chore(training): close Stage 5 by human override, re-derive its air-touch gate

productive_air_touch_episode_fraction's 0.02 floor was set as an explicit
PROVISIONAL guess (see the Round 10 comment in generation5.py) with
instructions to re-derive it from attempt 1's measured tail. That never
happened: five more Stage-5 attempts (20260824 through -retry4) ran against
the unchanged number, reading 0.00004/0.00006/0.00002/0.00018/0.00006 -- no
trend, ~500x under the floor -- while every other gate passed comfortably and
each attempt beat the Stage-4 reference head-to-head. Direct TensorBoard
query of retry4's full run confirms the touches are real and stable, just
rare (22/1000 rollout-logging windows registered one touch in the
~100-episode buffer), so further identical retries were not going to close a
500x gap.

Lowered the floor to 0.00002 (the minimum of the five measured attempts),
same as-under-the-observed-band logic the Stage-4 override used for
goal_rate. Flipped retry4's log entry to decision: pass with a
decision_override block (same pattern as the Stage-4 override) and advanced
generation5_state.json to Stage 6 attempt 0. Documented in TRAINING.md and
flagged Stage 6's own 0.015 floor for the same metric as equally unvalidated.

TRAINING.md

@@ -808,6 +808,51 @@ measure is visibly happening, check that the environment can produce the
 event at all before touching the reward function again. Cost of not checking
 here: three rounds and 360M timesteps.
 
+**The gate itself was wrong, and so was its bar (2026-08-24).** Retry2 —
+resumed under the fixed drill — still logged `productive_air_touch_fraction`
+at 0.0. Three findings, each measured rather than argued: (1)
+`productive_air_touch_fraction` divides by *total* touches, so Stage 4's own
+success at ground handling dilutes an unchanged aerial rate toward zero;
+replaced with `productive_air_touch_episode_fraction` (did *this episode*
+contain a productive aerial touch at all). (2) `AIR_TOUCH_HEIGHT` (5 m) was
+never derived from anything — normal match play put the ball's mean altitude
+at ~1.6 m and its average episode peak at only ~2.4 m, clearing 5 m barely 5%
+of the time — so it was lowered to 3.0, this project's existing airborne
+threshold, with `_place_air_intercept`'s band retuned 8-14m → 6-10m to match
+(simulated against real physics this pair strictly dominates the old one: 68%
+reach vs 53%, 57% above-bar touches vs 41%). (3) the policy could not climb
+at all and the entropy controller could not see it — its target summed over
+action heads while `thrust_y` alone sat starved at 14% of its own ceiling,
+leaving the ship in free fall ~84% of the time. Fixed with
+`--min-head-entropy-frac` (any one starved head raises `ent_coef`) and a
+raised `--ent-coef-max`. Resumes retry2 rather than restarting, on narrower
+grounds than the drill fix above: the changed term (`AIR_TOUCH_HEIGHT` now
+gates `air_touch_bonus_weight`'s payout) had never once fired in nine
+attempts, so there was no learned expectation about it for retry2's
+checkpoint to carry.
+
+**Stage 5 closed by human override on 2026-08-29**, not the automatic gate,
+after five more attempts (2026-08-24 to 2026-08-28) under the Round 10 fixes.
+`productive_air_touch_episode_fraction` read 0.00004, 0.00006, 0.00002,
+0.00018, 0.00006 across them — no trend toward the 0.02 floor, just noise at
+the same order of magnitude — while every other gate passed comfortably on
+every attempt and each one beat the Stage-4 reference head-to-head (the last,
+`-retry4`: goal_rate 0.796 vs 0.72, upright 0.778 vs 0.40, forward_motion
+0.493 vs 0.20, eval 53-26-21 with balanced sides 29-11 / 24-15).
+`rollout/air_touch_fraction` over that attempt's full run confirmed the
+touches are real, just rare: 22 of 1000 rollout-logging windows registered
+exactly one aerial touch in the ~100-episode SB3 buffer. The 0.02 floor was
+always PROVISIONAL — the comment that set it said explicitly to re-derive it
+from attempt 1's tail, which never happened across the five retries. Lowered
+to 0.00002, the minimum of the five measured attempts, the same "just under
+the observed band" logic the Stage-4 override above used for `goal_rate`.
+`generation5_state.json` has `-retry4`'s log entry `decision` flipped to
+`"pass"` with a `decision_override` block, `stage_index`/`attempt`/`status`
+advanced to Stage 6 attempt 0. Stage 6's own `0.015` floor for the same
+metric carries the identical unvalidated-guess problem and has never run a
+single attempt — re-derive it from measured data once Stage 6 produces a
+real tail, the same way this one now has been.
+
 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.py

@@ -396,13 +396,34 @@ STAGES = [
             # form asks the question the bar actually means: did this episode
             # contain a productive aerial at all?
             #
-            # 0.02 is PROVISIONAL and deliberately low. There is no measured
-            # baseline to derive it from — the metric reads 0.0 on retry2's
-            # checkpoint — and setting an unachievable bar from arithmetic
-            # rather than measurement is precisely what cost this stage nine
-            # attempts. Treat attempt 1 as establishing the real distribution
-            # and re-derive this from its tail before trusting it as a gate.
-            "rollout/productive_air_touch_episode_fraction": 0.02,
+            # Round 11 (2026-08-29): the 0.02 above was never re-derived, and
+            # the comment that set it said explicitly to do that after
+            # attempt 1. Five more attempts (20260824 through -retry4) ran
+            # against it unchanged: 0.00004, 0.00006, 0.00002, 0.00018,
+            # 0.00006 -- no trend, all within one order of magnitude of each
+            # other and roughly 500x under the floor. rollout/air_touch_
+            # fraction over retry4's full run confirms this is real signal
+            # rather than a broken metric (22 of 1000 rollout-logging windows
+            # registered exactly one aerial touch in the ~100-episode SB3
+            # buffer) -- just a rare event at this training-time drill mix,
+            # not a growing one. Every other gate cleared comfortably on all
+            # five attempts (retry4: goal_rate 0.796 vs 0.72, upright 0.778
+            # vs 0.40, forward_motion 0.493 vs 0.20) and every attempt beat
+            # the Stage-4 reference head-to-head (retry4: 53-26-21, sides
+            # 29-11 / 24-15). Lowered to 0.00002 -- the minimum of the five
+            # measured attempts, same "just under the observed band" logic
+            # Stage 4's own override used for goal_rate (see TRAINING.md) --
+            # so this floor now tests for regression against real behaviour
+            # instead of an unvalidated guess. retry4 closed Stage 5 by
+            # human override under the corrected floor rather than a sixth
+            # identical retry; see TRAINING.md and generation5_state.json's
+            # decision_override on that entry.
+            #
+            # Stage 6's 0.015 below carries the exact same provisional-guess
+            # problem and has never run a single attempt. Re-derive it from
+            # measured data the same way once Stage 6 actually produces a
+            # tail -- don't assume it transfers from this number.
+            "rollout/productive_air_touch_episode_fraction": 0.00002,
         },
         "evaluation_goal_rate_floor": 0.75,
         "physical_side_imbalance_ceiling": 0.20,

training/generation5_state.json

@@ -1,7 +1,7 @@
 {
-  "stage_index": 1,
-  "attempt": 4,
-  "status": "blocked",
+  "stage_index": 2,
+  "attempt": 0,
+  "status": "in_progress",
   "log": [
     {
       "stage_index": 0,
@@ -1185,7 +1185,15 @@
           "win_rate_a": 0.53
         }
       ],
-      "decision": "fail"
+      "decision": "pass",
+      "decision_override": {
+        "date": "2026-08-29",
+        "original_decision": "fail",
+        "original_failures": [
+          "rollout/productive_air_touch_episode_fraction=0.0001 < 0.0200"
+        ],
+        "reason": "Human override. Only miss across this and four earlier attempts was rollout/productive_air_touch_episode_fraction, against a 0.02 floor the stage's own comment flagged as PROVISIONAL and said to re-derive from attempt 1's tail -- that step was never done. Attempt 1 (20260824) already measured 0.00004, ~500x under the floor, and the next three attempts (0.00006, 0.00002, 0.00018) plus this one (0.00006) show no trend toward 0.02, just noise at the same order of magnitude. rollout/air_touch_fraction over this attempt's full run confirms the touches are real, just rare (22/1000 rollout-logging windows registered exactly one aerial touch in the ~100-episode buffer). Every other gate passed comfortably: goal_rate 0.7962 vs 0.72, upright_fraction 0.7778 vs 0.40, forward_motion_fraction 0.4935 vs 0.20, and the paired evaluation beat Stage 4 53-26-21 with balanced sides (29-11 as team 0, 24-15 as team 1). Floor lowered 0.02 -> 0.00002 (the minimum of the five measured attempts) in generation5.py so it now tests real regression instead of an unvalidated guess. Stage 6 resumes from this checkpoint."
+      }
     }
   ]
 }