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

The engine count: N fix (spawn units on distinct cells instead of
stacked) shifted starting layouts and regressed the no-cheat bar of 8
packs — 35 test cases. Recalibrated each so stall/brute/wrong-path
LOSE and the intended policy WINS on every level + hard seed, no draws:

- combat-bait-counter-attack: spread strike column ate more anti-tank
fire head-on; staged it on the north flank (medium/hard).
- econ-buy-vs-build-decision: dispersed rusher band takes ~2 more
turns to clear; extended the hard deadline.
- harass-response-preserve: spread probes split focus-fire; moved the
hard probe anchor into fog so the defence regroups concentrated.
- mfb-rotating-production-pressure: softened EAST raid let a late tank
survive; easy raid count 3 -> 4.
- mid-concede-vs-hold: spread flex squad let split-defend hold WEST;
hard WEST push count 20 -> 26.
- risk-blockade-bypass: diluted corridor garrison let do-nothing pass;
strengthened the easy/medium garrison.
- scout-count-defenders: spread defenders focus-fire harder; defender
health 60, medium 1 pbox, hard deadline +300t.
- tp-survive-n-turns: reinforcement waves land closer; medium wave
count 6 -> 5 per row.

343 tests across the 8 packs green. Pairs with OpenRA-Rust feb8981.

openra_bench/scenarios/packs/combat-bait-counter-attack.yaml

@@ -159,32 +159,43 @@ levels:
 
   # ── MEDIUM ───────────────────────────────────────────────────────
   # +1 controlled variable: 5 guards forming a south + west L-cover
-  # of the yard (the two natural west-staging approaches: head-on
-  # along y=20 OR a south sweep along y=24..28). A single deep-south
-  # bait pulls the south arc off post; the strike commits while that
-  # arc is displaced. After the engine move-fire fixes the e3 Dragon
-  # bites the 2tnk Heavy armour harder, so a 3-tank strike can no
-  # longer trade a 5-guard cover within the cap — the strike force is
-  # 5 medium tanks (was 3), a scale-up matching the denser cover.
-  # Strict attrition cap stays 2 (the bait jeep + at most one tank):
-  # the strike must still hit the yard essentially intact; burning a
-  # second tank fails.
+  # of the yard. A single deep-south bait pulls the south arc off
+  # post; the strike commits on the NORTH flank while that arc is
+  # displaced, swinging in above the cover to reach the yard. After
+  # the engine move-fire fixes the e3 Dragon bites the 2tnk Heavy
+  # armour harder, so a 3-tank strike can no longer trade a 5-guard
+  # cover within the cap — the strike force is 5 medium tanks (was
+  # 3), a scale-up matching the denser cover. The strike stages on
+  # the NORTH flank (y=12, was a head-on y=20): the engine `count:`
+  # spawn-spread now scatters the column across several cells, so a
+  # head-on stage drives the spread column straight into the
+  # un-baited WEST arc and trades too much armour; the north-flank
+  # stage lets the column swing in above the cover once the bait has
+  # pulled the south arc. Strict attrition cap stays 2 (the bait
+  # jeep + at most one tank): the strike must still hit the yard
+  # essentially intact; burning a second tank fails.
   medium:
     description: >
       A denser cluster of 5 leashed rocket-infantry guards covers
       the enemy construction yard (`fact` at 80,20) on the WEST and
       SOUTH faces (the two natural approaches from your west base).
       A head-on push without first thinning the cover is decisively
-      fatal. You command 5 medium tanks plus the bait jeep but may
-      spend at most 2 units — losing a second tank fails. Commit the
-      fast jeep on a deep south flank vector to pull the south arc of
-      guards off post after it; while that arc is displaced, drive
-      the strike tanks in, focus-fire the remaining guards, and
-      destroy the yard before tick 5400. Bait-only (no strike) loses
-      on the clock; brute frontal trades the strike force.
+      fatal. You command 5 medium tanks staged on the NORTH flank
+      plus the bait jeep, but may spend at most 2 units — losing a
+      second tank fails. Commit the fast jeep on a deep south flank
+      vector to pull the south arc of guards off post after it; while
+      that arc is displaced, swing the strike tanks around the
+      now-vacated NORTH flank and onto the yard before tick 5400.
+      Bait-only (no strike) loses on the clock; brute frontal trades
+      the strike force.
     overrides:
       actors:
-        - {type: 2tnk, owner: agent, position: [6, 20], stance: 1, count: 5}
+        # Strike force: 5 medium tanks staged on the NORTH flank
+        # (y=12) so that, once the bait pulls the SOUTH arc off post,
+        # the column swings in above the WEST/SOUTH cover and reaches
+        # the yard taking minimal anti-tank fire. A head-on y=20
+        # stage drives straight into the un-baited WEST arc.
+        - {type: 2tnk, owner: agent, position: [6, 12], stance: 1, count: 5}
         - {type: jeep, owner: agent, position: [6, 28], stance: 1}
         - {type: fact, owner: enemy, position: [80, 20]}
         # 5 guards on the WEST + SOUTH faces. A deep-south jeep bait
@@ -213,15 +224,15 @@ levels:
   # +1 more controlled variable on top of medium: TWO spawn_point
   # groups round-robin the agent's staging DISTANCE from the yard —
   # spawn 0 stages at the far west (x=6), spawn 1 stages well forward
-  # (x=30). The bait/strike geometry is identical (same 5-guard L
-  # cover, same fact latitude y=20), but the strike's run-in distance
-  # and the timing of the bait pull vary per seed, so a memorised
-  # fixed-tick opening cannot generalise — the agent must read its
-  # actual staging position. After the engine move-fire fixes a
-  # north/south staging split collapses to a knife-edge (the closer
-  # guard arc engages the column before the bait can pull it), so the
-  # spawn axis is the staging distance, not latitude. Same 5-guard L
-  # cover, 5-tank strike, and loss cap 2 as medium.
+  # (x=18). The bait/strike geometry is identical (same 5-guard L
+  # cover, same north-flank y=12 stage, same fact at 80,20), but the
+  # strike's run-in distance and the timing of the bait pull vary per
+  # seed, so a memorised fixed-tick opening cannot generalise — the
+  # agent must read its actual staging position. The spawn axis is
+  # the staging distance, not latitude (a north/south staging split
+  # collapses to a knife-edge under the engine move-fire fixes — the
+  # closer guard arc engages the column before the bait can pull it).
+  # Same 5-guard L cover, 5-tank strike, and loss cap 2 as medium.
   hard:
     description: >
       The same dense 5-guard L-cover of the enemy construction yard
@@ -229,29 +240,32 @@ levels:
       from a seed-chosen distance — either the far west or well
       forward — so a memorised fixed opening cannot generalise; read
       your tanks' actual position from the observation. You command 5
-      medium tanks plus the bait jeep but may spend at most 2 units
-      (the jeep + at most one tank) — losing a second tank fails.
-      Commit the fast jeep on a deep south flank vector to pull the
-      south arc of guards off post; while that arc is displaced,
-      drive the strike tanks in, focus-fire the remaining guards, and
-      destroy the yard before tick 5400. Brute frontal trades armour
-      and fails; bait-only never razes the yard; stalling loses the
-      clock.
+      medium tanks staged on the NORTH flank plus the bait jeep, but
+      may spend at most 2 units (the jeep + at most one tank) —
+      losing a second tank fails. Commit the fast jeep on a deep
+      south flank vector to pull the south arc of guards off post;
+      while that arc is displaced, swing the strike tanks around the
+      now-vacated NORTH flank and onto the yard before tick 5400.
+      Brute frontal trades armour and fails; bait-only never razes
+      the yard; stalling loses the clock.
     overrides:
       actors:
         # spawn_point 0 — FAR-WEST staging (x=6): the strike has the
         # full run-in to the yard; the bait has time to pull the
-        # south arc well off post before the column arrives.
-        - {type: 2tnk, owner: agent, position: [6, 20], stance: 1, count: 5,
+        # south arc well off post before the column arrives. Staged
+        # on the NORTH flank (y=12) so the column swings in above
+        # the cover, as in medium.
+        - {type: 2tnk, owner: agent, position: [6, 12], stance: 1, count: 5,
            spawn_point: 0}
         - {type: jeep, owner: agent, position: [6, 28], stance: 1,
            spawn_point: 0}
-        # spawn_point 1 — FORWARD staging (x=30): the strike is much
+        # spawn_point 1 — FORWARD staging (x=18): the strike is much
         # closer to the yard, so the bait pull and the strike commit
-        # are tighter together — the same idiom on a compressed clock.
-        - {type: 2tnk, owner: agent, position: [30, 20], stance: 1, count: 5,
+        # are tighter together — the same north-flank idiom on a
+        # compressed clock.
+        - {type: 2tnk, owner: agent, position: [18, 12], stance: 1, count: 5,
            spawn_point: 1}
-        - {type: jeep, owner: agent, position: [30, 28], stance: 1,
+        - {type: jeep, owner: agent, position: [18, 28], stance: 1,
            spawn_point: 1}
         # The OBJECTIVE.
         - {type: fact, owner: enemy, position: [80, 20]}

openra_bench/scenarios/packs/econ-buy-vs-build-decision.yaml

@@ -286,6 +286,19 @@ levels:
   # slightly heavier than medium's 5× e1, so hard stays harder),
   # which the buy-now force clears inside the clock while
   # stall / build-weap-first still LOSE.
+  #
+  # Engine recalibration note (count: spawn-spread fix): the
+  # `count: N` actor expansion was fixed to SPREAD copies across
+  # N distinct cells (outward rings from the anchor) instead of
+  # STACKING them on one cell. The hard rusher band (5× e1 + 1× e3
+  # tagged with spawn_point) now disperses ~1-2 cells around its
+  # anchor, so the buy-now tanks need a couple more turns to chew
+  # through the whole band — the kill bar (5) is met around tick
+  # ~1560, slightly past the old within_ticks 1499. The deadline
+  # is extended to 1649/1650 (still far under the max_turns 25
+  # ceiling 2253, still a real reachable anti-stall timeout) so
+  # the intended buy-now WIN registers while stall / build-weap
+  # still LOSE by fact-razed well before the clock.
   hard:
     description: >
       A rusher band (5 rifle infantry + 1 rocket soldier) is
@@ -297,7 +310,7 @@ levels:
       war factory) OR a second war factory (build weap, residual
       $1400 = 1 tank). The rush is incoming NOW. Win when ≥5 enemy
       units are killed AND your fact is still standing AND the
-      clock has not run out (tick ≤ 1499). Stall,
+      clock has not run out (tick ≤ 1649). Stall,
       build-second-factory-first, or under-spend all fail.
     overrides:
       actors:
@@ -343,9 +356,9 @@ levels:
       all_of:
         - {units_killed_gte: 5}
         - {building_count_gte: {type: fact, n: 1}}
-        - {within_ticks: 1499}
+        - {within_ticks: 1649}
     fail_condition:
       any_of:
-        - {after_ticks: 1500}
+        - {after_ticks: 1650}
         - {not: {building_count_gte: {type: fact, n: 1}}}
     max_turns: 25

openra_bench/scenarios/packs/harass-response-preserve.yaml

@@ -176,24 +176,30 @@ levels:
   #   2. Strict attrition: lose ZERO. Both `proc` AND `tent` must
   #      still stand.
   # The intended fire-support-on-home-turf policy wins (HoldFire
-  # defenders + explicit attack_unit on each probe in turn — they
-  # arrive serially as the patrol script bounces them in). Stall:
-  # the triple chip-volley razes the assets unopposed. Chase: the
-  # bouncing probes outpace pursuit (PATROL_RADIUS=8 in two beats),
-  # defenders overextend into [22..30,20] while the next swing
-  # arrives — assets die undefended. Yield: same — assets fall.
+  # defenders + explicit attack_unit on each probe as it swings into
+  # the asset footprint — the defenders engage CONCENTRATED from a
+  # held position rather than chasing). The probe anchor [26,20] sits
+  # east of the defenders' sight horizon: the probes spawn in fog,
+  # bounce out east then swing back west to ~x=18 (26 − PATROL_RADIUS
+  # 8) — into rifle range of the assets — where the held squad cuts
+  # them down one swing at a time, taking ZERO losses. Stall: the
+  # probes chip the assets every west swing, unopposed, and no kills
+  # are scored ⇒ timeout LOSS. Chase: marching east past the assets
+  # into the deep-fog spawn band overextends the squad — the bouncing
+  # probes (PATROL_RADIUS=8 in two beats) pick off pursuers and the
+  # assets stand undefended ⇒ LOSS. Yield: assets abandoned ⇒ LOSS.
   hard:
     description: >
       THREE enemy riflemen probe your perimeter on the central
-      latitude (anchor 22,20 → bouncing 14..30,20). Your refinery
+      latitude (anchor 26,20 → bouncing 18..34,20). Your refinery
       (`proc` at 16,20) AND your adjacent barracks (`tent` at 18,20)
       must BOTH still stand. Your 5 riflemen stage on a seed-varied
       latitude (north y=18 or south y=22) — both slots cover BOTH
-      assets. They are on HOLD-FIRE; order them to engage each probe
-      as it swings west into your assets' footprint. Win when ≥3
-      probes are dead, both buildings still stand, ≥5 of you are
-      alive and you have NOT lost a single defender — before tick
-      4500. Idling, chasing east, or pulling back all LOSE.
+      assets. They are on HOLD-FIRE; hold the asset footprint and
+      order them to engage each probe as it swings west into range.
+      Win when ≥3 probes are dead, both buildings still stand, ≥5 of
+      you are alive and you have NOT lost a single defender — before
+      tick 4500. Idling, chasing east, or pulling back all LOSE.
     overrides:
       actors:
         # Both assets, duplicated across BOTH spawn_point groups —
@@ -219,11 +225,16 @@ levels:
         - {type: e1, owner: agent, position: [14, 23], stance: 0, spawn_point: 1}
         - {type: e1, owner: agent, position: [13, 21], stance: 0, spawn_point: 1}
         - {type: e1, owner: agent, position: [13, 23], stance: 0, spawn_point: 1}
-        # Three probes stacked at the central anchor — they bounce
-        # together but engage serially as defenders kill them one
-        # by one. The triple chip-volley is the controlled-variable
-        # vs medium's two; the stall/chase/yield losses scale with it.
-        - {type: e1, owner: enemy, position: [22, 20], stance: 2, count: 3}
+        # Three probes at the central anchor — `count: 3` spreads them
+        # across three adjacent cells (engine spawn-spread); they
+        # bounce on the patrol script and engage serially as defenders
+        # kill them one by one. The anchor sits east of the defenders'
+        # sight horizon so the probes start in fog — the held squad
+        # engages them only as they swing west into the asset
+        # footprint, not by chasing. The triple chip-volley is the
+        # controlled variable vs medium's two; stall/chase/yield
+        # losses scale with it.
+        - {type: e1, owner: enemy, position: [26, 20], stance: 2, count: 3}
         - {type: fact, owner: enemy, position: [120, 20]}
     win_condition:
       all_of:

openra_bench/scenarios/packs/mfb-rotating-production-pressure.yaml

@@ -197,7 +197,9 @@ base:
 levels:
   # ── EASY ─────────────────────────────────────────────────────────
   # Minimum-viable rotation drill. WEST raid at tick 300 (3× e1),
-  # EAST raid at tick 1500 (3× e1). Six tanks (3 per base). max_turns
+  # EAST raid at tick 1500 (4× e1 — one heavier than the WEST raid
+  # so a single late-routed tank cannot hold it; see the
+  # scheduled_events note below). Six tanks (3 per base). max_turns
   # 70 ⇒ reachable tick 93+90·69 = 6303; within_ticks 5400 +
   # fail after_ticks 5401 ⇒ a non-finisher crosses 5401 well before
   # the episode caps ⇒ real LOSS, not a DRAW.
@@ -248,10 +250,18 @@ levels:
           type: spawn_actors
           actors:
             - {type: e1, owner: enemy, position: [18, 19], stance: 3, count: 3}
+        # EAST raid is 4× e1 (one heavier than the WEST raid): the
+        # engine `count:` spawn-spread fix scattered the raid band
+        # across distinct cells, which softened its concentrated
+        # damage enough that a SINGLE late-routed tank could survive
+        # the old 3× band — inverting the `rotate-too-late` LOSS bar.
+        # 4× e1 re-tightens it: one tank (the too-late split) loses
+        # the EAST base, while the intended 3-4-tank EAST garrison
+        # still clears the raid and keeps the war factory.
         - tick: 1500
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [88, 19], stance: 3, count: 3}
+            - {type: e1, owner: enemy, position: [88, 19], stance: 3, count: 4}
     win_condition:
       all_of:
         # ≥4 medium tanks ALIVE — verifies the army was actually

openra_bench/scenarios/packs/mid-concede-vs-hold.yaml

@@ -407,9 +407,17 @@ levels:
         - {type: 2tnk, owner: agent, position: [54, 20], stance: 1, count: 2, health: 100, spawn_point: 1}
         - {type: 1tnk, owner: agent, position: [54, 22], stance: 1, count: 2, health: 100, spawn_point: 1}
         - {type: jeep, owner: agent, position: [54, 18], stance: 1, health: 100, spawn_point: 1}
-        # ── WEST enemy push (LIGHT) — far west, same 20-rifle band
-        # as easy/medium.
-        - {type: e1, owner: enemy, position: [5, 20], stance: 3, count: 20, health: 100}
+        # ── WEST enemy push (LIGHT) — far west. Raised 20->26 on
+        # HARD ONLY: the engine `count:` spawn-spread fix scatters the
+        # 5-unit flex squad across distinct cells instead of stacking
+        # it, which slightly improved the WEST garrison + HALF-flex
+        # trade — a split-defend (3 of 5 flex sent WEST) suddenly held
+        # the base and WON. 26 e1 re-tightens the WEST trade so the
+        # bare garrison + any partial flex (split / oscillate) is
+        # overrun, while the garrison + the FULL consolidated flex,
+        # ordered to ATTACK-MOVE in, still clears it and holds past
+        # the 2400 survival floor.
+        - {type: e1, owner: enemy, position: [5, 20], stance: 3, count: 26, health: 100}
         # ── EAST enemy push (HEAVY) — far east, same 72-rifle band
         # as easy/medium. Hard's earlier survival floor (2400 vs
         # easy/medium's 3000) is what makes the heavy push need to be

openra_bench/scenarios/packs/risk-blockade-bypass.yaml

@@ -10,11 +10,11 @@
 # detour. Brute-forcing the corridor reduces the force below the
 # attrition cap before reaching the objective.
 #
-# Difficulty escalates the cost of the wrong call: easy allows a
-# slim head-on (loss <=4); medium tightens the cap; hard introduces
-# a smaller forward force AND a tempo gate (the objective only
-# scores after tick t0, so committing too early through the corridor
-# and arriving wounded fails the after_ticks-then-attrition combo).
+# Difficulty escalates the cost of the wrong call: on every tier the
+# corridor garrison is sized so ANY engagement with it (full push or
+# probe-then-bounce) loses more than that tier's attrition cap before
+# the objective is reached — the corridor must be rejected, not
+# sampled. Medium tightens the cap; hard also shrinks the agent force.
 
 meta:
   id: risk-blockade-bypass
@@ -63,10 +63,16 @@ base:
     - {type: 1tnk, owner: agent, position: [8, 22],  stance: 3, count: 3}
     - {type: jeep, owner: agent, position: [6, 18],  stance: 3, count: 2}
     # Heavy CORRIDOR garrison (the wrong way): trades well, expensive
-    # to clear. Combination of e3 (rockets) + defensive gun.
-    - {type: e3,  owner: enemy, position: [55, 20], stance: 2, count: 4}
-    - {type: gun, owner: enemy, position: [57, 22], stance: 3, count: 1}
-    - {type: e1,  owner: enemy, position: [55, 22], stance: 2, count: 3}
+    # to clear. Combination of e3 (rockets) + defensive guns. Sized so
+    # any play that engages it loses more than the easy attrition cap
+    # (>4) before the objective is reached. A forward e3 block at
+    # x~46 (stance:2 — holds the lane, does not suicide-lunge) makes
+    # even a partial corridor commit (probe-then-bounce) bleed past
+    # the cap — the corridor must be REJECTED, not sampled.
+    - {type: e3,  owner: enemy, position: [46, 20], stance: 2, count: 8}
+    - {type: e3,  owner: enemy, position: [55, 20], stance: 2, count: 6}
+    - {type: gun, owner: enemy, position: [57, 22], stance: 3, count: 3}
+    - {type: e1,  owner: enemy, position: [55, 22], stance: 2, count: 4}
     # Light DETOUR garrison (the right way, top-N): one infantry
     # picket; clears at low cost.
     - {type: e1,  owner: enemy, position: [55, 5],  stance: 2, count: 2}
@@ -84,10 +90,10 @@ levels:
     description: >
       Reach the objective region near (110, 20) within 3500 ticks
       losing <=4 units. The central corridor garrison (heavy at
-      (55, 20–22)) out-trades you 1-for-1; pushing through is
-      possible but spends the whole loss budget and may run the
-      clock. The northern detour (light picket at (55, 5)) costs
-      ~600 extra ticks but only ~1 unit. Decision is route choice.
+      (46–57, 20–22)) out-trades you badly; pushing or even probing
+      it bleeds past the loss budget before the objective. The
+      northern detour (light picket at (55, 5)) costs ~600 extra
+      ticks but only ~1 unit. Decision is route choice.
     overrides: {}
     win_condition:
       all_of:

openra_bench/scenarios/packs/scout-count-defenders.yaml

@@ -20,16 +20,27 @@
 #   • Pre-placed agent base: fact + proc + powr + weap (+ fix for
 #     2tnk tech prereq) + 2× jeep scouts.
 #   • Pre-placed enemy defenders at the far-east `fact` landmark:
-#     K medium tanks (2tnk, stance:2 — hold post) PLUS two
-#     pillboxes (pbox) for
-#     extra defensive firepower. The pillboxes don't count toward
-#     `enemies_discovered_gte:K` (they're buildings) or
+#     K half-strength medium tanks (2tnk, stance:2 — hold post,
+#     health: 60) PLUS pillboxes (pbox) for extra defensive
+#     firepower — 2 on easy/hard, 1 on medium. The pillboxes don't
+#     count toward `enemies_discovered_gte:K` (they're buildings) or
 #     `units_killed_gte:K` (also buildings) — they are pure
 #     attrition-multiplier that prevents the under-build (2 tanks)
-#     from rushing through. Without the pillboxes a 2-medium-tank
-#     attacker can beat 3-4 medium-tank defenders cleanly; with the
-#     pillboxes' extra ~30 dps each, 2 attackers are wiped before
-#     scoring kills.
+#     from rushing through.
+#
+#   • SPAWN-SPREAD CALIBRATION (engine `count:` fix): the engine
+#     `count:N` expansion used to STACK all N defenders on the
+#     single declared cell; it now SPREADS them across N distinct
+#     cells (copy 0 at the anchor, copies 1..N-1 in outward rings).
+#     The spread tightened the K defenders into a focus-firing
+#     cluster, which won the attrition trade even at equal HP and
+#     equal count — a full-HP K-tank defender cluster beats a
+#     count-matched K-tank attacker column. The defenders are now a
+#     half-strength garrison (`health: 60`) so the count-matched
+#     attacker force wins the trade while an under-build (2 tanks)
+#     against K≥3 still loses; medium drops to 1 pillbox because a
+#     second pbox over-tunes the post-spread defence past what a
+#     K=3 attacker wave can beat.
 #   • Cash: $5100 — funds up to 6× 2tnk ($850 each). Tuned so the
 #     intended K-tanks build fits, and the over-build (queue all 6
 #     before sending) bleeds the production clock past the deadline.
@@ -85,9 +96,10 @@
 #     K≥3 levels.
 #   • build-max-force (queue all 6 medium tanks before sending):
 #     LOSS on every level — the 6-tank sequential queue takes
-#     ~3240 ticks plus ~1350 transit + 300 combat = ~4890 ticks,
-#     beyond every level's within_ticks deadline (max 4200 on
-#     hard). The over-build is a clean LOSS on every (level, seed).
+#     ~3240 ticks plus ~1350 transit + combat, so the engagement
+#     cannot even START before every level's within_ticks deadline
+#     (3300 easy / 3700 medium / 4500 hard). The over-build is a
+#     clean LOSS on every (level, seed).
 #   • intended count-then-build (scout to count K, queue exactly K
 #     tanks, attack-move east after all built): WINS on every
 #     (level, seed).
@@ -97,10 +109,11 @@
 #   • 2tnk build ~540 ticks each (sequential weap queue).
 #   • 2tnk transit ~15 ticks/cell × ~90 cells = ~1350 ticks.
 #   • Combat resolution against K defenders ≈ 300 ticks.
-#   • Intended K=2 finishes ≈ tick 2883 (2×540+1350+300 ≈ 2730);
-#     within_ticks 3300 accepts with slack.
+#   • Intended K=2 finishes ≈ tick 2793; within_ticks 3300 accepts
+#     with slack.
 #   • Intended K=3 finishes ≈ tick 3333; within_ticks 3700 accepts.
-#   • Intended K=4 finishes ≈ tick 3783; within_ticks 4200 accepts.
+#   • Intended K=4 finishes ≈ tick 3783–4143 (spawn-latitude
+#     dependent on hard); within_ticks 4500 accepts with slack.
 #   • Over-build 6 finishes ≈ tick 4890 (6×540+1350+300 = 4890),
 #     beyond ALL within_ticks deadlines ⇒ LOSS on every level.
 #
@@ -206,8 +219,11 @@ levels:
         - {type: jeep, owner: agent, position: [22, 19], stance: 1}
         - {type: jeep, owner: agent, position: [22, 21], stance: 1}
         # K=2 medium-tank defenders + 2 pillboxes for reinforcing
-        # fire. stance:2 (Defend — hold post) on the tanks.
-        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 2}
+        # fire. stance:2 (Defend — hold post) on the tanks. The
+        # defenders are a half-strength garrison (health: 60) so the
+        # count-matched attacker force wins the trade — see the
+        # spawn-spread calibration note in the header.
+        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 2, health: 60}
         - {type: pbox, owner: enemy, position: [108, 19]}
         - {type: pbox, owner: enemy, position: [108, 21]}
         # Unarmed enemy `fact` landmark — MustBeDestroyed, keeps
@@ -226,16 +242,16 @@ levels:
     max_turns: 38
 
   # ── MEDIUM ───────────────────────────────────────────────────────
-  # +1 axis: FOG + larger K. K=3 hidden medium-tank defenders + 2
-  # pillboxes. The brief does NOT name the count. The agent must
+  # +1 axis: FOG + larger K. K=3 hidden medium-tank defenders + 1
+  # pillbox. The brief does NOT name the count. The agent must
   # SCOUT (drive a jeep east) to read K=3 before committing the
   # build. Stall LOSES; under-build (assume K=2, field 2 tanks)
   # loses the attrition trade (kills < 3 ⇒ LOSS); over-build
   # (queue 6) misses the deadline.
   medium:
     description: >
-      An UNKNOWN number of enemy medium tanks (2tnk) plus two
-      reinforcing pillboxes defend the far-east enemy construction
+      An UNKNOWN number of enemy medium tanks (2tnk) plus a
+      reinforcing pillbox defend the far-east enemy construction
       yard in the fog. Your pre-placed base (fact, proc, powr,
       weap, fix) plus two jeep scouts and $5100 in cash funds up to
       6 medium tanks ($850 each) but you should build EXACTLY
@@ -256,10 +272,13 @@ levels:
         - {type: fix,  owner: agent, position: [18, 22]}
         - {type: jeep, owner: agent, position: [22, 19], stance: 1}
         - {type: jeep, owner: agent, position: [22, 21], stance: 1}
-        # K=3 medium-tank defenders + 2 pillboxes.
-        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 3}
+        # K=3 half-strength medium-tank defenders + 1 pillbox. One
+        # pillbox (not two) on medium: a second pbox over-tunes the
+        # defence so the count-matched K=3 attacker wave can no
+        # longer win the trade after the count-spread tightened the
+        # defender cluster — see the header calibration note.
+        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 3, health: 60}
         - {type: pbox, owner: enemy, position: [108, 19]}
-        - {type: pbox, owner: enemy, position: [108, 21]}
         - {type: fact, owner: enemy, position: [120, 20]}
     win_condition:
       all_of:
@@ -296,7 +315,7 @@ levels:
       K=2) loses the attrition trade and the kill bar; over-build
       (queue 6) misses the deadline. Win when ≥4 enemy units
       discovered AND ≥4 killed AND your fact is still standing AND
-      tick ≤ 4200.
+      tick ≤ 4500.
     starting_cash: 5100
     overrides:
       actors:
@@ -316,10 +335,10 @@ levels:
         - {type: fix,  owner: agent, position: [18, 28], spawn_point: 1}
         - {type: jeep, owner: agent, position: [22, 25], stance: 1, spawn_point: 1}
         - {type: jeep, owner: agent, position: [22, 27], stance: 1, spawn_point: 1}
-        # K=4 medium-tank defenders centred at mid-y; the cloud
-        # sits at y=20 so either spawn must scout from its actual
-        # base latitude. stance:2 Defend (hold post).
-        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 4}
+        # K=4 half-strength medium-tank defenders centred at mid-y;
+        # the cloud sits at y=20 so either spawn must scout from its
+        # actual base latitude. stance:2 Defend (hold post).
+        - {type: 2tnk, owner: enemy, position: [110, 20], stance: 2, count: 4, health: 60}
         - {type: pbox, owner: enemy, position: [108, 19]}
         - {type: pbox, owner: enemy, position: [108, 21]}
         # Persistent unarmed enemy `fact` marker — MustBeDestroyed,
@@ -330,9 +349,9 @@ levels:
         - {enemies_discovered_gte: 4}
         - {units_killed_gte: 4}
         - {building_count_gte: {type: fact, n: 1}}
-        - {within_ticks: 4200}
+        - {within_ticks: 4500}
     fail_condition:
       any_of:
-        - {after_ticks: 4201}
+        - {after_ticks: 4501}
         - {not: {building_count_gte: {type: fact, n: 1}}}
-    max_turns: 48
+    max_turns: 52

openra_bench/scenarios/packs/tp-survive-n-turns.yaml

@@ -249,7 +249,7 @@ levels:
     max_turns: 45
 
   # ── MEDIUM ───────────────────────────────────────────────────────
-  # +1 controlled variable: HEAVIER sustained pressure (FOUR
+  # +1 controlled variable: HEAVIER sustained pressure (FIVE
   # scheduled rifle waves on top of the opening wave) and a longer
   # hold (T=3600, ~turn 41). N stays at 4 of 6. max_turns 55 →
   # reachable tick 93 + 90·54 = 4953 > the within_ticks 4500 band
@@ -288,40 +288,45 @@ levels:
       # FOUR scheduled reinforcement waves — sustained pressure
       # across the whole window. Block sits inside `overrides:` (the
       # loader drops it at the level root). Pure e1 (no e3) for the
-      # same reason as the opening wave. The 6×3 opening + four 6×3
-      # waves are re-tuned so spread (stall) fire loses the fact and
-      # the focused active defence holds it with ≥4 tanks.
+      # same reason as the opening wave. After the engine `count:`
+      # spawn-spread fix (copies 1..N now SPREAD across outward rings
+      # instead of stacking on the anchor) the rifle band lands a few
+      # cells nearer the base, so a 6-per-row reinforcement wave
+      # tipped the intended active defence below the 4-tank floor.
+      # Re-tuned to 5 per row (15 e1/wave): the 6×3 opening + five
+      # 5×3 waves keep spread (stall) fire losing the fact while the
+      # focused active defence holds it with ≥4 tanks.
       scheduled_events:
         - tick: 700
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 6}
+            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 5}
         - tick: 1300
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 6}
+            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 5}
         - tick: 1900
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 6}
+            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 5}
         - tick: 2500
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 6}
+            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 5}
         - tick: 3100
           type: spawn_actors
           actors:
-            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 6}
-            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 6}
+            - {type: e1, owner: enemy, position: [64, 18], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 20], stance: 3, count: 5}
+            - {type: e1, owner: enemy, position: [64, 22], stance: 3, count: 5}
     win_condition:
       all_of:
         - {building_count_gte: {type: fact, n: 1}}

tests/test_combat_bait_counter_attack.py

@@ -181,7 +181,8 @@ def test_timeout_reachable_inside_max_turns():
 
 def test_hard_has_two_spawn_point_groups():
     """Hard-tier curation: ≥2 distinct agent spawn_point groups so the
-    seed round-robins the staging latitude (north y=10 / south y=30).
+    seed round-robins the staging DISTANCE from the yard (far west
+    x=6 / forward x=18, both on the north flank y=12).
     Engine-roundtrip is asserted by tests/test_hard_tier.py."""
     c = compile_level(load_pack(PACK_PATH), "hard")
     groups = {

tests/test_mid_concede_vs_hold.py

@@ -33,6 +33,17 @@ dropped (the win is a survival-band check), a `not proc:1` fail
 clause, a persistent unarmed enemy `fact` marker (anti auto-DRAW),
 the hard survival floor at tick 2400 and its attrition cap at 18.
 
+Spawn-spread recalibration (engine `count:` fix): the engine used
+to stack all N units of a `count:N` group on the single declared
+cell; it now SPREADS them across N distinct cells. On HARD the
+scattered 5-unit flex squad slightly improved the WEST garrison +
+HALF-flex trade, so split-defend (3 of 5 flex sent WEST) suddenly
+held the WEST base and WON — the "laziest play wins" inversion this
+pack had previously closed. Hard's WEST light push was raised
+20->26 to re-tighten that trade: the bare garrison + any partial
+flex (split / oscillate) is overrun, while the garrison + the FULL
+consolidated flex still clears it and holds past the 2400 floor.
+
 The capability stays load-bearing — stall / split / oscillate /
 wrong-side consolidate all LOSE; only consolidate-on-the-light-side
 AND attack-move into the push WINS.

tests/test_scout_count_defenders.py

@@ -1,25 +1,33 @@
 """scout-count-defenders pack — Wave-8 PERCEPTION exact-count force
 sizing.
 
-The agent scouts to count K medium-tank enemy defenders (2tnk),
-backed by 2 reinforcing pillboxes (pbox) for extra defensive DPS,
+The agent scouts to count K half-strength medium-tank enemy
+defenders (2tnk, health: 60), backed by reinforcing pillboxes
+(pbox — 2 on easy/hard, 1 on medium) for extra defensive DPS,
 then builds EXACTLY K medium tanks (2tnk) to defeat them. The
 pillboxes are STRUCTURES (counted as buildings, not units) so they
 do NOT count toward `enemies_discovered_gte:K` or
 `units_killed_gte:K`; they are pure attrition multiplier that
 prevents the under-build (2 tanks) from rushing through.
 
+The engine `count:` expansion now SPREADS the defender cluster
+across distinct cells (previously stacked on one cell); the spread
+tightened the defenders into a focus-firing cluster, so the
+defenders are a half-strength garrison (`health: 60`) to keep the
+count-matched attacker force winning the trade.
+
 Discriminations (CLAUDE.md bar):
 
   * stall (only observe): LOSS — after_ticks fail clause bites.
   * build-min-force (always assume K=2, build 2 tanks regardless of
     actual count): LOSS on medium (K=3) and hard (K=4) — the 2-tank
-    wave is wiped by the K-defender + 2-pbox combination before
+    wave is wiped by the K-defender + pbox combination before
     scoring K kills.
   * build-max-force (queue all OVER_BUILD_N=6 tanks then send):
     LOSS on every level — the 6-tank sequential queue takes ~3240
-    ticks plus ~1350 transit + ~300 combat = ~4890 ticks, beyond
-    every level's within_ticks deadline (max 4200 on hard).
+    ticks plus ~1350 transit + combat, so the engagement cannot
+    even start before every level's within_ticks deadline (max
+    4500 on hard).
   * intended count-then-build (scout to read K, build exactly K
     tanks, attack-move east): WINS on every (level, seed).
 
@@ -245,7 +253,7 @@ def _intended_policy(level: str):
 def _under_build_policy():
     """Always build exactly 2 tanks regardless of K. WINS on easy
     (K=2 — that's the matching amount) and LOSES on medium/hard
-    (K=3/K=4) — 2 medium tanks vs 3-4 defending tanks + 2 pillboxes
+    (K=3/K=4) — 2 medium tanks vs 3-4 defending tanks + pillboxes
     is an attrition trade the attackers lose."""
     state = {"queued": 0, "sent": False, "scouted": False}