Defensive fix: strip agent.cash=0 when starting_cash>0 in tmp YAML

Quality drive: schema fix, 5 new/revised packs, 4 engine tests, scenario audit

Engine-feature integration: 4 commands + 9 scenario packs + 4 test suites

Add perception ablation grid (observation channel × fog of war)

Phase 1: unified Controller interface for the eval stack

feat(scenario): scout-cycle-keep-info-fresh — re-scout to detect mid-episode reinforcement

feat(bench): forbidden_tools + tool_violations_gte — strict-toolban / procedural-compliance primitive (BFCL V4 / τ²-bench / IFBench anchor)

Default spawn_mcvs:false — stop engine auto-seeding phantom MCVs

playback: append a terminal 'episode end' frame (the resolved win/loss board)

Structured-fog text mode, premium routing, codex descriptions, minimap colour-by-difficulty

playback: save the SAME _minimap_v2 the model receives (fog-accumulating) — viewer now shows exactly what the model saw, not the legacy matplotlib render

Training-parity minimap (real terrain + legend) + viewer (system/thinking/debrief)

Unified Battle Viewer in app.py + run/model playback identity

Wire goal tracker into scoring + leaderboard

Playback: capture model reasoning + per-turn goal tracker + viewer

S7 bench: surrender tool + loss outcome (tool schema 1:1, 15==15)

Step 4 (bench): interrupt-driven loop + playback/trace capture

Pipeline step 7: per-episode playback persistence

#12: real custom-map terrain via dynamic .oramap registry

Bench: economy scenario pack + full-loop integ test + starting_cash constraint

Add scoring + P/R/A diagnostics + run_eval CLI

Add Rust-backed eval stack: scenario packs, adapter, spine, integration tests