Wave A10: head is at its local optimum on every dial; wiener .25 sub-bar trend shelved

agent_reports_bgkf/c3_findings.md

@@ -225,3 +225,14 @@ alpha 1.0 best everywhere). Cost is dominated by late/early, not travel:
 the scaling verdict is geography-insensitive. Stopped early per user (two
 k=4 alpha arms unfinished - pattern already unambiguous). Artifacts:
 /tmp/c5/travel_map.py, travel_cluster.py, scalem_*/scalec_* CSVs.
+
+## Wave A10 (post-ship re-tuning on the alpha-1.0 base, 1623.8)
+
+Autopsy at the new operating point: recall 53% (+5pts, exactly the
+scaling-study prediction); near-miss wall unchanged (188 deaths / 3,056
+late-days at gap 1-10); silent still marginal. Dial sweep paired vs the
+current model: wiener .25/.30 = -22.7±14.2 / -25.8±24.6 (sub-bar trend,
+SHELF - candidate for a spare public trigger); alpha bowl bottoms at 1.0
+(1.05/1.10 sign-negative); day_q exactly 0; ec top-up shipped values
+optimal (off +23, wider +11); fixedref +3.3 sign-worse (stays a dial).
+VERDICT: the head is at its local optimum; no further clean gain today.