Hostile Validation Evidence

Hostile-validation evidence archive

The hash index points to the original review receipts and supporting manifests. It includes Phase0–6, administration, PWA, maintenance, UI documentation and the ongoing UI implementation review. Original files are retained in place. JSONL record hashes include their original line endings and preserve rejected, superseded and approved rounds without filtering by outcome.

The current UI review history is append-only. Earlier rounds reconstructed from available conversation summaries are expressly labeled; missing transcripts, source hashes and scores are not invented. Append corrections and later dispositions as new events. Never edit an old approval to imply it covers later source changes.

Before submission, freeze the exact source/configuration/package/assets and relevant test inventory; record its immutable commit/tree or complete content-hash manifest. Join each reported test execution and final hostile review to that exact inventory. Preserve failing and superseded evidence alongside passing runs. Review changes after a tested snapshot before extending approval to them. Refresh the index after the final append and freeze its digest with the submission artifacts; the index cannot hash itself. A changed receipt hash is a reason to inspect the append history, not to discard earlier records.

This index is an archive inventory, not approval of the current application, entrant eligibility, contest entry or deployment. Historical approvals retain their original scope and limitations. At initial archive creation, the UI review required race repairs and matching evidence; later dated dispositions below supersede that status. The current actual multiplayer UI obligation covers every player count 2 through 10 under the 31 October release amendment; earlier 15/25-player obligations and failures remain historical. The target is 9 eight-round games, 54 player participations and 432 accepted solves, with 25 observers unchanged and maximum audience 35. All 28 changed QBrain criteria remain unsatisfied pending fresh evidence. Disposable SQL service games do not establish that UI result. Mock HTTP browser fixtures, actual database tests, real transport tests, visual inspection and physical-phone/provider validation must remain distinct evidence tiers. No physical-device or authenticated-mission result is inferred.

Only repository review artifacts and supporting manifests are indexed. Do not add private session logs, runtime credentials, database URLs containing secrets, browser profiles or raw core dumps to submission material. Existing receipts remain unchanged; links to local runtime evidence are historical references and do not imply that private files should be copied into the submission.

Review update: repaired races

The subsequent ui-source-rereview-races-repaired event records that source inspection resolved the two outstanding races. This supersedes the earlier race-repair status above without erasing it. Final approval still awaits complete browser/build evidence, synchronized documentation/readback and a frozen source/evidence snapshot.

Final-evidence review update

The ui-final-evidence-room-error-finding event preserves a new rejection for UI-HV-007: raw RoomSession parse/network exception rendering. Prior race fixes remain resolved. Final approval awaits that repair, matching source/evidence and export linkage; earlier status paragraphs are historical.

Scoped implementation approval

The ui-implementation-final-approval event resolves UI-HV-007 and approves the UI extension at99accuracy/99completeness for the234-file manifest with SHA256 2728c24a17214fa06d1e4de9af4742f1a466ba736e922e88bb6a6c932784b0f8. Earlier rejections remain preserved. This supersedes the implementation pending status above for that candidate only. Wiki publication linkage, contest readiness and real live2–25UI execution were separate obligations at that review; the player-count obligation was later amended to 2 through 15 without changing this historical approval.

Historical release amendment: 30 October 2026

The current completion plan and guest/admin contract define a guest-only public game for 2 through 15 players, a separate 25-observer limit, private GitHub administration for sharpninja, and an admin-only PWA. Retired public OpenAI sign-in, member and public-PWA reviews retain their original source scope; no old approval extends to the current release by implication.

The current scale matrix uses actual private Realtime with ordinary safety polling. Its 14 complete eight-round UI games require 119 player participations and 952 accepted solves with at most two simultaneous answers and verified cleanup. Keep strict two-player/observer push, reconnect and renewal evidence with polling suppressed separate from the scale matrix and from the actual Realtime integration lane. Historical 20/25-player failures remain available and are not rewritten as successful runs.

The frozen-source serial validation receipt records local typecheck, lint, build and native reports for 632 unit, 61 PostgreSQL and 79 browser cases at b8400dc6f7aeb42a991861ac98e36f297eae8bb3, each with zero failures/skips. It establishes its stated local checks, not the outstanding actual multiplayer lanes, a hostile verdict or hosted deployment. Later edits need their own provenance and review.

The operator's current reviewer selection is Cline with model Beam and High effort until Monday, 2 November 2026, with no exact cutoff hour specified. Verify actual runtime identity, retain full request/response JSONL and the entire verdict in a read-back MCP turn, and require independent AGREE with accuracy and completeness each at least 98 percent before tracked completion. Selection is not execution or approval. Historical Claude rounds, including the last DISAGREE at 97/95, remain preserved. Fresh actual Realtime/UI evidence, review and hosted verification are still release obligations at this amendment.

Refresh the archive inventory after new retained review receipts are appended, then bind its digest and the current contest checklist to the actual submission candidate. Neither this amendment nor any review accepts legal terms or submits the contest entry; authenticated mission/conduct content and manual entrant actions remain explicit gaps.

31 October candidate update

The observed 0e8d31a chain is retained FAILED evidence, run 27a362dc-4720-4116-8640-a1f2016ab2c7. Its full source/build lane passed 894 unit/component/contract, 63 PostgreSQL and 114 browser cases, and its PowerShell and actual Realtime lanes exited 0. Its native normal matrix has 13 passed, 1 failed, 0 skipped: 2 through 14 players completed all eight rounds, while 15 players failed during round 1 category-command acknowledgement with zero solves. Strict UI and four preparation steps did not run. Preserve the earlier failed 11-player run, the separate passing 11-player diagnostic, actual private-log capture metadata and their evidence boundaries. No new independent review, hosted result or release acceptance is asserted.

The earlier b8400dc counts above remain a dated historical baseline. The submission checklist records the operator's 31 October working-deadline instruction, preserved source-date conflict and limited shared planning/design follow-up. The 96-file historical HV inventory is separate from this later local test evidence and must be refreshed after an actual new review; no prior approval transfers to the current candidate.

Current 10-player amendment: 31 October 2026

The authoritative cap10 readback and current plan supersede the earlier 15-player execution target. New rooms/matches support 2 through 10 players, creation default 4, rules v2.8 dated 31 October 2026, and 25 observers for a maximum audience of 35. The complete service/browser matrix target is 9 games, 54 player participations and 432 accepted round solves. Existing larger running matches remain finishable and historical evidence remains unchanged. Changed criteria remain unsatisfied; current-source full/PowerShell/Realtime/matrix/strict evidence, independent Cline Beam High review, exact hosted candidate and staged deployment are not established by this documentation.

The later private API diagnostic is also failed history: 11 passed games (2 through 12), one failed 13-player readiness case and two skipped counts (14/15). Its actual native message is Malformed rendered command version. An earlier derivative misclassified adjacent quoted source as an Unknown-pending failure; the correction and both original artifacts remain retained. The causal report separates observations from holdout-removal inference and unresolved browser/HTTP scheduling attribution. Reducing the supported cap does not convert this run or the earlier 11/15-player failures into passing evidence. The original Claude round6 DISAGREE 97/95 and its FAIL/UNKNOWN findings remain unsuperseded until an actual independent acceptance review. No wiki export or publication is claimed for this amendment while current source/tests remain pending.