Plan detailed 0.9.0 production-quality steps

Expand the next milestone into animation, audio, localization, accessibility,
performance/platform, QA/visual baselines, Godot-version policy and exit
criteria, with concrete verification work per workstream.

TODO.md

@@ -5,10 +5,10 @@ see `AGENTS.md` for the workflow and `MEMORY.md` for the decisions behind these
 
 ## Gap Analysis and v1.0.0 Roadmap
 
-Baseline: **v0.7.2** (2026-11-01). Three playable venues, a complete escalation/access/radio
+Baseline: **v0.8.1** (2026-11-01). Three playable venues, tutorial/campaign/checkpoint resume,
-loop, shift review, a Blender character/gear/prop library, 231 headless + 94 Python tests, a
+difficulty/help, a complete escalation/access/radio loop, shift review, the Blender asset library,
-13/13 clean screen-design judge and an optional Forward+ Desktop HQ build. This is a strong
+249 headless + 94 Python tests, a clean screen-design judge and an optional Forward+ Desktop HQ
-vertical slice, not yet a 1.0 product.
+build. This is a strong first-hour vertical slice, not yet a 1.0 product.
 
 ### Documentation audit (completed 2026-11-01)
 
@@ -52,22 +52,116 @@ vertical slice, not yet a 1.0 product.
 Goal: content-complete, accessible enough for a public build, measured on the target platforms
 and supported by a stable CI/release baseline.
 
-- [ ] Animation/presentation pass: negotiation/reaction poses and transitions, readable faces in
+### Animation and presentation
-      portrait, resolved camera height, and no coat/skirt leg clipping in normal movement.
+
-- [ ] Audio pass: music/ambience direction, event SFX, radio/crowd mix, caption hooks and a VO
+- [ ] Decide the first-person camera height against the concept frames; record the decision and
-      decision with production scope.
+      update the capture poses if the camera moves.
-- [ ] Localization pass: all player-facing strings externalized, pseudo-locale layout test, font
+- [ ] Add negotiation/reaction poses for warning, blocking, recognition, de-escalation and refusal,
-      fallback and a contributor translation workflow.
+      with short transitions instead of instant pose snaps.
-- [ ] Accessibility pass: subtitles/captions, colour-independent access/escalation cues, motion
+- [ ] Fix coat/skirt leg clipping in normal walking and blocking; adjust skin weights, hem
-      and flash options, full keyboard/gamepad parity and documented assist options.
+      geometry or a small corrective shape and verify from front/side/portrait angles.
-- [ ] Performance pass: profile 24 live NPCs, all venues, Web and Desktop HQ; record frame-time,
+- [ ] Run a portrait/close-up readability pass on all eight character variants, including hood,
-      draw-call and memory budgets; add regression checks where automation is possible.
+      helmet, cap and sunglasses; capture a repeatable portrait set.
-- [ ] QA pass: automated end-to-end shift for each venue, full smoke matrix, screenshot/judge
+- [ ] Add pose/animation regression checks where practical (bone/pose state plus the portrait
-      baselines, and a documented manual judge budget/model pin.
+      capture) and re-measure the skinned-avatar frame cost.
-- [ ] Resolve the Godot baseline: choose one supported version (or an explicit 4.3-CI/4.7-local
+- [ ] Exit: normal movement, blocking and portrait frames have no obvious clipping or pose
-      matrix), align export templates and visual smoke results, and document the policy.
+      pops; the camera decision is documented and visual baselines are refreshed.
+
+### Audio
+
+- [ ] Write the audio direction: crowd ambience, venue mood, UI/radio ticks, footsteps, warning
+      and escalation cues, and whether music is needed.
+- [ ] Map every gameplay event to an SFX/ambience hook, including tutorial prompts, campaign
+      transitions, radio send/failure, door lock and responder arrival.
+- [ ] Add Music, SFX and Ambience volume controls to Settings and persist them; keep the dummy
+      audio server safe for headless tests.
+- [ ] Add caption metadata to audio cues and tie it to the accessibility captions.
+- [ ] Make the VO decision: text-only for 0.9, or a scoped voice pipeline with content list,
+      recording/casting plan, licensing and fallback text.
+- [ ] Add audio smoke coverage for the new hooks and mix levels; document the final mix targets.
+
+### Localization
+
+- [ ] Audit hardcoded player-facing text, starting with Help, campaign/status screens, tutorial
+      overlay, review labels and new dialogue; move everything into `strings.en.json`.
+- [ ] Add a pseudo-locale or long-string expansion test that runs the UI at 1.5–2× text length and
+      flags clipped buttons, labels and HUD rows.
+- [ ] Verify font coverage for punctuation, arrows, accented characters and non-Latin sample
+      glyphs; document the fallback font policy.
+- [ ] Write the translator workflow: key naming, source-of-truth file, placeholders, review and
+      screenshot checks, and how to add a locale without code changes.
+- [ ] Add localization coverage tests for every named input action and every new 0.9 UI surface.
+- [ ] Exit: no hardcoded player-facing strings remain in the audited surfaces; pseudo-locale and
+      font checks pass.
+
+### Accessibility
+
+- [ ] Add subtitles/captions for tutorial prompts, radio/phone dialogue and key audio cues, with
+      a Settings toggle and readable defaults.
+- [ ] Audit all colour-dependent cues (access, escalation, crowd, door, flashlight) and add a
+      colour-independent shape/text/icon alternative where missing.
+- [ ] Add motion/flash options: disable camera bob/shake, reduce rapid flashes, and verify the
+      precision-walk/HUD transitions remain usable.
+- [ ] Verify complete keyboard-only and gamepad-only paths for menus, tutorial, inspection, wheel,
+      radio (including hold timing), review and Help.
+- [ ] Test HUD/font scaling at minimum and maximum settings, including long localized strings and
+      the body-camera overlay.
+- [ ] Write an accessibility statement and a manual accessibility checklist; add automated coverage
+      for input parity and caption/hint availability.
+- [ ] Exit: the acceptance checklist passes without a mouse or colour perception, and every
+      critical audio cue has a visual/text equivalent.
+
+### Performance and platform
+
+- [ ] Define the target hardware matrix (minimum/recommended desktop, Web browser and Desktop HQ
+      Vulkan) and record baseline frame-time, draw calls and memory per venue.
+- [ ] Add a repeatable 24-NPC performance smoke across all venues, including crowd panic and the
+      studio arrival; keep the 16-NPC result as the regression floor.
+- [ ] Profile Web and Desktop HQ separately; investigate LOD/impostor thresholds, prop draw calls,
+      shadow cost and skinning CPU before adding more avatars.
+- [ ] Add frame-time/memory regression gates where automation is possible and document manual
+      capture steps for the rest.
+- [ ] Verify the Compatibility/Web fallback and HQ Vulkan fallback on the target matrix; record
+      startup failures and GPU-driver requirements.
+- [ ] Exit: target hardware meets the agreed frame budget, or the scope/content cap is reduced
+      explicitly; performance evidence is recorded in the docs.
+
+### QA automation and visual baselines
+
+- [ ] Add an automated end-to-end shift for each venue that covers spawn, access, escalation,
+      radio, review and campaign advancement.
+- [ ] Consolidate the smoke matrix into one documented command/checklist; include tutorial,
+      checkpoint resume, campaign transition, Help, radio and all venue paths.
+- [ ] Refresh the screenshot gallery and realism baselines; run `compare` and the pinned judge,
+      recording the model, spend and any accepted review verdicts.
+- [ ] Keep the vision judge optional and cost-capped; document the manual judge budget/model pin
+      and never make an external API a CI release gate.
+- [ ] Add a release-candidate dry run that builds every artifact, verifies installers/Web/HQ and
+      records checksums.
+- [ ] Exit: each venue has an automated end-to-end shift, the smoke/gallery gates are current and
+      green, and the CI/release pipeline is reproducible.
+
+### Godot baseline and version policy
+
+- [ ] Choose the supported engine policy: one pinned version, or an explicit CI/local matrix
+      (currently 4.3 CI vs 4.7.2 local).
+- [ ] If upgrading, run import/headless/screenshot/HQ checks, remove temporary version-specific
+      light-falloff branches where possible, and align export templates, `.uid` policy and CI.
+- [ ] Document the upgrade/rollback procedure and the visual-smoke requirements for engine bumps.
+- [ ] Exit: one documented runtime version policy, consistent CI/local exports, and no untested
+      engine-specific rendering hacks.
+
+### Verification and exit criteria
+
+- [ ] Add/verify headless tests for every new 0.9 system, plus localization, accessibility and
+      performance checks that can run without a display.
+- [ ] Run the full smoke matrix and capture/label a clean 0.9 screenshot set.
+- [ ] Run a manual accessibility/localization review and record findings; fix P0/P1 blockers.
+- [ ] Update architecture, ADRs, status, README, changelog and roadmap in the same change as the
+      implementation.
 - [ ] Exit: production content is feature-complete, accessibility/localization acceptance criteria
-      pass, target hardware matrix is measured, and no P0/P1 defect remains open.
+      pass, target hardware is measured, the engine policy is documented, and no P0/P1 defect is
+      open.
 
 ## Milestone 0.9.5 — Release candidate