH1/WR11 retain anchorage exposure warnings. This is not complete wing geometry/anchorage approval.
Verification and unresolved scope
- Shared mesh contact tests: closure, idempotence, real-gap preservation, ambiguous-host rejection.
- Body quantity tests: explicit source authorization, shared topology-based dispatch, no alternate-policy fallback, rounded contacts and enclosure gate.
- Existing wing review, progressive preservation, fresh rebar, parapet and wing module regression tests.
- JavaScript quantity report and placement editor regression scripts pass.
- AirExposure regression suite: 21 checks pass; real gaps remain detectable.
- R3/R4 failures listed above and remaining wing anchorage warnings stay open. This release does not claim that every model or every reinforcement row is approved.