IMP-15 Step 14 visual_check 보강 #15

New Issue

2026-05-12T17:50:19+09:00

Kyeongmin commented

2026-05-12 17:50:19 +09:00

IMP-15 Step 14 visual_check 보강 (parent / decomposition issue)

Status

CLOSED — 2026-05-19. All 5 sub-issues delivered, integrated, and verified on main.
Verification baseline: pytest -q tests = 200 passed, 0 failed at HEAD 614c533.

This issue's original Stage 2 monolithic plan grew too large (~29 KB) for single-pass
Codex verification (Codex hang during Stage 2 review). Decomposed into 5 execution
sub-issues, each scoped + testable + with its own Codex verification.

Execution checklist (locked SHAs)

#45 [IMP-15 실행-1] image_aspect_mismatch detection + runtime test — e9b3d2e
#46 [IMP-15 실행-2] table overflow + element-identity dedup + Selenium test — 2827622 (IMP-16 prefix = known naming quirk; lineage = IMP-15 실행-2)
#47 [IMP-15 실행-3] classifier consumer (image + table) + pure-dict test — 535c484
#48 [IMP-15 실행-4] debug.json + spec doc trace + regression — 614c533
#49 [IMP-15 실행-5] final integration + parent close — verification-only (no new SHA)

Final integration evidence

HEAD = 614c53358ecaa92e9f6fc979b2399da028e15d41 (branch main)
pytest -q tests = 200 passed in ~40.5s, 0 failed (vs pre-#15 baseline 139, issue #49 acceptance ≥143)
Step 14 visual_check spot-check on samples/mdx/03* (run issue49_u2_spotcheck):
- top-level image_events / table_events keys surfaced from visual_runtime_check ✓
- docs/architecture/PHASE-Z-FIT-CLASSIFIER-ROUTER-SPEC.md has the image_aspect_mismatch Post-render taxonomy row (excluded from §3.2 priority / §4 router map by design) ✓
- empty event payloads on MDX 03 expected (no image/table content_objects with overflow); schema lift is the surface, not the payload

Out-of-scope follow-up candidates (kept out of #15)

Router action for image_aspect_mismatch (separate follow-up)
Step 21 viewer surfacing of table_self_overflow (separate follow-up)

Decomposition rationale

Stage 2 monolithic verification of a ~29 KB plan was unreliable in orchestrator runs
Codex hung during verification (workload too large for single-pass review)
Each sub-issue keeps its plan ≤ ~5 KB → Codex verification feasible
Each sub-issue must include its own tests (no "tests:[] defer to later unit" pattern)

Original scope (from issue body)

관련 step: Step 14, 21
source: INSIGHT-MAP §3 (H1 Reference Only)
priority: medium

scope:

image_aspect_mismatch / tabular_overflow 검사 추가
Step 14 의 image / table 검사 부재 보강

guardrail / validation:

AI/Kei classification 회귀 X
deterministic 검사 + trace

dependency: soft link: IMP-01 (Step 14 측정/trace layer 공유)

cross-ref:

review loop:

Codex 1차 review
Claude 재검토
Codex 재검증
scope-locked
ready-for-implementation
implemented
verified

References

Backup of original 11 comments: D:/tmp/issue15_backup_20260518_1521/
Codex #4 NO findings extract: D:/tmp/issue15_backup_20260518_1521/drafts/step2_codex4_findings.md
Claude #6 plan (u1a~u6) extract: D:/tmp/issue15_backup_20260518_1521/drafts/step2_claude6_plan_extract.md

# IMP-15 Step 14 visual_check 보강 (parent / decomposition issue) ## Status **CLOSED — 2026-05-19**. All 5 sub-issues delivered, integrated, and verified on `main`. Verification baseline: `pytest -q tests` = **200 passed, 0 failed** at HEAD `614c533`. This issue's original Stage 2 monolithic plan grew too large (~29 KB) for single-pass Codex verification (Codex hang during Stage 2 review). Decomposed into 5 execution sub-issues, each scoped + testable + with its own Codex verification. ## Execution checklist (locked SHAs) - [x] #45 [IMP-15 실행-1] image_aspect_mismatch detection + runtime test — `e9b3d2e` - [x] #46 [IMP-15 실행-2] table overflow + element-identity dedup + Selenium test — `2827622` *(IMP-16 prefix = known naming quirk; lineage = IMP-15 실행-2)* - [x] #47 [IMP-15 실행-3] classifier consumer (image + table) + pure-dict test — `535c484` - [x] #48 [IMP-15 실행-4] debug.json + spec doc trace + regression — `614c533` - [x] #49 [IMP-15 실행-5] final integration + parent close — verification-only (no new SHA) ## Final integration evidence - HEAD = `614c53358ecaa92e9f6fc979b2399da028e15d41` (branch `main`) - `pytest -q tests` = **200 passed in ~40.5s, 0 failed** (vs pre-#15 baseline 139, issue #49 acceptance ≥143) - Step 14 visual_check spot-check on `samples/mdx/03*` (run `issue49_u2_spotcheck`): - top-level `image_events` / `table_events` keys surfaced from `visual_runtime_check` ✓ - `docs/architecture/PHASE-Z-FIT-CLASSIFIER-ROUTER-SPEC.md` has the `image_aspect_mismatch` Post-render taxonomy row (excluded from §3.2 priority / §4 router map by design) ✓ - empty event payloads on MDX 03 expected (no image/table content_objects with overflow); schema lift is the surface, not the payload ## Out-of-scope follow-up candidates (kept out of #15) - Router action for `image_aspect_mismatch` (separate follow-up) - Step 21 viewer surfacing of `table_self_overflow` (separate follow-up) ## Decomposition rationale - Stage 2 monolithic verification of a ~29 KB plan was unreliable in orchestrator runs - Codex hung during verification (workload too large for single-pass review) - Each sub-issue keeps its plan ≤ ~5 KB → Codex verification feasible - Each sub-issue must include its own tests (no "tests:[] defer to later unit" pattern) ## Original scope (from issue body) **관련 step**: Step 14, 21 **source**: INSIGHT-MAP §3 (H1 Reference Only) **priority**: medium **scope**: - image_aspect_mismatch / tabular_overflow 검사 추가 - Step 14 의 image / table 검사 부재 보강 **guardrail / validation**: - AI/Kei classification 회귀 X - deterministic 검사 + trace **dependency**: `soft link: IMP-01` (Step 14 측정/trace layer 공유) **cross-ref**: - [backlog §2 IMP-15](https://gitea.hmac.kr/Kyeongmin/C.E.L_Slide_test2/src/branch/main/docs/architecture/PHASE-Z-IMPLEMENTATION-ISSUE-BACKLOG.md) - [INSIGHT-MAP §3 Step 14 / Step 21](https://gitea.hmac.kr/Kyeongmin/C.E.L_Slide_test2/src/branch/main/docs/architecture/PHASE-Q-INSIGHT-TO-22STEP-MAP.md) - [22-step pipeline Step 14 / Step 21](https://gitea.hmac.kr/Kyeongmin/C.E.L_Slide_test2/src/branch/main/docs/architecture/PHASE-Z-PIPELINE-OVERVIEW.md) **review loop**: - [x] Codex 1차 review - [x] Claude 재검토 - [x] Codex 재검증 - [x] scope-locked - [x] ready-for-implementation - [x] implemented - [x] verified ## References - Backup of original 11 comments: `D:/tmp/issue15_backup_20260518_1521/` - Codex #4 NO findings extract: `D:/tmp/issue15_backup_20260518_1521/drafts/step2_codex4_findings.md` - Claude #6 plan (u1a~u6) extract: `D:/tmp/issue15_backup_20260518_1521/drafts/step2_claude6_plan_extract.md`

Kyeongmin added the needs-codex-review label 2026-05-12 17:50:19 +09:00

Kyeongmin added this to the B-2 §2 separate axis registry (9) milestone 2026-05-12 18:16:12 +09:00