Turn the Queen dashboard into a one-screen strategy cabinet

[gHashTag]

Objective

Turn the existing Queen control room into a one-screen web strategy cabinet in the original T27 visual language. Use the already shipped React Three Fiber research city as the game scene; do not add a Unity WebGL runtime.

Observable contract

Desktop

• At a viewport of at least 1280x720, the Queen route presents one bounded command cabinet with a top status bar, strategic view controls, a central operational viewport, a live Bee fleet rail, and an activity/review HUD.
• Kanban, Mission Map, Factory/Research City, and Technology Tree remain switchable views of the same live public ledger.
• The primary cabinet fits one viewport; large datasets scroll inside their own panels rather than turning the route into a long marketing page.

Bee fleet

• Render eight physical hangar positions because the verified runtime policy supports a maximum of eight.
• Only provider-backed slots may appear busy or idle. Positions above the live workers.capacity must be labelled locked/unpowered and must not be counted as workers.
• Keep canonical issue links, live activity, review counts, and the 5-second/2-second polling contracts.

Visual language

• Preserve the original TRINITY logo, black/green/gold palette, triangular geometry, monospace telemetry, and T27 typography.
• Borrow strategy-game information architecture, not third-party artwork, terminology, or trade dress.

Accessibility and responsive behavior

• RU and EN labels remain equivalent.
• Controls are keyboard operable and expose pressed/selected state.
• Mobile uses a stacked cabinet with horizontal/contained panel scrolling.
• prefers-reduced-motion: reduce removes ambient orbit, scan, and construction motion while retaining all information.

Non-regression

• No fake tasks, synthetic Bees, private messages, credentials, or visual motion presented as throughput.
• Public-ledger-only boundary remains intact.
• Existing Queen factory, research-city, language, cycle-geometry, hardware-foundry, typecheck, and production build gates pass.

Boundary

apps/website/src/pages/Queen.tsx
apps/website/src/pages/Queen.css
apps/website/src/components/QueenFactory.tsx
apps/website/qa/queen-game-cabinet-contract.mjs
apps/website/package.json

[gHashTag]

TDD RED recorded before production changes.

Command: cd apps/website && node qa/queen-game-cabinet-contract.mjs
Result: exit 1. The contract reports the expected missing behaviors: five cabinet views, default Factory view, bounded one-screen shell, persistent game menu, central viewport, live HUD/ticker, localized locked hangars, mobile/reduced-motion rules, and the executable eight-hangar model. No production UI code had been changed at this point.

[gHashTag]

Hive coverage provenance — continuation of #960

Observed / scope

Baseline: 323b569024e5b7e71ffbcb448e4db789010b68f7.
PR #960 is merged. The module snapshot names gHashTag/trios, while the
coverage hook selects trinity. The manifest does not list trios. Folder
prefixes and punctuation-stripped basenames can nevertheless turn cells yellow.
This is one bounded correction to coverage provenance, not a migration or deploy.

Observable contract (before implementation)

1. Coverage belongs to the module snapshot's repository, never the board's
   repository or a hard-coded other repository. Missing repo or missing corpus
   is UNKNOWN, not a measured zero or manual-code verdict.
2. Yellow requires an exact repository-relative module path explicitly claimed
   by a .t27 manifest row in that repository. Neither folder naming, a spec
   display name, another module's basename nor punctuation removal is evidence.
3. A known corpus with no exact claim leaves a module red (migration debt,
   not proof of how all its code was authored). An empty cell stays blue.
   Unknown coverage is blue with an explicit RU/EN unknown label, not red/yellow.
4. A repository change must not retain another repository's claims. A changed
   claim set of the same size must invalidate the scene's coverage signature.
5. Keep the original logo, facing wall, retina scaling, honey hover/tap lift,
   live event cards, all navigation modes and public-read-only boundary.

Plan

• Inspect merged GitHub work and current artifact; isolate clean checkout.
• Record this contract before production changes.
• Reproduce false yellow with an executable RED test.
• Implement repository-scoped parsing and exact matching; GREEN tests.
• Build and verify the exact artifact on desktop/mobile/reduced motion.
• Independent read-only review; record evidence and remaining risks.

Boundaries

Manifest coverage is SOURCE-CLAIM, not generated-code parity or Queen acceptance.
Module and closed-issue layers share positions, not identities: do not label an
unrelated closed issue as covered merely because a module occupies that position.
No production publication, scheduler, worker capacity, secret or KIE changes.
The old workspace TRI runtime is unavailable (iCloud dataless tri_core.py;
bounded ./tri factory --json exited 142 after five seconds). Do not rewrite its
history or report a completed TRI wave without running it.

[gHashTag]

🧪 [TEST] Coverage provenance TDD: before production changes the new executable contract failed 10/12 checks (directory heuristics, basename collision, punctuation identity, exact-path claim, missing unknown state). The first minimal fix passes 24 checks. Scope: gHashTag/trios module snapshot must not borrow trinity manifest coverage. Clean isolated checkout at main 323b569; build, browser checks and independent review are next. No deployment or secret changes.

[gHashTag]

🔍 [RESEARCH] Independent review caught a deeper schema mismatch: the sync producer parses module as a display label, not a repository path. Fixed the consumer to require an explicit versioned modulePath mapping; the legacy corpus is UNKNOWN. Also validated canonical spec paths (including primary t27 unprefixed paths) and clear stale hover text on scene teardown. Review regressions: 10/34 RED before fix, now 34/34 GREEN; honesty 113/113. Fresh build and re-review underway. No fabricated trios coverage and no production changes.

[gHashTag]

Queen hive continuation — 2026-09-07

Outcome

Inspected merged agent work (PR #960 and subsequent main at
323b569024e5b7e71ffbcb448e4db789010b68f7). Continued with one verified local
correction: repository-scoped, explicit coverage provenance. No deployment,
branch push, worker-capacity change, secret change or scheduler was performed.

OBSERVED: the production field shows the gHashTag/trios snapshot but labels
2 module cells T27-covered and 113 manual. The old hook borrowed trinity
claims; path-name heuristics could create yellow without evidence. The producer
also uses module as a display label, not a mapping to a code directory.

Now the consumer requires a versioned explicit modulePath mapping. Legacy or
unavailable corpus data remains UNKNOWN, visibly localized in RU/EN. No names,
basenames or folder prefixes manufacture coverage. Current local result:
115 module cells UNKNOWN, zero falsely yellow/manual cells. Empty cells stay
blue. Hover text names the module separately from the colocated closed issue;
scene changes clear previous coverage text.

SOURCE-CLAIM: even a future valid manifest is not proof of generated-code parity
or Queen acceptance. TARGET: teach the producer to emit evidence-backed mappings
for trios; this patch does not fabricate those mappings.

Verification

Working directory: apps/website in this isolated checkout.

• npm ci --no-audit --no-fund: 247 packages installed.
• npm run check:queen-coverage: initial RED 10/12; review regressions RED10/34;
  final GREEN 34 checks.
• npm run check:queen-honesty: 113 checks.
• npm run check:queen-languages: 246 EN / 246 RU keys.
• npx eslint src/components/queenHud.ts src/components/QueenCombBabylon.tsx src/pages/Queen.tsx: exit0.
• npm run typecheck:ratchet: no increase; 179 existing errors in26 files.
• npm run build: exit0, final build20.42s; inherited font/import/chunk warnings.
• git diff --check: exit0.
• Independent read-only review: initial blockers reproduced and fixed; final
  review has no blocking findings, tests and independent probes passed.
• BrowserOS Neo on the relaunched production preview: desktop1440×1000 DPR2,
  mobile390×844 DPR2 and reduced-motion media. Correct unknown legend, no
  horizontal overflow, original logo/facing wall/models0,12 event cards.
  Hover names issue/module; language rebuild clears the old tooltip. A120ms
  touch gesture selects the module. Desktop/mobile screenshots inspected.

Exact artifact: index-BnanlRBx.js, Queen-B3tb7_0h.js,
QueenCombBabylon-CeHqHYyD.js, Queen-i3Cbtdey.css.
SHA256 dist/index.html:
ccb22ae3bcf9fed4aa9c248d52601487343c3963b5b09948858c4d60c6fe6413.
SHA256 Babylon chunk:
bfdc8727cf3563881909a95278206eb5541a9eb2b2a88bdfdd813659fb0211c9.

Limits and resume point

Preview: http://127.0.0.1:4178/#/queen. These changes are local and uncommitted.
The user's older dirty worktrees and append-only ledgers were preserved.

The root workspace's tri_core.py and state files are iCloud dataless.
Both ./tri factory --json and ./tri doctor --json, run with a5-second bound,
timed out (exit142). No tri wave or tri learn completion is claimed. This
checkout's evidence JSONL is the continuation record; do not reconstruct or
overwrite inaccessible historical ledgers.

The first zero-duration touch injection did not produce a pick; the subsequent
120ms gesture did. Reduced-motion layout renders correctly, but the existing
Babylon drift is unconditional: do not claim full reduced-motion compliance.
The live worker readout was0/4 during inspection; this correction does not launch
Bees and visual events are not worker throughput.

Next continuation starts from this checkout, this contract and the evidence
JSONL. Do not replay the old cabinet prototype or overwrite the merged hive.

Three distinct next waves

1. Real migration evidence: generate a versioned trios module-path ledger
   from specifications, generated artifacts and test/review proof.
2. Bee execution and review: diagnose the observed0/4 and waiting-review
   state from scheduler decisions and real worker logs; fix one dispatch blocker.
3. Hive interaction/accessibility: stop canvas drift under reduced motion
   and verify touch/hover lifecycle across repeated scene rebuilds.

[gHashTag]

Issue displays and close-up — #890

One vertical slice: a hive cell is a readable display of an actual GitHub issue
or epic, with exact-number events and operator-controlled close-up.

Research / design decision

• Babylon GUI has a separate texture resolution/renderScale; a256px event
  texture enlarged with the scene cannot recover missing text detail.
  Source: https://doc.babylonjs.com/features/featuresDeepDive/gui/gui/
• CSS and backing-store pixels differ on high-DPI screens. Keep scene rendering
  capped at2x, but use browser-native DOM text for the displays, outside bloom.
  Source: https://developer.mozilla.org/en-US/docs/Web/API/Window/devicePixelRatio
• Provide zoom/inspect/reset buttons and keyboard access, not a gesture-only
  interface. Source: https://www.w3.org/WAI/WCAG21/Understanding/pointer-gestures.html

Chosen visual: graphite hexagonal instrument displays, neon cyan information,
honey hover/focus, original hub logo. No generated artwork or invented metrics.
Do not migrate this existing t27.ai/GitHub Pages site to another hosting system.

Acceptance before code

1. Merge public board, foundation and epics only within the same repository.
   Identity is repo+issue number; live board state wins over historical closure.
   Keep existing issue positions across polls. The hub remains reserved.
2. A cell's events join by issue number, never by module title or cell position.
   Epics show explicit child completion and recent own/child events, with the
   original event issue number visible. No events means an explicit empty state.
3. Semantic zoom: overview -> number/state -> title -> full issue/epic display.
   Native text stays sharp at close-up; render only a bounded visible subset.
   Full title/events remain accessible in the selected display without clipping.
4. A click/tap on a cell and a keyboard-operable Inspect control can frame it
   at useful reading size on phone and desktop. Wheel/buttons reach beyond the
   old8x cap. Fit restores the overview. Close-up survives ordinary polls.
5. Reduced-motion stops idle wall drift and camera interpolation. Dispose old
   event listeners and overlays on scene rebuild. Preserve other views,
   RU/EN, original logo and the previously corrected coverage provenance.

Plan

• Read current source and previous local corrections; remote main unchanged.
• Research primary docs and record this observable contract.
• Deterministic RED for identities/events/projection/zoom/epic counts.
• Implement issue display model, bounded DOM rendering and camera controls.
• GREEN/regressions, exact build, interaction review and independent review.
• Record outcome and production boundary; no fabricated deployment claim.

The earlier coverage patch is preserved. The root TRI runtime is still known
to have dataless iCloud files; no historical ledger rewrite is authorized.

[gHashTag]

🧪 [TEST] Native hive displays: RED recorded before implementation; model now passes stable issue placement, exact-number events, epic counts, canonical URLs and close-up beyond8x. Browser development preview frames real issue1540 at29.41x with a361x417 CSS-pixel native display, full title and latest event. Independent review found body/wheel interception, legacy event placement and camera persistence issues; fixes are now in place and being verified. Prior coverage correction preserved. No publication yet.

[gHashTag]

✅ [TEST] Issue-display implementation is complete locally and independently reviewed. Local commit: 86f03f4 on codex/queen-hive-displays-890 (not pushed or deployed). Native issue/epic cards replace 256px text textures; Inspect/Whole hive and manual zoom up to 128x work with RU/EN, keyboard and mobile tap/inner scrolling. Cards use exact issue events, explicit epic children and canonical repository links. No positional module coverage, foreign foundation or unmapped Bee fallback. Final checks: display contract PASS, coverage 34, honesty 113, languages 246/246, changed-file ESLint PASS, type ratchet unchanged at 179 existing errors, build PASS. BrowserOS verified final assets index-DUONrynA.js, Queen-qf8uYWJb.js and QueenCombBabylon-Cmt3JeB4.js on desktop and mobile DPR2. Epic #1334 displays 5/7 children independently of Closed; missing events remain empty. Evidence and resume instructions: docs/QUEEN_HIVE_DISPLAY_REPORT.md and append-only QUEEN_HIVE_DISPLAY_EVIDENCE.jsonl. Production is unchanged; release approval is still required. No new automation or root-ledger rewrite.

[gHashTag]

🔍 [RESEARCH] Release now explicitly authorized by the user. Reconciled previous QUEEN NEW history, merged PR #960, the untouched dirty trinity-game tree and current main 323b569. The other task cannot answer (provider 429); read-only review will assess its uncommitted camera-alpha change against the new native displays. Release candidate 86f03f4. Existing t27.ai target is ghashtag.github.io main/root Pages; use its concurrency-controlled publisher, preserve static content-loss checks and wait for actual live assets. No DNS, hosting, secrets, worker capacity or new automation changes. Contract: docs/QUEEN_HIVE_RELEASE_2026-09-07.md.

[gHashTag]

[FIX] PR #969 release validation found a real boundary defect: historical GitHub titles in the new issue chooser were treated as untranslated interface copy by the EN audit. Narrow fix preserves canonical source titles and exempts only their individual option/h3 nodes using the existing source-language convention; all controls/states remain audited. Contract recorded first, regression RED then GREEN, independent read-only review found no blockers; local built EN page has zero Cyrillic outside quoted source nodes. Full Website checks rerun is required before merge. Unrelated CI findings are documented rather than bypassed: Brain Health command reports TODO/not measured, docs consistency references absent tjepa files, Project update has HTTP 401, Claude review returned is_error:true. No secrets or unrelated workflows changed.

[gHashTag]

Release extension 0613922: the user requested the original point-down logo, homepage typography, an unclipped viewport, translucent displays and a real star catalog. Added a shared wall transform with projection/picking tests, Outfit/JetBrains tokens, a full-width outer grid track, and a reproducible 4,965-star HYG4.1/J2000 subset (CC BY-SA4.0; pinned source hashes and attribution included). Observed RED→GREEN defects: 832px viewport but only 756px implicit track; background-tab catalog loaded without painting. Exact desktop/mobile DPR2 artifact now fills its viewport and shows stars through native issue cards; mobile reduced-motion/tap passed. Independent read-only review reproduced all catalog rows and found no scoped blocker. Display/orientation/star, coverage34, honesty113, RU/EN246, API and type ratchets, ESLint and build pass (existing baselines remain). PR969 is being rechecked before the already-authorized existing apex publication. No DNS, secrets, scheduler, or other-agent worktree changes.

[gHashTag]

LIVE VERIFIED: PR969 merged normally as ab8c2c4; Website checks34117265633 passed. Existing apex publisher34117982409 passed and produced artifact96f5481089bf616ed7f963a13a56ef347debcfba; Pages34118254855 build/deploy passed with CNAME t27.ai unchanged. Public t27.ai/#/queen now serves index-D5f5alzV.js, Queen-DkVys0TV.js, QueenCombBabylon-BCtPNXkn.js, Queen-tRuWT4YQ.css and hyg-v41-bright-DtBO8j0_.json: HTTP200 and every SHA256 matches the verified local candidate. Browser verified actual translucent native cards over catalog stars, Outfit typography, point-down logo, full viewport width, exact #1540 Finished/refused event with Backlog state, and canonical GitHub links. Public RU mobile390x844/DPR2/reduced-motion has no overflow and preserves selection. Original logo rules and repeatable read-only release QA are saved in skills. The initial old page was browser cache, resolved by owned-tab hard reload without another deploy. This UI slice is complete; issue890 remains open for its broader campaign. Existing unrelated CI failures and real idle/review backlog remain separate, not hidden by animation.

[gHashTag]

🧪 [TEST] QUEEN hive status-filled cells — local verification

User requested full-cell colors visible from project overview. Implemented on
local branch codex/queen-hive-status-fill based on ab8c2c49:

• Initial/problem/unresolved goal = red, running/review = neon blue,
  done/closed with T27 proof = honey. Closing alone is not T27 evidence.
• One paint resolver drives scene palette batches, native cards and counts.
  Same-identity state/proof changes invalidate paint without losing selection.
• Hover/focus keeps honey outline/lift, not a misleading honey status fill.
• Original point-down logo, translucent space, real HYG catalog, RU/EN and
  canonical issue identity/events remain intact.

TDD: missing paint regression RED, then GREEN. Display/orientation/starfield,
coverage34, honesty113, language246 EN/RU and changed-file ESLint pass.
Type/API ratchets remain at existing179-errors/26files and110-gap baselines.
Build exit0 (1131modules,16.31s). Desktop1400x900 and mobile390x844, DPR2,
reduced-motion and touch overview verified with the exact local build.
Independent read-only reviewer found no blockers.

Observed public-source data in local preview:1206 cells;1014 red,192 review/work
blue,0 honey;two GPU batches. #1540 Backlog/refused is red; #1551 Review/wait is
blue. Coverage remains UNKNOWN because no issue-to-module proof is published.
This UI change does not solve worker runtime policy or claim running Bees.

☣️ [VERDICT] LOCAL_VERIFIED, not published. No push/merge/deploy, DNS, secrets or
runtime worker changes. New artifact root index-BZc_gtZI.js, Queen
Queen-CX43wZjz.js, scene QueenCombBabylon-BB0XpR19.js, CSS Queen-D-QnQ5SW.css.
Do not reuse PR969 deployment evidence as proof for this artifact.

[gHashTag]

Local hive signal language verified

Implemented separate lifecycle fills, fresh event rings, selection rims and
board/activity health. Red marks pending work/proof; stronger red marks blockers;
cyan is board-running only; violet is Queen review; slate is paused; honey is
complete plus explicit T27 proof. Result events are white, approved reviews green,
and finished/refused is red with a visible failure label. No public honey proof
or running worker was fabricated.

Fixed initial/reconnect replay, late older failure replay, recovery ring ordering,
historical review falsely counted as live work, and selection-generated event
flares. Independent read-only review has no remaining blocking findings.

Tests: displays PASS, honesty 113 PASS, coverage 34 PASS, EN/RU 246 keys each PASS,
focused ESLint PASS, build PASS (39.13 s). Type/API ratchets remain at their existing
179-error / 110-gap baselines. Exact local assets: index-DWmqLBld.js,
Queen-DwAgC4OG.js, QueenCombBabylon-DNxDYlmF.js, Queen-C9RsfDFr.css.

Browser verification: desktop/mobile DPR 2, RU/EN, reduced-motion, canonical
trios#1540 close-up, and page-local connection failure/recovery. Task fills were
retained during stale data and badges cleared after recovery. No server mutation.
Final sampled counts: 1206 issues; 1007 pending goal/proof, 192 review, 7 paused,
0 running, 0 honey. These are observed snapshots, not throughput.

Not published. Source branch codex/queen-hive-status-fill also contains the earlier
status-fill work. Next choices: exact-artifact release; public T27 proof adapter;
operator filters/minimap consistency and actual runtime refusal investigation.