Board: NEW FRONTIER ft_vmin7r_wnr20 (-71.8, 3.05 sigma, fully clean); entire vrel family quarantined as appetite-in-disguise

agent_reports_bgkf/FRONTIER.md

@@ -2,22 +2,29 @@
 
 Updated: 2026-08-22 ~21:40. Baseline: champion logbag/base = 1760.2 (swaps 9.5).
 
-## Standings (paired on 48 origin-sums vs 1760.2; * = >= 2 sigma and count-clean-pending-checks)
-- md_bl_hgbd1l25_vrel  -117.8 +/- 46.1 (2.6s, 30W/18L, swaps 9.2)  * current leader
-- md_hgb_d1_vrel       -119.5 +/- 63.5 (1.9s)                       high variance twin
-- bl_v_wnr20           -100.9 +/- 41.4 (2.4s, 36W/12L, swaps 9.3)  * vrel + wiener .20
-- ft_triple_rtau        -96.5 +/- 37.5 (2.6s, 34W/13L, swaps 9.0)  *
-- ft_vrel_rstale        -91.4 +/- 38.0 (2.4s, 37W/11L)             *
-- bl_wnr20 (champion+wiener.20)  -50.3 +/- 19.7 (2.55s, count-clean corr -0.005) VALIDATED PIECE
-- vrel alone            -66.8 +/- 41.4 -- SEE CAUTION BELOW
+## Standings (paired on 48 origin-sums vs 1760.2)
+- **ft_vmin7r_wnr20  -71.8 +/- 23.5 (3.05 sigma, wilcoxon p=.037, swaps 9.58)
+  = NEW FRONTIER, the campaign's one fully-clean survivor.**
+  Composition: FEATURES + vmin7_rank (fleet pct-rank of 7d-min voltage,
+  already in the table) with the wiener .20 value-preserving order blend.
+  Count-clean by every control (corr +0.048; count-flat scenarios carry the
+  full win; anchored-ordering verified; components additive: wiener -50.3 +
+  vmin7r -23.2 ~= -73.5 predicted vs -71.8 measured).
+  To compose on it: cv.VARIANTS['vm7'] = lambda t: cv.BASE_FEATURES +
+  ['vmin7_rank']; then cv.run_bgkf(variant='vm7', blend=('wiener', 0.20)).
+  A second order source needs a 3-way wrapper (wiener slot is occupied).
+- Clean pieces: wiener .20 blend (-50.3 champion base); vmin7_rank feature
+  (-23.2 solo, -21.5 +/- 12.3 marginal ON the wiener base).
 
-## CAUTION - vrel is count-entangled (bl lane finding, needs ft adjudication)
-corr(count delta, cost delta) = -0.544 for vrel; on count-flat scenarios vrel
-reads +36.4 (WORSE). Its -66.8 may be a hidden appetite channel (the exact
-phase-56 frontier failure mode). EVERY vrel-based leader inherits this doubt
-until the count-matched (value-anchored) vrel twin is adjudicated - assigned
-to the ft lane. The WIENER increment is independently clean (additive
--34..-50 on every base with corr -0.005).
+## QUARANTINED - the entire vrel family is appetite-in-disguise (ft lane
+value-anchored adjudication, matches bl's corr finding exactly):
+vrel anchored twin +21.4 +/- 23.9 (raw -66.8 collapses); triple anchored
++37.9; within-test-fleet recomputation flips to +89 (the "win" needed the
+full 24-building shared reference - not computable in deployment).
+ALL vrel-based standings rows are quarantined: md_bl_hgbd1l25_vrel,
+md_hgb_d1_vrel, bl_v_wnr20, ft_triple_*, ft_vrel_* - pending each one's own
+anchored twin, assume the vrel component contributes nothing transferable.
+Also dead: z-scores (weaker vrel), building relatives, dynamics ranks.
 
 ## Watch list (unconfirmed leads)
 - wd_aux_all (aux usage-dynamics features rng_90/vstd_30/rng_ratio/gap_max_60