chore(training): add air_approach_weight and restart stage-5 intercepts

Stage 5 blocked all three attempts on productive_air_touch_fraction
stuck exactly at 0.0 across a continuous 180M-step lineage, while
goal_rate/upright_fraction/forward_motion_fraction kept improving on
the same budget. forward_velocity_to_ball_weight (the term that solved
Stage 4's ground pursuit) is hard-gated below GROUND_HANDLING_HEIGHT
and does nothing in the air, so Stage 5's air_intercept_chance had no
matching aerial incentive to learn from. air_approach_weight adds the
airborne mirror (nose-first 3D closing speed, no uprightness
multiplier) and folds into HANDLING_REWARD_FLAGS so Stage 6 inherits
it too. Deleted the three blocked attempts and reset state to resume
Stage 5 from the Stage-4 checkpoint with the new term.

training/generation5.py

@@ -151,9 +151,40 @@ STANDING_ARGS = ["--ent-coef", "0.01", "--entropy-floor"]
 # floor, and air-drill-chance goes to 0. The reward terms already built
 # are left exactly as they were — they should finally pull in a direction
 # the ship can go.
+#
+# Round 7 (2026-08-18): Stage 4 closed by human override (see TRAINING.md).
+# Stage 5 (intercepts) then blocked all three attempts on the same single
+# floor every time — rollout/productive_air_touch_fraction stayed exactly
+# 0.0 across a continuous 180M-step lineage (each retry resumes the
+# previous attempt's checkpoint, not a fresh run), while air_touch_fraction
+# sat at noise level (0.00008 -> 0.00006 -> 0.00006) and goal_rate/
+# upright_fraction/forward_motion_fraction all kept improving on the same
+# budget. A dead-flat metric across that much continued training, next to
+# metrics that keep moving, is the missing-mechanism signature from Round 6
+# again, not a slow-learning one: forward_velocity_to_ball_weight -- the
+# term that actually solved ground handling -- is hard-gated to
+# ship.global_position.y < GROUND_HANDLING_HEIGHT and does nothing in the
+# air, so air_intercept_chance (added for Stage 5) was asking for aerial
+# pursuit with only the generic, orientation-agnostic velocity_to_ball_
+# weight (0.04) to learn it from -- the same class of gap as Stage 4's
+# missing ground-tilt/non-forward pressure before those were added.
+#
+# air_approach_weight (ship_ai_controller.gd) is the airborne mirror:
+# nose-first 3D closing speed on the ball, active above
+# GROUND_HANDLING_HEIGHT instead of below it (mutually exclusive with
+# forward_velocity_to_ball_weight by altitude), with no uprightness
+# multiplier since a real aerial requires pitching away from level. Set to
+# 0.15 to match forward_velocity_to_ball_weight's proven-effective
+# magnitude; added to HANDLING_REWARD_FLAGS (not just Stage 5's flags) so
+# it also carries into Stage 6, which reuses these flags and its own
+# air_intercept_chance. Stage 5 restarts from Stage 4's checkpoint rather
+# than continuing retry2's, same reasoning as every previous mechanism
+# change in this file: don't resume a policy shaped by an absent term into
+# one where it now exists.
 HANDLING_REWARD_FLAGS = [
     "--velocity-to-ball-weight", "0.04",
     "--forward-velocity-to-ball-weight", "0.15",
+    "--air-approach-weight", "0.15",
     "--ball-distance-penalty", "0.01",
     "--ball-touch-reward", "0.7",
     "--ball-velocity-to-goal-weight", "0.06",

training/generation5_state.json

@@ -1,7 +1,7 @@
 {
   "stage_index": 1,
-  "attempt": 2,
-  "status": "blocked",
+  "attempt": 0,
+  "status": "in_progress",
   "log": [
     {
       "stage_index": 0,
@@ -258,254 +258,6 @@
         ],
         "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."
       }
-    },
-    {
-      "stage_index": 1,
-      "stage_number": 5,
-      "stage_name": "intercepts",
-      "experiment": "20260817-0750-gen5-s5-intercepts",
-      "attempt": 0,
-      "telemetry_tail": {
-        "rollout/air_touch_fraction": 7.999999821186065e-05,
-        "rollout/airborne_fraction": 0.24998261865973473,
-        "rollout/ep_len_mean": 106.38228009033203,
-        "rollout/ep_rew_mean": 10.283133600234985,
-        "rollout/forward_motion_fraction": 0.3337678092420101,
-        "rollout/goal_rate": 0.7703999981880189,
-        "rollout/grounded_upright_fraction": 0.24178000067174435,
-        "rollout/mean_altitude": 3.19321390914917,
-        "rollout/productive_air_touch_fraction": 0.0,
-        "rollout/upright_fraction": 0.7911632860898972,
-        "rollout/vertical_thrust_mean": 0.03387699818262831
-      },
-      "telemetry_failures": [
-        "rollout/productive_air_touch_fraction=0.0000 < 0.0050"
-      ],
-      "evaluation_goal_failures": [],
-      "side_balance_failures": [
-        "20260816-2126-gen5-s4-handling-retry2.json: physical_side_imbalance=0.290 > 0.200"
-      ],
-      "eval": {
-        "timestamp": "2026-08-17T16:01:22+00:00",
-        "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260817-0750-gen5-s5-intercepts.json",
-        "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-        "seed": 1,
-        "episodes": 100,
-        "wins_a": 51,
-        "wins_b": 32,
-        "draws": 17,
-        "side_results": {
-          "a_team_0": {
-            "wins_a": 18,
-            "wins_b": 23,
-            "draws": 9
-          },
-          "a_team_1": {
-            "wins_a": 33,
-            "wins_b": 9,
-            "draws": 8
-          }
-        },
-        "physical_team_wins": {
-          "team_0": 27,
-          "team_1": 56
-        },
-        "win_rate_a": 0.51
-      },
-      "evals": [
-        {
-          "timestamp": "2026-08-17T16:01:22+00:00",
-          "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260817-0750-gen5-s5-intercepts.json",
-          "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-          "seed": 1,
-          "episodes": 100,
-          "wins_a": 51,
-          "wins_b": 32,
-          "draws": 17,
-          "side_results": {
-            "a_team_0": {
-              "wins_a": 18,
-              "wins_b": 23,
-              "draws": 9
-            },
-            "a_team_1": {
-              "wins_a": 33,
-              "wins_b": 9,
-              "draws": 8
-            }
-          },
-          "physical_team_wins": {
-            "team_0": 27,
-            "team_1": 56
-          },
-          "win_rate_a": 0.51
-        }
-      ],
-      "decision": "fail"
-    },
-    {
-      "stage_index": 1,
-      "stage_number": 5,
-      "stage_name": "intercepts",
-      "experiment": "20260817-1701-gen5-s5-intercepts-retry1",
-      "attempt": 1,
-      "telemetry_tail": {
-        "rollout/air_touch_fraction": 5.9999998658895494e-05,
-        "rollout/airborne_fraction": 0.25157142809033395,
-        "rollout/ep_len_mean": 95.91548010253906,
-        "rollout/ep_rew_mean": 10.333840370178223,
-        "rollout/forward_motion_fraction": 0.3461968092918396,
-        "rollout/goal_rate": 0.8227999938726425,
-        "rollout/grounded_upright_fraction": 0.20529999974370003,
-        "rollout/mean_altitude": 3.198590657234192,
-        "rollout/productive_air_touch_fraction": 0.0,
-        "rollout/upright_fraction": 0.7962958091497421,
-        "rollout/vertical_thrust_mean": 0.053312998113571666
-      },
-      "telemetry_failures": [
-        "rollout/productive_air_touch_fraction=0.0000 < 0.0050"
-      ],
-      "evaluation_goal_failures": [],
-      "side_balance_failures": [],
-      "eval": {
-        "timestamp": "2026-08-18T01:10:15+00:00",
-        "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260817-1701-gen5-s5-intercepts-retry1.json",
-        "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-        "seed": 1,
-        "episodes": 100,
-        "wins_a": 58,
-        "wins_b": 30,
-        "draws": 12,
-        "side_results": {
-          "a_team_0": {
-            "wins_a": 26,
-            "wins_b": 20,
-            "draws": 4
-          },
-          "a_team_1": {
-            "wins_a": 32,
-            "wins_b": 10,
-            "draws": 8
-          }
-        },
-        "physical_team_wins": {
-          "team_0": 36,
-          "team_1": 52
-        },
-        "win_rate_a": 0.58
-      },
-      "evals": [
-        {
-          "timestamp": "2026-08-18T01:10:15+00:00",
-          "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260817-1701-gen5-s5-intercepts-retry1.json",
-          "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-          "seed": 1,
-          "episodes": 100,
-          "wins_a": 58,
-          "wins_b": 30,
-          "draws": 12,
-          "side_results": {
-            "a_team_0": {
-              "wins_a": 26,
-              "wins_b": 20,
-              "draws": 4
-            },
-            "a_team_1": {
-              "wins_a": 32,
-              "wins_b": 10,
-              "draws": 8
-            }
-          },
-          "physical_team_wins": {
-            "team_0": 36,
-            "team_1": 52
-          },
-          "win_rate_a": 0.58
-        }
-      ],
-      "decision": "fail"
-    },
-    {
-      "stage_index": 1,
-      "stage_number": 5,
-      "stage_name": "intercepts",
-      "experiment": "20260818-0210-gen5-s5-intercepts-retry2",
-      "attempt": 2,
-      "telemetry_tail": {
-        "rollout/air_touch_fraction": 5.9999998658895494e-05,
-        "rollout/airborne_fraction": 0.24753280937671662,
-        "rollout/ep_len_mean": 103.67796003723144,
-        "rollout/ep_rew_mean": 11.149126368522644,
-        "rollout/forward_motion_fraction": 0.363580904841423,
-        "rollout/goal_rate": 0.7755999964475632,
-        "rollout/grounded_upright_fraction": 0.23967999944090843,
-        "rollout/mean_altitude": 3.170829694747925,
-        "rollout/productive_air_touch_fraction": 0.0,
-        "rollout/upright_fraction": 0.800830381155014,
-        "rollout/vertical_thrust_mean": 0.024330998776718044
-      },
-      "telemetry_failures": [
-        "rollout/productive_air_touch_fraction=0.0000 < 0.0050"
-      ],
-      "evaluation_goal_failures": [],
-      "side_balance_failures": [],
-      "eval": {
-        "timestamp": "2026-08-18T10:17:39+00:00",
-        "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260818-0210-gen5-s5-intercepts-retry2.json",
-        "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-        "seed": 1,
-        "episodes": 100,
-        "wins_a": 53,
-        "wins_b": 33,
-        "draws": 14,
-        "side_results": {
-          "a_team_0": {
-            "wins_a": 23,
-            "wins_b": 20,
-            "draws": 7
-          },
-          "a_team_1": {
-            "wins_a": 30,
-            "wins_b": 13,
-            "draws": 7
-          }
-        },
-        "physical_team_wins": {
-          "team_0": 36,
-          "team_1": 50
-        },
-        "win_rate_a": 0.53
-      },
-      "evals": [
-        {
-          "timestamp": "2026-08-18T10:17:39+00:00",
-          "model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260818-0210-gen5-s5-intercepts-retry2.json",
-          "model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
-          "seed": 1,
-          "episodes": 100,
-          "wins_a": 53,
-          "wins_b": 33,
-          "draws": 14,
-          "side_results": {
-            "a_team_0": {
-              "wins_a": 23,
-              "wins_b": 20,
-              "draws": 7
-            },
-            "a_team_1": {
-              "wins_a": 30,
-              "wins_b": 13,
-              "draws": 7
-            }
-          },
-          "physical_team_wins": {
-            "team_0": 36,
-            "team_1": 50
-          },
-          "win_rate_a": 0.53
-        }
-      ],
-      "decision": "fail"
     }
   ]
 }

training/train.py

@@ -355,6 +355,11 @@ def parse_args():
         "--forward-velocity-to-ball-weight", type=float, default=None,
         help="Low-altitude dense reward for nose-led planar approach toward the ball",
     )
+    curriculum.add_argument(
+        "--air-approach-weight", type=float, default=None,
+        help="Aerial mirror of forward-velocity-to-ball-weight: high-altitude dense reward for "
+        "nose-led 3D closing speed toward the ball",
+    )
     curriculum.add_argument(
         "--ball-distance-penalty", type=float, default=None,
         help="Overrides ShipAIController.ball_distance_penalty (dense per-tick cost scaled by distance to the ball)",
@@ -419,6 +424,7 @@ def _curriculum_kwargs(args) -> dict:
         "ai_grounded_upright_reward": args.grounded_upright_reward,
         "ai_velocity_to_ball_weight": args.velocity_to_ball_weight,
         "ai_forward_velocity_to_ball_weight": args.forward_velocity_to_ball_weight,
+        "ai_air_approach_weight": args.air_approach_weight,
         "ai_ball_distance_penalty": args.ball_distance_penalty,
         "ai_ball_touch_reward": args.ball_touch_reward,
         "ai_airborne_penalty": args.airborne_penalty,