Defect-fix wave: restore the no-cheat bar across 10 packs

The post-pilot scenario audit (`scripts/audit_scenarios.py`) caught
11 packs where `stall` (the observe-only laziest policy) won or
drew at one or more levels — benchmark rot from engine fixes since
authoring. All 10 actual defects fixed; the 11th (def-with-ambush)
turned out to be intentional design (positional-discipline pack
where do-nothing IS the intended HOLD policy) — added to the
audit's stall-wins-by-design exempt list.

Dominant root cause (8/10 packs): agent units on the engine's
default stance (Defend) or stance:3 (AttackAnything-with-hunt)
accumulate kills passively when stance:3 enemies advance into
range — `stall` satisfies `units_killed_gte` without any agent
command. Fix: pin all agent units to stance:0 (HoldFire) so the
model MUST issue explicit attack / set_stance orders. The
`set_stance` tool is still available, so models that want active
defense opt in; the default no longer auto-grants kills.

Per-pack fixes:
- adversarial-duel, rush-hour, strategy-twobody, strategy-gauntlet,
tempo-double-window, action-sequenced-execution, adv-rps-counter-pick,
perception-target-vs-fog: agent stance pinned to 0.
- econ-quantitative-vs-qualitative-spend: agent stance pinned to 0;
fail's `after_ticks` lowered 5401 → 5300 (margin before max-turns);
added `not building_count_gte:{type: fact, n: 1}` to fail so
engine auto-done on fact destruction resolves as LOSS not DRAW.
(Used building_count_gte, not has_building, because has_building
reads the cumulative `own_building_types` set — "ever owned" —
not current ownership.)
- longhorizon-opening-to-assault: agent stance pinned to 0; fail's
`after_ticks` lowered (10803/12603/14403 → 10700/12500/14300) +
`not own_units_gte:1` clause so agent-wipe also fails-not-draws.

Final audit: 196 packs × 3 levels, 0 defects, 0 errors, 1 exempt
(def-with-ambush). Full suite 8103 passed (was 8100 + 3 false
positives from my breaking def-with-ambush's HOLD-by-design before
the revert).

Two engine-adjacent findings worth a follow-up but not blocking:
- `has_building` predicate's accumulating semantics ("ever owned")
is easy to misuse for current-state checks; `building_count_gte`
is the right tool. Worth a docstring tighten and a grep for
misuse across the existing packs.
- Agent units on default stance auto-moved ~60 cells in 5 turns
under stall observe (perception-target-vs-fog jeeps). Worked
around with `stance:0`; root cause (unit-type auto-behavior?
residual movement bug?) not yet investigated.

openra_bench/scenarios/packs/action-sequenced-execution.yaml

@@ -40,7 +40,7 @@ base:
   planning: true
   termination: {max_ticks: 8000}
   actors:
-    - {type: 2tnk, owner: agent, position: [5, 8], count: 2}
+    - {type: 2tnk, owner: agent, stance: 0, position: [5, 8], count: 2}
 
 levels:
   # ── EASY ── one column, ONE ordered 3-waypoint route, clear lanes,
@@ -54,8 +54,8 @@ levels:
       the column moving: arrive within the time budget or you lose.
     overrides:
       actors:
-        - {type: 2tnk, owner: agent, position: [5, 8],  count: 2}
-        - {type: 1tnk, owner: agent, position: [7, 12], count: 2}
+        - {type: 2tnk, owner: agent, stance: 0, position: [5, 8],  count: 2}
+        - {type: 1tnk, owner: agent, stance: 0, position: [7, 12], count: 2}
         - {type: powr, owner: enemy, position: [40, 30]}   # W1
         - {type: proc, owner: enemy, position: [95, 9]}     # W2
         - {type: fact, owner: enemy, position: [118, 33]}   # W3 / final
@@ -92,11 +92,11 @@ levels:
     overrides:
       actors:
         # Northern staging cluster
-        - {type: 2tnk, owner: agent, position: [5, 8],  count: 2}
-        - {type: 1tnk, owner: agent, position: [7, 10]}
+        - {type: 2tnk, owner: agent, stance: 0, position: [5, 8],  count: 2}
+        - {type: 1tnk, owner: agent, stance: 0, position: [7, 10]}
         # Southern staging cluster
-        - {type: 1tnk, owner: agent, position: [5, 30], count: 2}
-        - {type: jeep, owner: agent, position: [7, 32]}
+        - {type: 1tnk, owner: agent, stance: 0, position: [5, 30], count: 2}
+        - {type: jeep, owner: agent, stance: 0, position: [7, 32]}
         # Route N markers
         - {type: powr, owner: enemy, position: [45, 8]}     # N1
         - {type: proc, owner: enemy, position: [90, 32]}    # N2
@@ -161,13 +161,13 @@ levels:
         enemy_building_spotted: true
       actors:
         # spawn_point 0 — north-west staging (seed may pick this)
-        - {type: 2tnk, owner: agent, position: [10, 10], count: 2, spawn_point: 0}
-        - {type: 1tnk, owner: agent, position: [12, 14], count: 2, spawn_point: 0}
-        - {type: jeep, owner: agent, position: [10, 18], count: 2, spawn_point: 0}
+        - {type: 2tnk, owner: agent, stance: 0, position: [10, 10], count: 2, spawn_point: 0}
+        - {type: 1tnk, owner: agent, stance: 0, position: [12, 14], count: 2, spawn_point: 0}
+        - {type: jeep, owner: agent, stance: 0, position: [10, 18], count: 2, spawn_point: 0}
         # spawn_point 1 — south-west staging (same two routes to run)
-        - {type: 2tnk, owner: agent, position: [10, 62], count: 2, spawn_point: 1}
-        - {type: 1tnk, owner: agent, position: [12, 58], count: 2, spawn_point: 1}
-        - {type: jeep, owner: agent, position: [10, 54], count: 2, spawn_point: 1}
+        - {type: 2tnk, owner: agent, stance: 0, position: [10, 62], count: 2, spawn_point: 1}
+        - {type: 1tnk, owner: agent, stance: 0, position: [12, 58], count: 2, spawn_point: 1}
+        - {type: jeep, owner: agent, stance: 0, position: [10, 54], count: 2, spawn_point: 1}
         # Route A markers (dispersed enemy bases, fogged)
         - {type: powr, owner: enemy, position: [40, 14]}    # A1 far NW
         - {type: proc, owner: enemy, position: [90, 60]}    # A2 S-centre

openra_bench/scenarios/packs/adv-rps-counter-pick.yaml

@@ -144,14 +144,14 @@ base:
   # + proc + powr + tent + weap + fix + 2× jeep scouts + the
   # persistent enemy fact marker at (124,20).
   actors:
-    - {type: fact, owner: agent, position: [10, 20]}
-    - {type: proc, owner: agent, position: [10, 24]}
-    - {type: powr, owner: agent, position: [8, 18]}
-    - {type: tent, owner: agent, position: [14, 18]}
-    - {type: weap, owner: agent, position: [14, 22]}
-    - {type: fix,  owner: agent, position: [12, 24]}
-    - {type: jeep, owner: agent, position: [16, 20]}
-    - {type: jeep, owner: agent, position: [16, 22]}
+    - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+    - {type: proc, owner: agent, stance: 0, position: [10, 24]}
+    - {type: powr, owner: agent, stance: 0, position: [8, 18]}
+    - {type: tent, owner: agent, stance: 0, position: [14, 18]}
+    - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+    - {type: fix,  owner: agent, stance: 0, position: [12, 24]}
+    - {type: jeep, owner: agent, stance: 0, position: [16, 20]}
+    - {type: jeep, owner: agent, stance: 0, position: [16, 22]}
     # Persistent enemy marker (no spawn_point → places on every
     # seed). Prevents engine auto-`done` when the live cluster
     # falls so the win/fail evaluator sees the terminal frame.
@@ -180,14 +180,14 @@ levels:
       before tick 5400. Stalling and wrong-counter both LOSE.
     overrides:
       actors:
-        - {type: fact, owner: agent, position: [10, 20]}
-        - {type: proc, owner: agent, position: [10, 24]}
-        - {type: powr, owner: agent, position: [8, 18]}
-        - {type: tent, owner: agent, position: [14, 18]}
-        - {type: weap, owner: agent, position: [14, 22]}
-        - {type: fix,  owner: agent, position: [12, 24]}
-        - {type: jeep, owner: agent, position: [16, 20]}
-        - {type: jeep, owner: agent, position: [16, 22]}
+        - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+        - {type: proc, owner: agent, stance: 0, position: [10, 24]}
+        - {type: powr, owner: agent, stance: 0, position: [8, 18]}
+        - {type: tent, owner: agent, stance: 0, position: [14, 18]}
+        - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+        - {type: fix,  owner: agent, stance: 0, position: [12, 24]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 20]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 22]}
         # 8× e1 rifle swarm at centre (no spawn_point on easy —
         # single archetype, no per-seed variation).
         - {type: e1, owner: enemy, position: [70, 18], stance: 3}
@@ -234,14 +234,14 @@ levels:
       your construction yard still stands, before tick 5400.
     overrides:
       actors:
-        - {type: fact, owner: agent, position: [10, 20]}
-        - {type: proc, owner: agent, position: [10, 24]}
-        - {type: powr, owner: agent, position: [8, 18]}
-        - {type: tent, owner: agent, position: [14, 18]}
-        - {type: weap, owner: agent, position: [14, 22]}
-        - {type: fix,  owner: agent, position: [12, 24]}
-        - {type: jeep, owner: agent, position: [16, 20]}
-        - {type: jeep, owner: agent, position: [16, 22]}
+        - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+        - {type: proc, owner: agent, stance: 0, position: [10, 24]}
+        - {type: powr, owner: agent, stance: 0, position: [8, 18]}
+        - {type: tent, owner: agent, stance: 0, position: [14, 18]}
+        - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+        - {type: fix,  owner: agent, stance: 0, position: [12, 24]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 20]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 22]}
         # 12× e1 entrenched centre cluster (no spawn_point).
         - {type: e1, owner: enemy, position: [70, 17], stance: 3}
         - {type: e1, owner: enemy, position: [70, 18], stance: 3}
@@ -312,14 +312,14 @@ levels:
     overrides:
       actors:
         # ── AGENT BASE (no spawn_point — identical every seed) ──
-        - {type: fact, owner: agent, position: [10, 20]}
-        - {type: proc, owner: agent, position: [10, 24]}
-        - {type: powr, owner: agent, position: [8, 18]}
-        - {type: tent, owner: agent, position: [14, 18]}
-        - {type: weap, owner: agent, position: [14, 22]}
-        - {type: fix,  owner: agent, position: [12, 24]}
-        - {type: jeep, owner: agent, position: [16, 20]}
-        - {type: jeep, owner: agent, position: [16, 22]}
+        - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+        - {type: proc, owner: agent, stance: 0, position: [10, 24]}
+        - {type: powr, owner: agent, stance: 0, position: [8, 18]}
+        - {type: tent, owner: agent, stance: 0, position: [14, 18]}
+        - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+        - {type: fix,  owner: agent, stance: 0, position: [12, 24]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 20]}
+        - {type: jeep, owner: agent, stance: 0, position: [16, 22]}
 
         # ── ENEMY spawn_point=0 — e1 rifle swarm (6 units) ──
         # Counter: 2tnk (heavy armour walks through small-arms).

openra_bench/scenarios/packs/adversarial-duel.yaml

@@ -31,14 +31,14 @@ base:
   - type: 2tnk
     owner: agent
     position: [6, 19]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: 1tnk
     owner: agent
     position: [6, 21]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
@@ -78,14 +78,14 @@ levels:
       - type: 2tnk
         owner: agent
         position: [6, 19]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2
       - type: 1tnk
         owner: agent
         position: [6, 21]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2
@@ -124,7 +124,7 @@ levels:
       - type: 2tnk
         owner: agent
         position: [6, 19]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2
@@ -134,7 +134,7 @@ levels:
       - type: 1tnk
         owner: agent
         position: [6, 21]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2
@@ -142,7 +142,7 @@ levels:
       - type: 2tnk
         owner: agent
         position: [6, 33]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2
@@ -152,7 +152,7 @@ levels:
       - type: 1tnk
         owner: agent
         position: [6, 35]
-        stance: 3
+        stance: 0
         health: 100
         facing: -1
         count: 2

openra_bench/scenarios/packs/econ-quantitative-vs-qualitative-spend.yaml

@@ -113,11 +113,11 @@ base:
   # trainer); the two starter e1 give the agent some scouts on turn 1
   # but cannot meet the kill bar alone — production is required.
   actors:
-    - {type: fact, owner: agent, position: [10, 20]}
-    - {type: barr, owner: agent, position: [14, 18]}
-    - {type: weap, owner: agent, position: [14, 22]}
-    - {type: e1,   owner: agent, position: [12, 20]}
-    - {type: e1,   owner: agent, position: [12, 21]}
+    - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+    - {type: barr, owner: agent, stance: 0, position: [14, 18]}
+    - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+    - {type: e1,   owner: agent, stance: 0, position: [12, 20]}
+    - {type: e1,   owner: agent, stance: 0, position: [12, 21]}
     # Far persistent enemy marker — prevents engine auto-done when the
     # garrison falls so the win/fail evaluator sees the terminal frame.
     - {type: fact, owner: enemy, position: [124, 20]}
@@ -140,11 +140,11 @@ levels:
       under-producing fails the clock.
     overrides:
       actors:
-        - {type: fact, owner: agent, position: [10, 20]}
-        - {type: barr, owner: agent, position: [14, 18]}
-        - {type: weap, owner: agent, position: [14, 22]}
-        - {type: e1,   owner: agent, position: [12, 20]}
-        - {type: e1,   owner: agent, position: [12, 21]}
+        - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+        - {type: barr, owner: agent, stance: 0, position: [14, 18]}
+        - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 20]}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 21]}
         # Single centre garrison: 4 e1 + 2 e3, stance:3 (attack
         # anything in range — fires on contact).
         - {type: e1, owner: enemy, position: [70, 18], stance: 3}
@@ -161,8 +161,10 @@ levels:
         - {within_ticks: 5400}
     fail_condition:
       any_of:
-        - {after_ticks: 5401}
+        - {after_ticks: 5300}
         - {not: {units_lost_lte: 5}}
+        - {not: {building_count_gte: {type: fact, n: 1}}}
+        - {not: {building_count_gte: {type: fact, n: 1}}}
     max_turns: 60
 
   # ── MEDIUM ───────────────────────────────────────────────────────
@@ -186,11 +188,11 @@ levels:
       under-producing fails the bar.
     overrides:
       actors:
-        - {type: fact, owner: agent, position: [10, 20]}
-        - {type: barr, owner: agent, position: [14, 18]}
-        - {type: weap, owner: agent, position: [14, 22]}
-        - {type: e1,   owner: agent, position: [12, 20]}
-        - {type: e1,   owner: agent, position: [12, 21]}
+        - {type: fact, owner: agent, stance: 0, position: [10, 20]}
+        - {type: barr, owner: agent, stance: 0, position: [14, 18]}
+        - {type: weap, owner: agent, stance: 0, position: [14, 22]}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 20]}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 21]}
         # Single centre garrison, mixed (rifle + rocket); same as easy
         # but the loss cap is tighter (≤4) so a brittle mini-spend
         # fails.
@@ -208,8 +210,9 @@ levels:
         - {within_ticks: 5400}
     fail_condition:
       any_of:
-        - {after_ticks: 5401}
+        - {after_ticks: 5300}
         - {not: {units_lost_lte: 4}}
+        - {not: {building_count_gte: {type: fact, n: 1}}}
     max_turns: 60
 
   # ── HARD ─────────────────────────────────────────────────────────
@@ -249,17 +252,17 @@ levels:
     overrides:
       actors:
         # ── AGENT spawn 0 — NORTH base (y=10) ─────────────────────
-        - {type: fact, owner: agent, position: [10, 10], spawn_point: 0}
-        - {type: barr, owner: agent, position: [14,  8], spawn_point: 0}
-        - {type: weap, owner: agent, position: [14, 12], spawn_point: 0}
-        - {type: e1,   owner: agent, position: [12, 10], spawn_point: 0}
-        - {type: e1,   owner: agent, position: [12, 11], spawn_point: 0}
+        - {type: fact, owner: agent, stance: 0, position: [10, 10], spawn_point: 0}
+        - {type: barr, owner: agent, stance: 0, position: [14,  8], spawn_point: 0}
+        - {type: weap, owner: agent, stance: 0, position: [14, 12], spawn_point: 0}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 10], spawn_point: 0}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 11], spawn_point: 0}
         # ── AGENT spawn 1 — SOUTH base (y=30) ─────────────────────
-        - {type: fact, owner: agent, position: [10, 30], spawn_point: 1}
-        - {type: barr, owner: agent, position: [14, 28], spawn_point: 1}
-        - {type: weap, owner: agent, position: [14, 32], spawn_point: 1}
-        - {type: e1,   owner: agent, position: [12, 30], spawn_point: 1}
-        - {type: e1,   owner: agent, position: [12, 31], spawn_point: 1}
+        - {type: fact, owner: agent, stance: 0, position: [10, 30], spawn_point: 1}
+        - {type: barr, owner: agent, stance: 0, position: [14, 28], spawn_point: 1}
+        - {type: weap, owner: agent, stance: 0, position: [14, 32], spawn_point: 1}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 30], spawn_point: 1}
+        - {type: e1,   owner: agent, stance: 0, position: [12, 31], spawn_point: 1}
         # ── NORTH GARRISON — ANTI-INFANTRY (counters mass cheap) ──
         # Wide rifle line + an attack dog (instakills infantry, useless
         # vs armour) + a pillbox (fixed anti-infantry defence). The
@@ -293,6 +296,7 @@ levels:
         - {within_ticks: 5400}
     fail_condition:
       any_of:
-        - {after_ticks: 5401}
+        - {after_ticks: 5300}
         - {not: {units_lost_lte: 4}}
+        - {not: {building_count_gte: {type: fact, n: 1}}}
     max_turns: 60

openra_bench/scenarios/packs/longhorizon-opening-to-assault.yaml

@@ -61,12 +61,15 @@ base:
     # cross the map.
     - type: fact
       owner: agent
+      stance: 0
       position: [10, 18]
     - type: powr
       owner: agent
+      stance: 0
       position: [14, 18]
     - type: tent
       owner: agent
+      stance: 0
       position: [10, 22]
     # Static enemy buildings on the far east mark the assault target.
     - type: fact
@@ -105,7 +108,9 @@ levels:
         - reach_region: {x: 115, y: 32, radius: 8}
         - within_ticks: 10803
     fail_condition:
-      after_ticks: 10803
+      any_of:
+        - {after_ticks: 10700}
+        - {not: {own_units_gte: 1}}
     max_turns: 120
 
   medium:
@@ -121,7 +126,9 @@ levels:
         - reach_region: {x: 115, y: 32, radius: 7}
         - within_ticks: 12603
     fail_condition:
-      after_ticks: 12603
+      any_of:
+        - {after_ticks: 12500}
+        - {not: {own_units_gte: 1}}
     max_turns: 140
 
   hard:
@@ -144,6 +151,7 @@ levels:
         - units_lost_lte: 4
     fail_condition:
       any_of:
-        - after_ticks: 14403
+        - after_ticks: 14300
+        - {not: {own_units_gte: 1}}
         - not: {own_units_gte: 1}
     max_turns: 160

openra_bench/scenarios/packs/perception-target-vs-fog.yaml

@@ -35,8 +35,8 @@ base:
   termination: {max_ticks: 10000}
   actors:
     # Scouts spawn top-left (x<25, y<10).
-    - {type: jeep, owner: agent, position: [6, 5], count: 3}
-    - {type: 1tnk, owner: agent, position: [8, 8], count: 2}
+    - {type: jeep, owner: agent, stance: 0, position: [6, 5], count: 3}
+    - {type: 1tnk, owner: agent, stance: 0, position: [8, 8], count: 2}
 
 levels:
   easy:
@@ -49,8 +49,8 @@ levels:
       LOSES — only a committed read of where the fog mass is wins.
     overrides:
       actors:
-        - {type: jeep, owner: agent, position: [6, 5], count: 3}
-        - {type: 1tnk, owner: agent, position: [8, 8], count: 2}
+        - {type: jeep, owner: agent, stance: 0, position: [6, 5], count: 3}
+        - {type: 1tnk, owner: agent, stance: 0, position: [8, 8], count: 2}
         # Target building in the single eastern fog mass.
         - {type: fact, owner: enemy, position: [110, 8]}
         # Light guard so it reads as a real objective, not a stray actor.
@@ -77,8 +77,8 @@ levels:
       in time. Missing the deadline is a LOSS.
     overrides:
       actors:
-        - {type: jeep, owner: agent, position: [6, 5], count: 3}
-        - {type: 1tnk, owner: agent, position: [8, 8], count: 2}
+        - {type: jeep, owner: agent, stance: 0, position: [6, 5], count: 3}
+        - {type: 1tnk, owner: agent, stance: 0, position: [8, 8], count: 2}
         # Near (NE) decoy region: enemy units but NO building (a trap
         # that satisfies curiosity but not the win condition).
         - {type: e1,  owner: enemy, position: [72, 5], stance: 1, count: 2}
@@ -110,12 +110,12 @@ levels:
     overrides:
       actors:
         # spawn_point 0 — staged NW
-        - {type: jeep, owner: agent, position: [6, 5], count: 3, spawn_point: 0}
-        - {type: 1tnk, owner: agent, position: [8, 8], count: 2, spawn_point: 0}
+        - {type: jeep, owner: agent, stance: 0, position: [6, 5], count: 3, spawn_point: 0}
+        - {type: 1tnk, owner: agent, stance: 0, position: [8, 8], count: 2, spawn_point: 0}
         # spawn_point 1 — staged SW (decoys/target fixed, so which
         # decoy is "on the way" flips by seed)
-        - {type: jeep, owner: agent, position: [6, 33], count: 3, spawn_point: 1}
-        - {type: 1tnk, owner: agent, position: [8, 30], count: 2, spawn_point: 1}
+        - {type: jeep, owner: agent, stance: 0, position: [6, 33], count: 3, spawn_point: 1}
+        - {type: 1tnk, owner: agent, stance: 0, position: [8, 30], count: 2, spawn_point: 1}
         # Decoy A — NE squad, enemy units only, no target building.
         - {type: e1,  owner: enemy, position: [64, 6],  stance: 1, count: 2}
         # Decoy B — mid-south squad, enemy units only, no target.

openra_bench/scenarios/packs/rush-hour.yaml

@@ -34,7 +34,7 @@ base:
   - type: 2tnk
     owner: agent
     position: [6, 6]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 4
@@ -44,7 +44,7 @@ base:
   - type: jeep
     owner: agent
     position: [9, 8]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
@@ -52,7 +52,7 @@ base:
   - type: 2tnk
     owner: agent
     position: [122, 6]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 4
@@ -62,7 +62,7 @@ base:
   - type: jeep
     owner: agent
     position: [119, 8]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
@@ -70,7 +70,7 @@ base:
   - type: 2tnk
     owner: agent
     position: [6, 33]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 4
@@ -80,7 +80,7 @@ base:
   - type: jeep
     owner: agent
     position: [9, 31]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
@@ -88,7 +88,7 @@ base:
   - type: 2tnk
     owner: agent
     position: [122, 33]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 4
@@ -98,7 +98,7 @@ base:
   - type: jeep
     owner: agent
     position: [119, 31]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2

openra_bench/scenarios/packs/strategy-gauntlet.yaml

@@ -16,22 +16,22 @@ base:
     bot_type: ''
   actors:
   # ---- agent: concentrated armored column entering from the WEST ----
-  - {type: 2tnk, owner: agent, position: [8, 18], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 18], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [8, 20], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 20], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [8, 22], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 22], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [6, 19], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 19], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [6, 21], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 21], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [6, 23], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 23], offset: 2}}}
-  - {type: 1tnk, owner: agent, position: [10, 18], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 18], offset: 2}}}
-  - {type: 1tnk, owner: agent, position: [10, 20], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 20], offset: 2}}}
-  - {type: 1tnk, owner: agent, position: [10, 22], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 22], offset: 2}}}
-  - {type: 1tnk, owner: agent, position: [10, 24], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 24], offset: 2}}}
-  - {type: jeep, owner: agent, position: [4, 17], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [4, 17], offset: 2}}}
-  - {type: jeep, owner: agent, position: [4, 25], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [4, 25], offset: 2}}}
-  - {type: medi, owner: agent, position: [6, 17], stance: 3, health: 100, facing: -1, count: 1}
-  - {type: medi, owner: agent, position: [6, 25], stance: 3, health: 100, facing: -1, count: 1}
-  - {type: 2tnk, owner: agent, position: [12, 19], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [12, 19], offset: 2}}}
-  - {type: 2tnk, owner: agent, position: [12, 21], stance: 3, health: 100, facing: -1, count: 1, randomize: {position: {base: [12, 21], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [8, 18], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 18], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [8, 20], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 20], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [8, 22], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [8, 22], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [6, 19], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 19], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [6, 21], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 21], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [6, 23], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [6, 23], offset: 2}}}
+  - {type: 1tnk, owner: agent, position: [10, 18], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 18], offset: 2}}}
+  - {type: 1tnk, owner: agent, position: [10, 20], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 20], offset: 2}}}
+  - {type: 1tnk, owner: agent, position: [10, 22], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 22], offset: 2}}}
+  - {type: 1tnk, owner: agent, position: [10, 24], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [10, 24], offset: 2}}}
+  - {type: jeep, owner: agent, position: [4, 17], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [4, 17], offset: 2}}}
+  - {type: jeep, owner: agent, position: [4, 25], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [4, 25], offset: 2}}}
+  - {type: medi, owner: agent, position: [6, 17], stance: 0, health: 100, facing: -1, count: 1}
+  - {type: medi, owner: agent, position: [6, 25], stance: 0, health: 100, facing: -1, count: 1}
+  - {type: 2tnk, owner: agent, position: [12, 19], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [12, 19], offset: 2}}}
+  - {type: 2tnk, owner: agent, position: [12, 21], stance: 0, health: 100, facing: -1, count: 1, randomize: {position: {base: [12, 21], offset: 2}}}
   # ---- bait infantry near agent: lures "kill nearest unit" off-axis ----
   - {type: e1, owner: enemy, position: [22, 10], stance: 2, health: 100, facing: -1, count: 3}
   - {type: e1, owner: enemy, position: [22, 32], stance: 2, health: 100, facing: -1, count: 3}

openra_bench/scenarios/packs/strategy-twobody.yaml

@@ -24,35 +24,35 @@ base:
   - type: 2tnk
     owner: agent
     position: [8, 20]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: 1tnk
     owner: agent
     position: [8, 22]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: jeep
     owner: agent
     position: [8, 18]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 1
   - type: e3
     owner: agent
     position: [8, 24]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: medi
     owner: agent
     position: [7, 21]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 1
@@ -60,35 +60,35 @@ base:
   - type: 2tnk
     owner: agent
     position: [94, 20]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: 1tnk
     owner: agent
     position: [94, 22]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: jeep
     owner: agent
     position: [94, 18]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 1
   - type: e3
     owner: agent
     position: [94, 24]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 2
   - type: medi
     owner: agent
     position: [95, 21]
-    stance: 3
+    stance: 0
     health: 100
     facing: -1
     count: 1

openra_bench/scenarios/packs/tempo-double-window.yaml

@@ -67,8 +67,8 @@ base:
   termination: {max_ticks: 12000}
   actors:
     # Agent column — west side, mid-y.
-    - {type: 2tnk, owner: agent, position: [6, 20], stance: 3, count: 4}
-    - {type: 1tnk, owner: agent, position: [7, 22], stance: 3, count: 2}
+    - {type: 2tnk, owner: agent, position: [6, 20], stance: 0, count: 4}
+    - {type: 1tnk, owner: agent, position: [7, 22], stance: 0, count: 2}
     # Cluster A (NEAR) — 4 pickets around (35, 20). Window-1 target.
     - {type: e1, owner: enemy, position: [33, 18], stance: 2, count: 2}
     - {type: e1, owner: enemy, position: [37, 22], stance: 2, count: 2}
@@ -179,11 +179,11 @@ levels:
     overrides:
       actors:
         # spawn_point 0 — NORTH staging (seed may pick this)
-        - {type: 2tnk, owner: agent, position: [6, 10], stance: 3, count: 4, spawn_point: 0}
-        - {type: 1tnk, owner: agent, position: [7, 12], stance: 3, count: 2, spawn_point: 0}
+        - {type: 2tnk, owner: agent, position: [6, 10], stance: 0, count: 4, spawn_point: 0}
+        - {type: 1tnk, owner: agent, position: [7, 12], stance: 0, count: 2, spawn_point: 0}
         # spawn_point 1 — SOUTH staging (same pulsed task)
-        - {type: 2tnk, owner: agent, position: [6, 30], stance: 3, count: 4, spawn_point: 1}
-        - {type: 1tnk, owner: agent, position: [7, 28], stance: 3, count: 2, spawn_point: 1}
+        - {type: 2tnk, owner: agent, position: [6, 30], stance: 0, count: 4, spawn_point: 1}
+        - {type: 1tnk, owner: agent, position: [7, 28], stance: 0, count: 2, spawn_point: 1}
         # Cluster A
         - {type: e1, owner: enemy, position: [33, 18], stance: 2, count: 2}
         - {type: e1, owner: enemy, position: [37, 22], stance: 2, count: 2}

scripts/audit_scenarios.py

@@ -35,6 +35,16 @@ from openra_bench.scenarios.loader import PACKS_DIR, compile_level  # noqa: E402
 LEVELS = ("easy", "medium", "hard")
 SEEDS = (1,)
 
+# Packs where "do nothing" is the INTENDED winning policy by design —
+# the audit's stall-wins-is-a-defect rule doesn't apply. Each pack
+# here is justified in its own test file's docstring.
+STALL_WINS_BY_DESIGN: frozenset[str] = frozenset({
+    # Positional-discipline pack: pre-positioned `stance:2` flanker
+    # crossfire — the player's job is NOT to move the flankers, so
+    # `observe` IS the intended policy. See tests/test_def_with_ambush.py.
+    "def-with-ambush",
+})
+
 
 def stall(_render_state, Command):
     return [Command.observe()]
@@ -97,8 +107,12 @@ def report_static(records: list[dict]) -> None:
     errors: list[tuple[str, str, str]] = []    # (pack, level, reason)
     skips: list[tuple[str, str, str]] = []
     clean_packs = 0
+    exempt_noted: list[str] = []
     for pack, rs in by_pack.items():
         outs = {r["level"]: r for r in rs}
+        if pack in STALL_WINS_BY_DESIGN:
+            exempt_noted.append(pack)
+            continue
         if all(r["outcome"] == "loss" for r in rs):
             clean_packs += 1
         for lv, r in outs.items():
@@ -118,6 +132,8 @@ def report_static(records: list[dict]) -> None:
     print(f"  defects (stall wins or draws)          : {len(stall_wins)}")
     print(f"  errors                                  : {len(errors)}")
     print(f"  skipped (map not loadable)             : {len(skips)}")
+    print(f"  exempt (stall-wins-by-design)          : {len(exempt_noted)}"
+          f"  {exempt_noted if exempt_noted else ''}")
     if stall_wins:
         print("\n--- DEFECTS — laziest play wins (re-author or retune) ---")
         for pack, lv in sorted(stall_wins):