Polish cooperative learning and interior workface lighting

[NellInc]

Summary

• Give cooperative role answers immediate, authoritative inline clarification without treating correct answers as member consent.
• Extend replay comparison to include bounded agreement time and a discussion grounded in actual charters, ordered plans, commitments and evidence quality. No winner or inferred charter bonus.
• Rebalance interior illumination with bounded ceiling banks, real fitting/pipe clearance and a narrow same-page reference. Keep two/four lamps, fifteen fittings/pools, the 44 m cutoff and existing shadow budget.

Validation

• Cooperative feedback: 45 focused tests and 42 native views.
• Actual replay learning: 43 focused tests and 24 native views, using genuinely played and downloaded records.
• Lighting: 105 affected tests, 78 native views, including Medium day/dusk/night and High/Low night. Exact source/build seals retained locally.
• Required typecheck, ESLint, format, depth, shader, fresh bundle/reachability/community checks pass. 431 compiled files, 169.88 MiB below the unchanged 170 MiB cap.
• Muted sandboxed out-of-process native Metal. No actor, inventory, quality or exposure override; paused plant/member state and light identities restore exactly.
• Outgoing history secret scan and 100 MB Git size check pass.

Scope and limits

Interface and gameplay authority, consent, replay records, physical simulation, people routes and exterior layout remain preserved. Lighting is an authored bounded bank approximation; the reference swaps illumination only. Existing posts partly occlude two night close views. Pacing/adaptive-quality work remains open separately, with no waived budgets. Physical player/device checks remain deferred. No unrelated primary-checkout work or PR17 is included.

Ship through all five original PR checks, normal merge, main CI and Pages. Verify served code and assets against the accepted local assembly after deployment.