chore(training): promote stage-4 retry2 to medium and open stage 5

20260816-2126-gen5-s4-handling-retry2 exhausted its three attempts and
missed only the 0.80 training goal-rate floor, at 0.7731. Every
evaluation gate passed: 65-22-13 versus promoted/easy.json, 87% non-draw
against an 80% floor, 12.6% physical-side imbalance against a 20%
ceiling, and both handling telemetry floors clear. The round improved the
goal rate monotonically across attempts (0.537 -> 0.683 -> 0.773) and the
checkpoint plays well by hand, so close Stage 4 by human override.

Promote it to Game/bots/promoted/medium.json. Medium and Hard both point
at the new policy: Hard stays a label-only duplicate until a stronger one
earns hard.json, which keeps the tiers monotonic rather than leaving Hard
weaker than Medium.

generation5_state.json flips that log entry to "pass" with a
decision_override block preserving the original verdict and reasoning,
and advances to Stage 5 attempt 1. This is what passing_entry() needs to
resolve Stage 5's resume checkpoint and evaluation reference, and what
league_pool() will need at Stage 6; --skip-to-next-stage would advance
the stage without marking anything as passing and die immediately.

generation5.sh now pulls before launching. Each stage ends in
commit_progress()'s push, which fails and kills the run hours in if the
box is behind origin.

Game/scripts/main_menu.gd

@@ -11,14 +11,18 @@ extends Control
 
 const BOTS_DIR := "res://bots"
 
-# All three tiers deliberately point at the same promoted checkpoint at its
-# full trained capability. The labels are placeholders until genuinely
-# stronger models are promoted as medium.json/hard.json; difficulty is not
-# simulated with reaction delay or action noise in the meantime.
+# Every tier runs its promoted checkpoint at full trained capability —
+# difficulty is a genuinely different policy, never the same policy
+# handicapped with reaction delay or action noise. Easy and Medium are now
+# distinct models (medium.json beats easy.json 65-22-13 head-to-head); Hard
+# still points at medium.json, the strongest promoted policy, and stays a
+# label-only duplicate until a stronger one earns hard.json. Keep the tiers
+# monotonic: never leave a lower tier pointing at a stronger model than the
+# one above it.
 const DIFFICULTIES := [
 	{"name": "Easy", "model": "res://bots/promoted/easy.json", "reaction_ticks": 8, "action_noise": 0.0},
-	{"name": "Medium", "model": "res://bots/promoted/easy.json", "reaction_ticks": 8, "action_noise": 0.0},
-	{"name": "Hard", "model": "res://bots/promoted/easy.json", "reaction_ticks": 8, "action_noise": 0.0},
+	{"name": "Medium", "model": "res://bots/promoted/medium.json", "reaction_ticks": 8, "action_noise": 0.0},
+	{"name": "Hard", "model": "res://bots/promoted/medium.json", "reaction_ticks": 8, "action_noise": 0.0},
 ]
 
 @onready var difficulty_dropdown: OptionButton = %DifficultyDropdown

TRAINING.md

@@ -164,7 +164,14 @@ flat files there, never touching subdirectories.
 actually referenced by the shipped game — currently `easy.json` (promoted
 2026-08-08 from generation 4's `20260806-1939-curric-s3-gauntlet`; this is
 the 320M-step MultiDiscrete policy and the foundation for the planned
-generation-5 curriculum below) and `reference-grounded.json` (added for
+generation-5 curriculum below), `medium.json` (promoted 2026-08-17 from
+generation 5's `20260816-2126-gen5-s4-handling-retry2` — Stage 4 attempt 3,
+which the pipeline recorded as a *fail* on the 80% training-goal-rate gate at
+0.7731, but which beats `easy.json` 65-22-13 in the 100-episode paired
+evaluation, 87% non-draw and 12.6% physical-side imbalance, both inside the
+Stage-4 bars; promoted by hand on gameplay feel, so its recorded
+`decision: "fail"` in `generation5_state.json` is expected and not a
+bookkeeping error) and `reference-grounded.json` (added for
 generation 4 — a copy of generation 3's `curric-s5-aggression`, made before
 the flat `Game/bots/` dump was scrapped for the redesign, kept as the
 strongest grounded-era artifact and the fixed yardstick generations 1-3 were
@@ -174,10 +181,12 @@ promoted file is never touched by training scripts, never overwritten by a
 same-named future export, and never disturbed by pruning old experiment
 files from the flat dump.
 
-Until distinct `medium.json` and `hard.json` policies earn promotion, the
-three menu tiers all run this same `easy.json` policy at its full trained
-cadence (`reaction_ticks=8`, `action_noise=0`). The tiers are labels only;
-the game does not manufacture difficulty gaps by handicapping this model.
+Easy and Medium are now genuinely different policies. Hard still points at
+`medium.json` and remains a label-only duplicate until a stronger policy earns
+`hard.json`. Every tier runs at full trained cadence (`reaction_ticks=8`,
+`action_noise=0`) — the game does not manufacture difficulty gaps by
+handicapping a model. When promoting, keep the tiers monotonic: a lower tier
+must never point at a policy that beats the tier above it.
 
 To promote a new bot into a tier: copy the chosen `Game/bots/<experiment>.json`
 to `Game/bots/promoted/<tier>.json` (overwriting the old one), and note the
@@ -609,6 +618,27 @@ that fades to zero by 3m altitude. Its scoring gates deliberately run before
 Stage 5: becoming upright is not progress if the resulting policy stops
 finishing goals. Stage 5 adds moving high-ball intercept
 starts aimed toward real goals rather than a standalone air-touch reward.
+
+**Stage 4 was closed by human override on 2026-08-17**, not by the automatic
+gate. `20260816-2126-gen5-s4-handling-retry2` exhausted all three attempts and
+the pipeline recorded `decision: "fail"`, on the training goal-rate floor alone
+(0.7731 vs 0.80). Every other gate passed — 65-22-13 versus
+`promoted/easy.json`, 87% non-draw against the 80% floor, 12.6% physical-side
+imbalance against the 20% ceiling, and both handling telemetry floors clear
+(`upright_fraction` 0.772 vs 0.45, `forward_motion_fraction` 0.315 vs 0.25) —
+and the round's three attempts improved the training goal rate monotonically
+(0.537 → 0.683 → 0.773). The same checkpoint also plays well enough by hand to
+have been promoted to `medium.json`. `generation5_state.json` therefore has
+that log entry's `decision` flipped to `"pass"` with a `decision_override`
+block recording the original verdict and reasoning, and `stage_index`/
+`attempt`/`status` advanced to Stage 5 attempt 1 — which is what
+`passing_entry()` needs to resolve Stage 5's resume checkpoint and evaluation
+reference, and what `league_pool()` will later need at Stage 6. Prefer this
+edit over `--skip-to-next-stage`: that flag advances `stage_index` without
+marking anything as passing, so the run dies immediately with `RuntimeError:
+No passing generation-5 stage index 0`. Note that retry2 already clears Stage
+5's own 0.75 goal-rate floor; the 0.80 Stage-4 figure was always the stricter
+of the two.
 Stage 6's `league` opponent mode samples a historical exported policy at each
 episode reset. Each later stage preserves the preceding shaping and adds one
 new difficulty.

training/generation5.sh

@@ -14,6 +14,13 @@ if tmux has-session -t "$SESSION" 2>/dev/null; then
     exec tmux attach -t "$SESSION"
 fi
 
+# Stage start and resume point both come from committed state, and each stage
+# ends in commit_progress()'s `git push` — which fails, killing the run after
+# hours of training, if the box is behind origin. Pull before starting, same as
+# next_run.sh. Aborts here (set -e) on a dirty tree or a conflict, which is the
+# point: fix it before spending the compute, not after.
+git pull --rebase
+
 tmux new-session -d -s "$SESSION" -n curriculum \
     ".venv/bin/python generation5.py $*; echo; echo '=== generation5.py exited — press Enter to close ==='; read"
 

training/generation5_state.json

@@ -1,7 +1,7 @@
 {
-  "stage_index": 0,
-  "attempt": 2,
-  "status": "blocked",
+  "stage_index": 1,
+  "attempt": 0,
+  "status": "in_progress",
   "log": [
     {
       "stage_index": 0,
@@ -249,7 +249,15 @@
           "win_rate_a": 0.65
         }
       ],
-      "decision": "fail"
+      "decision": "pass",
+      "decision_override": {
+        "date": "2026-08-17",
+        "original_decision": "fail",
+        "original_failures": [
+          "rollout/goal_rate=0.7731 < 0.8000"
+        ],
+        "reason": "Human override. Only miss was the training goal-rate floor, at 0.7731 vs 0.80; every evaluation gate passed (65-22-13 vs promoted/easy.json, 87% non-draw, 12.6% physical-side imbalance) and the goal rate improved monotonically across all three attempts (0.537 -> 0.683 -> 0.773). Also promoted to Game/bots/promoted/medium.json on the same evidence. Stage 5 resumes from this checkpoint."
+      }
     }
   ]
 }