Report
Status: DONE
Result
The reported 17-frame suppression failure was a browser-check diagnosis error: !!frame.contentDocument.body also returns true for about:blank, and scrolling an iframe inside a closed <details> does not open it. After opening all intended details and waiting for each final same-origin document, all 29 child widgets were present and suppressed. The real deployed defect is the outer layout runtime overriding the comment toolbar from position:fixed to position:relative inside a document-positioned host. The staged source now keeps the toolbar fixed at the viewport top and makes the adapter wait for a matching final frame document.
Deliverables and checks
| Deliverable | Requirement or verification | Result | Evidence |
|---|---|---|---|
| staged master | Preserve the September document identity, 29 iframe paths, 267 unique critique targets, and original child widgets | PASS | source repair proof confirms the master ID and counts. All 29 child HTML files match their pre-repair SHA-256 hashes. |
| builder | Generate the viewport-fixed toolbar and attach suppression only to final, matching same-origin documents | PASS | python3 -m py_compile outputs/rita-journey-20261002/legacy/build_legacy_review.py passed. The generator replaces an existing adapter and skips an already assembled critique index, avoiding duplicate IDs and duplicate refresh listeners. |
| adapter fixture | Reject body-bearing about:blank, wait through navigation readiness, suppress one real same-origin document, and prevent inner shortcut re-enable |
PASS | node outputs/rita-journey-20261002/legacy/test_adapter.cjs passed. The fixture leaves cross-origin and wrong-path documents unchanged. No live provider record was created. |
| live diagnosis | Independently inspect final document URL, ready state, content title, suppression, and live toolbar geometry | PASS | The live master returned HTTP 200. After opening all details, 29/29 actual documents matched their staged source titles and had hidden roots plus suppression markers. Before the source fix, the toolbar was at y=8 when scrollY=0 and y=-33,715 at scrollY=33,723. All non-read requests were guarded; none were attempted. |
| staged browser metadata | Verify the repaired assembled page in a real headless browser | PASS | 29/29 final local iframe documents, roots, document markers, and shadow markers were present. All 29 roots computed to display:none, visibility:hidden, and pointer-events:none. After Ctrl+Space in the first child, aria-pressed remained false. The standalone persona widget remained visible. No non-read request was attempted. |
| header screenshot | Outer toolbar at header | PASS | 1440×1000, DPR 1, scrollY 0, fixed at y=8 and within the viewport. |
| critique screenshot | Outer toolbar at the deep critique target | PASS | 1440×1000, DPR 1, scrollY 33,723, still fixed at y=8 and within the viewport. |
| standalone persona screenshot | Preserve the original customer child toolbar | PASS | 390×844, DPR 1, scrollY 0. |
Limits or consequential decisions
No live publication or mirror was performed. The parent owns deployment and actual post-deployment screenshots. Provider persistence was not tested. The direct comment service quota state and same-origin fallback remain as previously documented.
The full builder could not be executed against its current SOURCE because /Users/agency/.agents/cf-review/komplex-rita-funnel-2026-09-26-v1/materials/content.json is absent, and the source index.html was absent when rechecked. I applied the builder's literal layout and adapter changes to the already assembled staged master using patch_existing_stage.py, verified that markup outside those runtimes stayed byte-identical after removing the duplicate generated critique copy, then ran the local browser verification. The three standalone screenshot viewports were closed with the headless browser; no user-facing tab was opened.
The delegated runtime exposed GPT-6, not Luna, and did not expose an exact variant or reasoning-effort label. I did not relabel the executor as Luna/max.