build-defensive-tower-cluster: tailor map to chokepoint-arena (96x40)

Replace the shared rush-hour-arena (128x40 open) with a per-pack
chokepoint-arena map (96x40) that physically funnels the rush band
through a single 4-cell corridor (y=18..21 through a vertical water
wall at x=43..52, pinch_width=10). The agent fact sits at x=10 in the
WEST lobe; attackers spawn in the EAST lobe at (70,20) and the rusher
bot's A* path to the agent centroid has no alternative but the
corridor — water elsewhere along x=43..52.

Why the new geometry strengthens the discrimination:
* Stall — rush funnels through the corridor and razes the fact.
* Pbox-line (cells [(20,18),(25,19),(30,20),(35,21)] in the agent's
west lobe, between fact and corridor) spreads thin along the
y=18..21 attack lane, fails the radius-4 region clause around
the fact (0 of 4 inside the disc), and dissipates kill rate
across the corridor mouth so the band reaches the fact.
* Pure-army — fails count clause and is overrun.
* Intended cluster (4 pbox wrapping the active fact within radius
4) sees the WHOLE band sequentially as it funnels out of the
corridor; overlapping pbox dps shreds it.

Map spec (chokepoint-arena, 96x40, cordon=2):
* pinch_x=48, pinch_width=10, corridor_width=4, corridor_y=20
* Single materialised map shared across all three tiers
(build-defensive-tower-cluster-chokepoint-96x40.oramap)

Actor changes:
* Drop the (120,20) anti-DRAW enemy fact (out of bounds on a
96-wide map); relocate to (90,20) in the east lobe.
* Move attacker spawn from (60..62, *) to (70..72, *) (still east
lobe; closer to the corridor mouth on the smaller map).
* Hard: add a scheduled_events spawn_actors wave at tick 2700
(4xe1 + 1xe3) so the cluster has to handle reinforcements after
it assembles — strengthens the bar against line/pure-army.

Same 4-pbox / 2400-cash budget and kill bars (3/5/6) as before; only
the map and per-tier actor positions move. Win/fail predicates
unchanged.

Validation (scripted policies, tests/test_build_defensive_tower_cluster.py):
21 tests pass (4 seeds x 3 tiers):
* Intended adaptive cluster WIN every level + every seed.
* Stall LOSS x12, pbox-line LOSS x12, pure-army LOSS x12.
* Wrong-centre cluster LOSS on hard (fact-flip discrimination).
* Determinism, invariants, fact-alive predicate, hard
spawn_points, reachable timeout all pass.

openra_bench/scenarios/packs/build-defensive-skirt-corners.yaml

@@ -123,7 +123,41 @@ meta:
     - "military quadrant doctrine"
   author: "openra-bench"
 
-base_map: rush-hour-arena
+base_map:
+  # Tailored arena: 128x40 open rectangle (cordon 2) so the agent fact at
+  # the map CENTRE (64,20) is EQUIDISTANT from all four diagonal corners
+  # — the geometric prerequisite for the skirt-vs-concentrate test. Four
+  # symmetric far-corner water markers (2x2 each, near the map's NE/NW/
+  # SE/SW edges) visually demarcate the four quadrants without touching
+  # any rusher path or the corner pbox discs themselves: the markers sit
+  # OUTSIDE every corner-region disc (radius-4 around (46,9)/(82,9)/(46,
+  # 31)/(82,31) for easy/medium, and (32,9)/(96,9)/(32,31)/(96,31) for
+  # the hard WEST/EAST seeds) and OUTSIDE every diagonal rush corridor
+  # from corner spawn to fact, so they affect neither the build placement
+  # nor the rush charge. They give each quadrant a distinct terrain
+  # signature so a "centre-only" or "mass-on-one-corner" defense reads
+  # visually wrong before the predicate even bites. The 128-wide canvas
+  # is the minimum that admits the hard EAST seed's NE corner at x=96
+  # with playable cells out to x=125. Bounds 2,2,124,36.
+  generator: arena
+  name: build-defensive-skirt-corners-128x40
+  width: 128
+  height: 40
+  cordon: 2
+  title: Defensive Skirt Four Corners 128x40
+  obstacles:
+    # NW quadrant marker — between map NW corner and the easy NW disc
+    # (46,9). At (22,5)-(23,6) it sits ~24 cells from the NW corner disc
+    # centre, well outside its radius-4 envelope, and ~10 cells off the
+    # straight diagonal from (46,9) toward the fact at (64,20) so it
+    # never blocks a rush.
+    - {x: 22, y: 5,  w: 2, h: 2}
+    # NE quadrant marker — symmetric, near map NE corner.
+    - {x: 104, y: 5, w: 2, h: 2}
+    # SW quadrant marker — symmetric, near map SW corner.
+    - {x: 22, y: 33, w: 2, h: 2}
+    # SE quadrant marker — symmetric, near map SE corner.
+    - {x: 104, y: 33, w: 2, h: 2}
 
 base:
   agent: {faction: allies}

openra_bench/scenarios/packs/build-defensive-tower-cluster.yaml

@@ -2,22 +2,43 @@
 #
 # REASONING focus: CONCENTRATED-DEFENSE topology. A single high-value
 # building (the agent fact) is the protected asset. A heavy rush band
-# charges the agent's centroid — which IS the fact. The right doctrine
-# is a TIGHT CLUSTER of pillboxes WRAPPING the fact (overlapping fields
-# of fire on the single chokepoint that matters: the cell the fact sits
-# on), NOT a thin line spread across the map.
+# is FORCED through a single narrow corridor (the chokepoint between
+# two open lobes), then converges on the agent's centroid — which IS
+# the fact. The right doctrine is a TIGHT CLUSTER of pillboxes
+# WRAPPING the fact (overlapping fields of fire on the single cell the
+# fact sits on), NOT a thin line spread between the fact and the
+# corridor — because the line dissipates its kill rate across the
+# 4-cell corridor mouth and cannot mass enough fire on the rush band
+# before the band reaches the unprotected fact.
 #
 # The win predicate makes the topology decision load-bearing — total
 # pbox count alone is not enough; ≥3 of the pillboxes must sit INSIDE
 # a radius-4 disc around the fact. A pbox-LINE policy (the obvious
-# sibling: pboxes strung along the east-west attack lane) satisfies the
-# count but FAILS the region bar AND lets the concentrated rifle/rocket
-# band overwhelm the thin frontage at the fact — line LOSES, intended
-# cluster WINS.
+# sibling: pboxes strung between the fact and the corridor mouth)
+# satisfies the count but FAILS the region bar AND lets the
+# concentrated rifle/rocket band overwhelm the thin frontage at the
+# fact — line LOSES, intended cluster WINS.
 #
-# Sibling to a hypothetical `build-defensive-tower-line` (which would
-# reward a long-frontage line across a wide threat-axis); on THIS pack
-# the threat is FOCUSED on the fact, so the doctrine inverts.
+# Sibling to `build-defensive-tower-line` (which rewards a long-frontage
+# line across a wide threat-axis); on THIS pack the threat is FOCUSED
+# on a single point (the agent fact) by the corridor geometry, so the
+# doctrine inverts.
+#
+# MAP (per tier, materialised via `chokepoint-arena` generator):
+#   • All tiers — 96×40, cordon=2, single pinch wall at x=43..52
+#     (pinch_width=10) with a 4-cell corridor at y=18..21 (the only
+#     passable cells between west and east lobes).
+#   • West lobe (x∈[2,42]) — agent base. Fact sits at the WEST EDGE
+#     (x=10) so the rush MUST traverse the corridor THEN the entire
+#     west lobe to reach the fact; the cluster wrapping the fact has
+#     time to assemble before the band arrives.
+#   • East lobe (x∈[53,93]) — attackers spawn here; the rusher bot
+#     charges the agent fact, so its A* path is forced through the
+#     single y=18..21 corridor (water elsewhere along x=43..52).
+#   • Hard tier flips the fact NORTH (y=14) / SOUTH (y=26) by seed;
+#     the corridor stays at y=20 so the rush still has to traverse it
+#     AND then turn north/south to reach the active fact — the
+#     cluster centre must follow the fact.
 #
 # Real-world anchor:
 #   • ERQA spatial-commit primitive — place a small, finite resource set
@@ -30,44 +51,54 @@
 #     uniform spread across all approaches.
 #
 # Discrimination (four-script bar — scripted, no model needed):
-#   • stall (observe-only): spends nothing, the rush razes the fact —
-#     fact-alive fail clause fires → LOSS (no draw degeneracy because
-#     `after_ticks` also fires as a backstop).
+#   • stall (observe-only): spends nothing, the rush funnels through
+#     the corridor and razes the fact — fact-alive fail clause fires
+#     → LOSS (no draw degeneracy because `after_ticks` also fires as a
+#     backstop).
 #   • pbox-line (the "build-defensive-tower-line" solution — pboxes
-#     strung along the attack lane far from the fact): meets the count
-#     clause but FAILS the region clause (at most 0-1 line-pbox lands
-#     inside the radius-4 disc around the fact) AND the line lets the
-#     rusher band slip through any gap to reach the fact → LOSS.
+#     strung along the y=18..21 attack lane far from the fact): meets
+#     the count clause but FAILS the region clause (0 line-pboxes land
+#     inside the radius-4 disc around the fact) AND the spread lets
+#     the rusher band slip through gaps to reach the fact → LOSS.
 #   • pure-army (only e1, no pbox): FAILS the count clause AND the
 #     region clause; the rifle wall alone cannot stop the heavier
 #     medium/hard waves either → LOSS.
 #   • intended CLUSTER (3-4 pboxes wrapping the fact within radius 4):
-#     overlapping pbox fire shreds the focused rush; count + region +
-#     kill + fact-alive clauses all satisfied → WIN.
+#     overlapping pbox fire shreds the focused rush at the fact;
+#     count + region + kill + fact-alive clauses all satisfied → WIN.
 #
 # Why the spec works (engine combat sheet, per CLAUDE.md):
 #   • pbox = pillbox (anti-infantry base defence; cost 600cr; dps very
 #     high vs rifle infantry; 1×1 footprint, easy to cluster). The
 #     defense queue is SEPARATE from infantry so parallel queues are
 #     allowed (CLAUDE.md).
-#   • A radius-4 disc around the fact at (10,20) contains ~50 cells;
-#     plenty of room for ≥3 pboxes after the fact's 2×2 footprint plus
-#     the pre-placed tent/powr are subtracted.
+#   • A radius-4 disc around the fact contains ~50 cells; plenty of
+#     room for ≥3 pboxes after the fact's 2×2 footprint plus the
+#     pre-placed tent/powr are subtracted.
 #   • `building_count_gte:{fact,1}` (PRESENT-TENSE) is the fact-alive
 #     check — `has_building` is a one-shot ever-seen set that stays true
 #     after the fact is razed (documented CLAUDE.md footgun).
 #   • `after_ticks: BUDGET+1` fail clause is reachable within max_turns
 #     (no interrupts ⇒ exactly 90 ticks/step ⇒ max tick = 93+90·(N-1));
 #     stallers hit it and LOSE rather than draw.
-#   • A persistent unarmed enemy `fact` at (120,20) keeps the episode
-#     alive past full rusher elimination so the win/fail evaluator sees
-#     the terminal frame (no DRAW collapse on enemy-wipe auto-done).
+#   • A persistent unarmed enemy `fact` at the far east edge keeps the
+#     episode alive past full rusher elimination so the win/fail
+#     evaluator sees the terminal frame (no DRAW collapse on
+#     enemy-wipe auto-done). On a 96-wide map it sits at (90, *) —
+#     well within the playable rectangle.
 #   • The `rusher` bot relentlessly charges the agent's centroid (the
 #     agent fact) — that's WHY a cluster at the fact wins and a line
-#     away from it doesn't: the rush focal point is the fact itself.
-#   • Cash is intentionally tight (exactly N×600cr for N pboxes, with a
-#     small slack on hard for one extra defender unit). A model that
-#     wastes cash on extra units cannot complete the pbox count.
+#     between the fact and the corridor doesn't: the rush focal point
+#     is the fact itself, and the corridor funnels the band so a
+#     cluster sees the WHOLE band sequentially.
+#   • Cash is intentionally tight (exactly 4×600cr = 2400 for the four
+#     pboxes). A model that wastes cash on extra units cannot complete
+#     the pbox count.
+#   • Hard tier adds a SECOND reinforcement wave via `scheduled_events`
+#     at tick 2700 (~30 turns in); the wave is heavier and timed AFTER
+#     the cluster has had time to assemble, so the bar discriminates:
+#     a working cluster shreds the wave too; a pbox-line / pure-army
+#     defence cannot survive both waves.
 
 meta:
   id: build-defensive-tower-cluster
@@ -75,13 +106,14 @@ meta:
   capability: reasoning
   real_world_meaning: >
     A single high-value building (the fact — your construction yard) is
-    the protected asset, and a concentrated enemy band charges directly
-    at it. Right doctrine: a TIGHT CLUSTER of pillboxes WRAPPING the
-    fact, overlapping fields of fire on the single point that matters.
-    Wrong doctrine: a thin LINE of pillboxes strung along the attack
-    lane far from the fact — the line satisfies a naive "we built N
-    defences" count but cannot stop a focused rush from reaching the
-    fact, because the firepower is too dispersed to mass on the threat.
+    the protected asset, and a concentrated enemy band funnelled
+    through a single narrow corridor charges directly at it. Right
+    doctrine: a TIGHT CLUSTER of pillboxes WRAPPING the fact,
+    overlapping fields of fire on the single point that matters. Wrong
+    doctrine: a thin LINE of pillboxes strung between the fact and the
+    corridor mouth — the line satisfies a naive "we built N defences"
+    count but cannot stop a focused rush from reaching the fact,
+    because the firepower is too dispersed to mass on the threat.
     The win predicate makes the topology decision load-bearing: total
     pbox count alone is not enough; ≥3 of the pillboxes must sit INSIDE
     the radius-4 disc around the fact, AND the fact must survive.
@@ -100,7 +132,20 @@ meta:
     - "defense-in-depth around critical infrastructure"
   author: "openra-bench"
 
-base_map: rush-hour-arena
+# Base map: per-tier override below. The default below is the
+# chokepoint geometry every level inherits (the only mode this pack
+# supports — see the topology rationale at the top of the file).
+base_map:
+  generator: chokepoint-arena
+  name: build-defensive-tower-cluster-chokepoint-96x40
+  width: 96
+  height: 40
+  cordon: 2
+  pinch_x: 48
+  pinch_width: 10
+  corridor_width: 4
+  corridor_y: 20
+  title: Cluster Chokepoint 96x40
 
 base:
   agent: {faction: allies}
@@ -116,9 +161,10 @@ base:
   planning: true
   # No interrupts — defense topology is a STATIC up-front decision; the
   # threat axis is known a priori (the rusher always charges the agent
-  # fact). Disabling interrupts also makes the tick budget deterministic
-  # (exactly 90 ticks/step ⇒ max_turns is a hard tick budget that the
-  # `after_ticks` fail clause reliably bites in).
+  # fact through the only passable corridor). Disabling interrupts also
+  # makes the tick budget deterministic (exactly 90 ticks/step ⇒
+  # max_turns is a hard tick budget that the `after_ticks` fail clause
+  # reliably bites in).
   interrupts: {}
   termination:
     max_ticks: 12000
@@ -128,36 +174,25 @@ levels:
   # ── EASY ── bare CONCENTRATED-DEFENSE skill: budget covers exactly
   # 4 pbox (2400cr). Win requires ≥3 of the 4 INSIDE the radius-4 disc
   # around the fact at (10,20). A line layout (pbox strung at x=20..35
-  # along the attack lane) lands ≤1 pbox inside the disc → region clause
-  # fails. Stall loses (clock OR fact razed); pure-army loses (count).
+  # along the corridor band y=18..21) lands 0 pbox inside the disc →
+  # region clause fails AND the spread can't mass enough fire on the
+  # corridor mouth to stop the band. Stall loses (clock OR fact razed);
+  # pure-army loses (count + fact razed).
   # max_turns 60 ⇒ reachable tick 93+90·59 = 5403; deadline 5400.
   easy:
-    # Original (pre-verbosity-sweep) description preserved for
-    # contributors. The trimmed version below removes scripted-policy
-    # spoilers and cell-coord dumps; load-bearing intent kept.
-    #
-    # A rusher band of rifle infantry charges your construction yard
-    # (fact, at (10,20)). Build 4 pillboxes (pbox — 600cr each, budget
-    # exactly 2400) AND place at least 3 of them INSIDE the radius-4
-    # disc around the fact, so their overlapping fields of fire shred
-    # the rush at the doorstep of the protected asset. A thin pillbox
-    # LINE strung along the attack lane (pboxes at x=20..35) fails the
-    # region clause (none land inside the radius-4 disc around the
-    # fact) and also lets the rifle band slip through. Stall, pure-
-    # army, and pbox-line all LOSE. The fact must survive; ≥3 enemy
-    # units must die before tick 5400.
     description: >
-      A rifle band charges your construction yard from the east. Budget
-      $2400 — build four pillboxes with at least three sitting within
-      four cells of the yard. Keep the yard alive, kill three enemies,
-      within about 60 turns.
+      A rifle band funnels through the single narrow corridor (y=18..21
+      through the wall at x=43..52) and charges your construction yard
+      on the west. Budget $2400 — build four pillboxes with at least
+      three sitting within four cells of the yard. Keep the yard alive,
+      kill three enemies, within about 60 turns.
     starting_cash: 2400
     overrides:
       actors:
-        # Pre-placed agent base. Fact at (10,20); tent and powr WEST
-        # of the fact (behind it) so they aren't in the rush line of
-        # fire and they leave plenty of cells inside the radius-4
-        # disc free for cluster pboxes.
+        # Pre-placed agent base in the WEST lobe. Fact at (10,20);
+        # tent and powr WEST of the fact (behind it) so they aren't in
+        # the rush line of fire and they leave plenty of cells inside
+        # the radius-4 disc free for cluster pboxes.
         - {type: fact, owner: agent, position: [10, 20]}
         - {type: tent, owner: agent, position: [6,  18]}
         - {type: powr, owner: agent, position: [6,  22]}
@@ -168,15 +203,16 @@ levels:
         - {type: e1,   owner: agent, position: [12, 19], stance: 2}
         - {type: e1,   owner: agent, position: [12, 20], stance: 2}
         - {type: e1,   owner: agent, position: [12, 21], stance: 2}
-        # Rusher band at mid-east — forced through the y=20 lane on
-        # the way to the agent centroid (the fact). 5× e1 is enough
-        # to overwhelm a thin line / no-pbox defence but a cluster of
-        # ≥3 pbox (dps very high vs infantry, overlapping fields of
-        # fire) shreds it. Off the (50,20) silently-failing column.
-        - {type: e1, owner: enemy, position: [60, 20], stance: 3, count: 5}
+        # Rusher band in the EAST lobe — forced through the y=18..21
+        # corridor on the way to the agent fact at (10,20). 5× e1 is
+        # enough to overwhelm a thin line / no-pbox defence; a cluster
+        # of ≥3 pbox (dps very high vs infantry, overlapping fields of
+        # fire) shreds it as it funnels out of the corridor mouth.
+        - {type: e1, owner: enemy, position: [70, 20], stance: 3, count: 5}
         # Unarmed high-HP marker (anti-DRAW): keeps the episode alive
         # past full rusher elimination so the win/fail check fires.
-        - {type: fact, owner: enemy, position: [120, 20]}
+        # Far east corner of the east lobe.
+        - {type: fact, owner: enemy, position: [90, 20]}
     win_condition:
       all_of:
         - {building_count_gte: {type: pbox, n: 4}}
@@ -196,29 +232,18 @@ levels:
   # line layout (1 at fact, 3 along the lane) still meets the count
   # but FAILS the region (1 of 4 inside the disc, not the required 3)
   # AND the line cannot mass enough firepower to stop the heavier
-  # band. The intended cluster (3 pbox INSIDE the disc + 1 spare
-  # nearby, OR all 4 inside the disc) survives. Kill bar raised to
-  # 5 to require the cluster to actually fight, not just absorb.
+  # band funnelled through the corridor. The intended cluster (3
+  # pbox INSIDE the disc + 1 spare nearby, OR all 4 inside the disc)
+  # survives. Kill bar raised to 5 to require the cluster to actually
+  # fight, not just absorb.
   # max_turns 60 ⇒ reachable tick 5403; deadline 5400.
   medium:
-    # Original (pre-verbosity-sweep) description preserved for
-    # contributors. The trimmed version below removes scripted-policy
-    # spoilers and cell-coord dumps; load-bearing intent kept.
-    #
-    # The full rush band — 6 rifle infantry + 2 rocket soldiers —
-    # charges your construction yard (fact, at (10,20)). Build 4
-    # pillboxes (budget exactly 2400cr = 4 pbox at 600 each) AND
-    # place at least 3 of them INSIDE the radius-4 disc around the
-    # fact. A pillbox LINE (1 at the fact, 3 along the lane) meets
-    # the count but FAILS the region bar (1 of 4 inside the disc, not
-    # the required 3) AND cannot mass enough firepower to blunt the
-    # heavier band. Stall, pure-army, and pbox-line all LOSE. The
-    # fact must survive; ≥5 enemy units must die before tick 5400.
     description: >
-      Heavier rush — rifles plus rocket soldiers — charges your
-      construction yard. Budget $2400, build four pillboxes with at
-      least three within four cells of the yard. Keep it alive, kill
-      five, within about 60 turns.
+      Heavier rush — rifles plus rocket soldiers — funnels through the
+      single narrow corridor and charges your construction yard.
+      Budget $2400, build four pillboxes with at least three within
+      four cells of the yard. Keep it alive, kill five, within about
+      60 turns.
     starting_cash: 2400
     overrides:
       actors:
@@ -231,16 +256,14 @@ levels:
         - {type: e1,   owner: agent, position: [12, 20], stance: 2}
         - {type: e1,   owner: agent, position: [12, 21], stance: 2}
         - {type: e1,   owner: agent, position: [8,  20], stance: 2}
-        # Canonical heavy rush: 6× e1 + 2× e3 at x=60 (mid-east), all
-        # at y=20 so the band converges through the protected lane on
-        # the way to the agent fact at (10,20). Off the (50,20)
-        # silently-failing column. Heavier band ⇒ a line / pure-army
-        # defence is overrun in ~6-8 turns; the cluster (overlapping
-        # pbox dps) holds.
-        - {type: e1, owner: enemy, position: [60, 20], stance: 3, count: 6}
-        - {type: e3, owner: enemy, position: [62, 20], stance: 3, count: 2}
-        # Anti-DRAW marker.
-        - {type: fact, owner: enemy, position: [120, 20]}
+        # Canonical heavy rush: 6× e1 + 2× e3 in the east lobe at
+        # y=20, forced through the y=18..21 corridor on the way to
+        # the agent fact. A line / pure-army defence is overrun in
+        # ~6-8 turns; the cluster (overlapping pbox dps) holds.
+        - {type: e1, owner: enemy, position: [70, 20], stance: 3, count: 6}
+        - {type: e3, owner: enemy, position: [72, 20], stance: 3, count: 2}
+        # Anti-DRAW marker (east edge).
+        - {type: fact, owner: enemy, position: [90, 20]}
     win_condition:
       all_of:
         - {building_count_gte: {type: pbox, n: 4}}
@@ -263,32 +286,21 @@ levels:
   # READ the fact's latitude from observation and place the cluster
   # relative to it. Enemies don't honour spawn_point (engine:
   # oramap.rs::expand_scenario_actors per CLAUDE.md), so the rusher
-  # band is staged symmetrically at the matching latitude on BOTH
-  # candidate routes; the rusher charges agent centroid so it
-  # converges on whichever fact is present this seed. Total pbox
-  # grows to 4 (still 2400cr, exact). Kill bar 6.
+  # band is staged at the corridor latitude (y=20) — the only
+  # passable cells through the wall — and converges on whichever fact
+  # is present this seed by turning north or south after emerging
+  # from the corridor mouth. Additionally a `scheduled_events`
+  # reinforcement wave fires at tick 2700 (~30 turns), giving the
+  # cluster time to assemble first; the wave overwhelms a line /
+  # pure-army defence but a real cluster handles both waves.
   # max_turns 60 ⇒ reachable tick 5403; deadline 5400.
   hard:
-    # Original (pre-verbosity-sweep) description preserved for
-    # contributors. The trimmed version below removes scripted-policy
-    # spoilers and cell-coord dumps; load-bearing intent kept.
-    #
-    # The agent construction yard (fact) flips between NORTH (y=14)
-    # and SOUTH (y=26) by seed; the cluster centre must follow.
-    # Build 4 pillboxes (budget 2400cr = 4 pbox at 600 each) AND
-    # place at least 3 of them INSIDE the radius-4 disc around the
-    # CURRENT fact (either (10,14) or (10,26) — read it from the
-    # observation). A line layout, a cluster around (10,20) (the
-    # old centre that no longer holds the fact), or any plan that
-    # doesn't follow the fact's latitude FAILS the region bar AND
-    # lets the rush reach the fact. Stall, pure-army, pbox-line
-    # all LOSE. The fact must survive; ≥6 enemy units must die
-    # before tick 5400.
     description: >
-      Same rush, but the yard sits in the NORTH or SOUTH west-edge slot
-      by seed. Budget $2400, build four pillboxes with at least three
-      within four cells of the current yard. Keep it alive, kill six,
-      within about 60 turns.
+      Same corridor and same focused rush, but the yard sits in the
+      NORTH or SOUTH west-edge slot by seed, and a reinforcement wave
+      arrives mid-episode. Budget $2400, build four pillboxes with at
+      least three within four cells of the current yard. Keep it
+      alive, kill six, within about 60 turns.
     starting_cash: 2400
     overrides:
       actors:
@@ -310,18 +322,23 @@ levels:
         - {type: e1,   owner: agent, position: [12, 27], stance: 2, spawn_point: 1}
         - {type: e1,   owner: agent, position: [8,  26], stance: 2, spawn_point: 1}
         # Enemies don't honour spawn_point (CLAUDE.md). Stage the rush
-        # band at BOTH candidate latitudes (y=14 AND y=26); the rusher
-        # charges agent centroid so the band whose latitude matches the
-        # active fact converges on it. (Both bands always place; the
-        # off-latitude band wanders harmlessly toward the unused
-        # latitude — but the on-latitude band is heavy enough on its
-        # own to overwhelm a thin / mis-placed defence.)
-        - {type: e1, owner: enemy, position: [60, 14], stance: 3, count: 4}
-        - {type: e3, owner: enemy, position: [62, 14], stance: 3, count: 1}
-        - {type: e1, owner: enemy, position: [60, 26], stance: 3, count: 4}
-        - {type: e3, owner: enemy, position: [62, 26], stance: 3, count: 1}
-        # Anti-DRAW marker on the symmetry axis.
-        - {type: fact, owner: enemy, position: [120, 20]}
+        # band at the corridor latitude (y=20) — the only passable
+        # cells through the wall — so it funnels through and then
+        # turns N/S after emerging into the west lobe toward the
+        # active fact. Heavier than medium: 5×e1 + 2×e3.
+        - {type: e1, owner: enemy, position: [70, 20], stance: 3, count: 5}
+        - {type: e3, owner: enemy, position: [72, 20], stance: 3, count: 2}
+        # Anti-DRAW marker on the symmetry axis (east edge).
+        - {type: fact, owner: enemy, position: [90, 20]}
+      # Mid-episode reinforcement wave (after the cluster has time to
+      # assemble) — a working cluster shreds both waves; a line /
+      # pure-army defence cannot survive the combined pressure.
+      scheduled_events:
+        - tick: 2700
+          type: spawn_actors
+          actors:
+            - {type: e1, owner: enemy, position: [70, 20], stance: 3, count: 4}
+            - {type: e3, owner: enemy, position: [72, 20], stance: 3, count: 1}
     win_condition:
       all_of:
         - {building_count_gte: {type: pbox, n: 4}}