providers: implement BedrockProvider via Converse API

conftest: skip image-primary tests that call _render_state helper

Add Playlist tab — cold-start non-gamer baseline UX

Bucket E: align synthetic predicate-test contexts with current YAML

Bucket D: discover archived packs + load adversarial-siege from _archive

Bucket C: classify the 11 unclassified active packs

Bucket A: drop tests referencing archived artofwar-decoy-sacrifice

conftest: add run_handoff / open_study_session to engine-token list

conftest: skip tests that import openra_train inside test bodies

Merge origin/main into pr13-revised

Land PR #19 static site infrastructure (rebased onto pr13-revised)

def-position-revealed-direction: tailored 112x40 lane arena

def-tower-line-vs-cluster: per-tier maps that match topology to threat geometry

combat-hold-chokepoint: rewrite as canonical chokepoint-arena generator use case

def-walls-vs-towers: tier-differentiated walls vs towers doctrine

no-cheat redesign: build-defensive-tower-line — wide-front LINE topology

def-multi-direction: tailor per-tier arenas + scale lanes per tier

def-position-expected-direction: tailor per-tier maps to the funnel idiom

combat-attack-from-behind-fog: per-tier maps reinforce fog-flank doctrine

feat(scenario): econ-contested-expansion — corridor map tailored to geographic contention

econ-second-base-race: redesign with custom map + spatial placement gate

mapgen: obstacles, bridges-arena, chokepoint-arena + per-scenario authoring

Fix P0 regression: explicit agent.cash in 3 test packs

Phase 1 engine audit: ENGINE_AUDIT.md + bench-side closures

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

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

Strip empty tool_calls from wire history (Together Qwen3.6-Plus fix)

Paper-experiment prep: human-study harness + pass^k + paper plan

Add handoff ablation (recover-from-deficit / capitalize-on-advantage)

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

Skip live engine tests without openra_train

Skip survive live tests without engine wheel

Skip live Rust rendezvous tests without engine wheel

Improve Play minimap clarity

Persist human Play-tab runs in the standard Playback format

test: HumanController drives a 1v1 side (Phase 2 x Phase 3)

Phase 2: Play tab — selection boundary, move arrows, clear selection

Recalibrate 8 packs regressed by the count: spawn-spread engine fix

Phase 2: Play tab — fix hp, reorder panel, cleaner units table

Phase 2: Play tab — show objective, units panel, sharper minimap

Phase 2: 'Play' tab in app.py — human-labeling UI

Phase 2: InteractiveSession — turn-steppable backend for GUI play

Phase 3: 1v1 full-macro adversarial harness

Phase 2: human-labeling machine core

Phase 1: unified Controller interface for the eval stack

fix(scenario): combat-target-priority-highvalue — recalibrate after engine movement fixes

fix(scenario): rob-multiple-simultaneous-pressures — recalibrate after engine movement fixes

fix(scenario): combat-tank-vs-tank-engagement — recalibrate after engine movement fixes

fix(scenario): combat-bait-counter-attack — recalibrate after engine movement fixes

fix(scenario): build-repair-priority-under-fire — add hard-tier spawn-witness