Campaign 3 CLOSED: ~45 candidates, frontier confirmed as pipeline ceiling; autopsy prize unreachable by ordering; alpha deflated (sign-test refusal); xlb micro-piece shelved; team-repo intel (Phase 60)

EXPERIMENTS.md

@@ -3249,3 +3249,36 @@ Private-selection matrix as of this phase: candidate head = 3f9bf12 (FT2
 frontier); fallback = 3c7b883 (measured 1661.45); never 698401a (1877.08).
 The user manually triggers public scoring; a public look at 3f9bf12 prices
 the residual ghost/composition risk the CV cannot see (phase 58).
+
+## Phase 60 - campaign 3: exhaustive post-frontier exploration; frontier confirmed as the pipeline ceiling
+
+User directive: improve the champion CV further, radical options included,
+review all assumptions (podium context: team-repo scale ~1400). ~45
+candidates over 7 waves + 2 device-level autopsies + full mining of the
+team repo (rul-mgt-zero, 1493.65 public vs our 1661.45). Full detail:
+agent_reports_bgkf/c3_findings.md.
+
+KEY FINDINGS:
+1. AUTOPSY overturns the silent-death assumption: frontier recall is 48%;
+   237 missed deaths at median rank 5 (gap quartiles 2/3/8 below a ~rank-2
+   cutoff) carry 70% of remaining cost; only 45/237 are voltage-silent.
+   False includes are plateau devices at 2.41-2.44V for 40+ days (36% die,
+   median 83d — just outside the window). The prize is top-band ordering.
+2. No candidate reaches it: every feature/order/model/count axis measured;
+   best survivor is a -6-point crossing-label micro-piece (xlb, Wilcoxon
+   .0475, shelved). ft2 (1688.4, 3f9bf12) is the ceiling of this pipeline
+   under the count-clean bar.
+3. Appetite re-audited and re-closed: alpha 1.0 reads -64.5 mean but
+   17W/24L, Wilcoxon .375 — warmup-carried count redistribution; gap-band
+   precision ~11% < ~16% break-even.
+4. Team repo decoded: their edge = multi-level self-supervised targets +
+   poskeep_target sampling (-155.6, not portable — we already train on all
+   rows) + CP-SAT planner over all nine cost components. Their CV is
+   in-fleet (1412.6 not comparable to our held-out 1688.4). Direct feature
+   transplant fails in our model form (+15..+167). Cross-pollination
+   candidates TO them: WienerFHT order blend, vmin7_rank, building-held-out
+   transfer check; their gen-10 state-dependent dV/dT (resistance channel)
+   is the strongest open lead anywhere.
+
+STANDING RECOMMENDATION unchanged: public-trigger 3f9bf12/head; private =
+3f9bf12 unless publicly refuted, else 3c7b883; never 698401a.

agent_reports_bgkf/c3_findings.md

@@ -81,6 +81,48 @@ measures the SHAPE of the approach.
   vmin7_rank, and a building-held-out transfer check (their harness cannot
   see cross-fleet failure; ours was built for it).
 
-## Wave X (in flight): crossing-label order sources (self-supervised
-## x25-in-42d, logbag + HGB forms), rttt_accel, raw-trajectory HGB,
-## cost-weighted order. Results appended on completion.
+## Wave X results (closes the radical lanes)
+
+| candidate | vs ft2 | verdict |
+|---|---|---|
+| xlb (crossing-label logistic, w=.15-.20 in order) | -5.8 ± 14.3, 15W/3L, Wilcoxon .0475 | REAL micro-piece, see below |
+| xhgb / xhgb35 (crossing-label HGB) | +39 / +285 | dead; capacity hurts cross-fleet |
+| acc (rttt_accel knee compounding) | +66 | dead in our form |
+| raw / rawo (raw-trajectory HGB) | +136..+167 | dead |
+| costo (cost-weighted logbag order) | +25 | dead |
+
+xlb bowl: w .10 -4.5 (.074) / .15 -5.8 (.0475) / .20 -6.1 (.0505) /
+.30 +23. A genuine, sign-consistent ~-6-point effect from self-supervised
+crossing labels (the one team-repo structural idea that survived
+transplant) — but 0.4 sigma on the mean, tie-dominated. SHELF: not worth a
+new deployment code path on deadline day; the recipe is
+XrossOrder in /tmp/c3/run6.py (labels = 7d-median crosses 2.50 within 42d;
+logistic on vm7 features at weight .15 inside the order blend).
+
+## ALPHA DEFLATED (correction to the audit section above)
+
+Scrutiny kills the appetite lever even locally: alpha 1.0 is 17W/24L,
+Wilcoxon 0.375 — most scenarios get WORSE; the -64.5 mean is a few
+warmup-heavy saves (count-redistribution, the phase-56 signature). The
+gap-band precision ~11% sits below the ~16% break-even, exactly as the
+phase-58-era memory recorded. Appetite stays CLOSED on local evidence too.
+
+## CAMPAIGN 3 CONCLUSION (~45 candidates, 2 autopsies, team-repo mining)
+
+**The frontier stands: ft2 = 1688.4 (commit 3f9bf12) is the ceiling of
+this pipeline under the count-clean bar.** Every axis is now measured:
+composition/weights/windows (plateau at shipped values), feature space
+(level, shape, rank, aux, physics, contagion, team-port), order sources
+(RankNet, HGB, dthaz, gbm, sim, raw-trajectory, cost-weighted,
+crossing-label), sampling (tr lane), planner (pl lane + day_q 0.0),
+appetite (deflated). The near-miss prize the autopsy exposed (211 deaths
+at rank-gap 1-10, 3,593 late-days) is real but unreachable by any
+ordering candidate tested — the false includes are genuinely
+harder-looking devices; separating plateau-at-threshold from fast-crossers
+needs information this fleet's 82 labels cannot teach cross-building.
+
+**The podium path is the TEAM repo** (1493.65 public vs our 1661.45):
+its multi-level supervision + CP-SAT planner is a structurally better
+pipeline. Recommended relays: our WienerFHT order blend, vmin7_rank, a
+building-held-out transfer check, and their own gen-10 resistance channel
+(state-dependent dV/dT) as the strongest open lead there.