Board: aux_ho survives anchored twin (-59.7, real ordering core); vm7+aux_ho+wnr20 -126.4 nominal lead; radical lane closed no survivor

agent_reports_bgkf/FRONTIER.md

@@ -26,13 +26,21 @@ 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
-  + hot/oob): -80.2 +/- 40.8 on the champion base - an instrument INVERSION
-  (these were +124 dead on the purge). wd lane isolating which aux feature
-  carries it; if it survives, compose with vrel/wiener.
-- Radical standalone sim pipeline: +636.8 @ swaps 21.3 (count-exploded) -
-  first radical arm negative, hybrids pending.
+## Watch list -> ADJUDICATED (wd lane, restored from pre-outage records;
+## deterministic re-verification auto-runs when the baseline CSV reappears)
+- aux_all isolates to HOT_90 + OOB_90 ("aux_ho"): raw -114.7 +/- 41.1
+  (31W/17L, swaps 9.6, stable across bag sizes). ANCHORED TWIN HOLDS:
+  -59.7 +/- 35.9, 29W/9L (contrast vrel's collapse to +21.4) - a real
+  ordering core; ~half the raw effect is value-channel (corr -0.23, mild).
+- Compositions: vm7+aux_ho+wiener.20 = -126.4 +/- 43.8 (31W/17L, swaps 9.6)
+  - nominally BELOW the ft frontier; head-to-head vs ft_vmin7r_wnr20
+  -54.6 +/- 45.7 (n.s., favorable). Anchored composition -58.5 +/- 36.7.
+  ADOPTION FORM OPEN: raw (includes value channel) vs anchored (ordering
+  only, shipping-bar conservative) - decided at consolidation.
+- RADICAL LANE CLOSED, no survivor: sim standalone +604 at MATCHED count
+  (physics loses to the 82 labels cross-fleet, final - artifacts destroyed,
+  not worth rebuilding); EC planner = count knob (agrees with pl closure);
+  direct death-day rejected; sim order loses the wiener slot (+18.8).
 
 ## Validated composable pieces
 - wiener FHT order blend w=0.20: additive on champion (-50.3), vrel (-34.1

agent_reports_bgkf/md_findings.md

@@ -1,21 +1,67 @@
 # md — Model families & hyperparameters (building-GroupKFold)
 
-STATUS: running. Baseline logbag SCORE 1760.2, swaps 9.5. HGB+monotone axis in progress.
+STATUS: resumed after outage (deadline-day morning). All pre-crash reads below are
+valid — harness is deterministic, deltas are vs the same baseline recipe (1760.2,
+swaps 9.50). Remaining queue: d2 l2-bowl completion (l2=2, l2=10), value-anchored
+twin of bowl winner, 3-way order source on vm7 frontier base.
+
+## Results (base variant, paired vs /tmp/cv_bgkf.csv, 48 origin-sums)
 
 | label | SCORE | delta | SE | W/L | swaps |
 |---|---|---|---|---|---|
-| md_hgb_d2 (mono, depth2, l2=1) | 1673.7 | -86.4 | 72.7 | 28/20 | 9.25 |
-| md_hgb_l25 (mono, l2=5) | 1691.8 | -68.3 | 69.2 | 26/22 | 9.40 |
-| md_hgbmono_l2 (mono, l2=1) | 1715.1 | -45.1 | 70.2 | 27/21 | 9.35 |
+| md_hgb_d2 (mono HGB, depth2, l2=1) | 1673.7 | -86.4 | 72.7 | 28/20 | 9.25 |
+| md_hgb_d1 (mono stumps, lr.05x400, l2=1) | 1681.2 | -79.0 | 54.0 | 28/20 | 9.52 |
+| md_bl_hgbd1 (mean logbag+d1, w=.5) | 1684.5 | -75.7 | 39.0 | 27/21 | 9.65 |
+| md_bl_hgbd1l25 (mean logbag+d1l25, w=.5) | 1686.1 | -74.0 | 41.6 | 28/20 | 9.69 |
+| md_hgb_d1l25 (mono stumps, l2=5) | 1687.7 | -72.4 | 57.2 | 29/19 | 9.58 |
+| md_hgb_l25 (mono, depth3, l2=5) | 1691.8 | -68.3 | 69.2 | 26/22 | 9.40 |
+| md_hgb_d1f (mono stumps, lr.1x200) | 1695.1 | -65.1 | 60.8 | 23/25 | 9.77 |
+| md_bl_hgbd1_w3 (mean blend w=.3) | 1702.0 | -58.1 | 38.4 | 27/18 | 9.52 |
+| md_hgb_d2s (mono d2, lr.05x400) | 1704.6 | -55.5 | 72.2 | 29/19 | 9.50 |
+| md_hgb_d2l25 (mono d2, l2=5) | 1710.3 | -49.9 | 66.1 | 27/21 | 9.52 |
+| md_hgbmono_l2 (mono, depth3, l2=1) | 1715.1 | -45.1 | 70.2 | 27/21 | 9.35 |
 | md_hgb_leaf50 (mono, leaf50, l2=1) | 1716.1 | -44.1 | 71.4 | 25/23 | 9.44 |
-| md_hgb_l2s (mono, l2=1, lr.05 x400) | 1726.2 | -34.0 | 69.9 | 23/25 | 9.46 |
+| md_blcm_hgbd1 (blend, count-matched) | 1719.7 | -40.5 | 31.2 | 23/22 | 9.65 |
+| md_hgb_l2s (mono, l2=1, lr.05x400) | 1726.2 | -34.0 | 69.9 | 23/25 | 9.46 |
+| md_rk_hgbd1_w5 (rank-only twin, w=.5) | 1744.4 | -15.8 | 33.0 | 16/25 | 9.71 |
+| md_robust (logbag + RobustScaler) | 1747.8 | -12.4 | 13.8 | 24/8 | 9.46 |
+| md_impmean (logbag, mean imputer) | 1747.7 | -12.4 | 11.5 | 16/5 | 9.50 |
+| md_logbag_c03 (C=0.3) | 1750.5 | -9.7 | 8.5 | 23/9 | 9.48 |
+| md_enet (saga elasticnet l1r=.5) | 1756.5 | -3.7 | 3.4 | 15/5 | 9.48 |
 | logbag (baseline) | 1760.2 | 0.0 | - | - | 9.50 |
-| md_hgbmono_slow (mono, lr.05 x400) | 1768.6 | +8.5 | 73.6 | 23/25 | 9.38 |
+| md_l1 (saga L1, C=1) | 1766.7 | +6.6 | 10.4 | 13/6 | 9.50 |
+| md_hgbmono_slow (mono, lr.05x400) | 1768.6 | +8.5 | 73.6 | 23/25 | 9.38 |
+| md_logbag_c3 (C=3) | 1768.8 | +8.6 | 10.2 | 7/9 | 9.50 |
+| md_logbag_b30 (bags=30) | 1781.2 | +21.0 | 13.7 | 11/12 | 9.50 |
+| md_logbag_b60 (bags=60) | 1783.2 | +23.0 | 13.7 | 13/16 | 9.52 |
 | md_hgbmono (mono, gbm defaults) | 1788.6 | +28.4 | 71.0 | 22/26 | 9.54 |
 | md_gbm (unconstrained control) | 1809.5 | +49.3 | 85.7 | 24/24 | 9.94 |
 
-- gbm control reproduces prior read exactly (+49). Monotone constraints alone: +49 -> +28.
-- Monotone + regularization is a clean monotone trend: l2=1 -45, l2=5 -68, depth2+l2=1 -86.
-- None past 2*SE yet; swaps drift slightly low (9.25-9.45) — count-coupling under watch.
+## vrel-family runs (base QUARANTINED by coordinator — historical record only)
+
+| label | SCORE | delta | SE | W/L | swaps | vs vrel-logbag |
+|---|---|---|---|---|---|---|
+| md_hgb_d1_vrel | 1640.7 | -119.5 | 63.5 | 26/22 | 9.15 | -52.7 ± 53.3 |
+| md_bl_hgbd1l25_vrel | 1642.4 | -117.8 | 46.1 | 30/18 | 9.17 | -51.0 ± 34.2 |
+| md_bl_hgbd1_vrel | 1667.2 | -93.0 | 45.5 | 30/18 | 9.19 | -26.2 ± 34.3 |
+| md_blcm_hgbd1_vrel (count-matched) | 1667.6 | -92.6 | 46.8 | 31/17 | 9.29 | ~-25.8 |
+| md_logbag_vrel (reference) | 1693.4 | -66.8 | 41.4 | 32/15 | 9.29 | 0 |
+
+## One-liners
+- gbm control reproduces prior read exactly (+49.3). Monotone constraints alone: +49 -> +28.
+- Constrained + regularized is a clean monotone trend down: depth3 l2=1 -45, l2=5 -68;
+  depth2 l2=1 -86.4 (best solo); stumps l2=1 -79.0 ± 54 with PERFECTLY flat swaps 9.52.
+- Depth2 bowl so far: l2=1 -86.4, l2=5 -49.9 — l2=2 and l2=10 pending to bracket.
+- Mean blend logbag+stumps halves the SE: -75.7 ± 39.0 (1.94 sigma), swaps 9.65.
+- QUARANTINE checks (base): rank-only twin keeps -15.8 ± 33.0 of -75.7 (ordering is a
+  minority share); count-matched twin keeps -40.5 ± 31.2 with swaps unchanged at 9.65
+  (win is NOT sum(p42)-count inflation — the value channel is per-device magnitudes
+  feeding ec-margin, not expected count).
+- Linear family: flat. C=0.3 mildly good (-9.7 ± 8.5, W/L 23/9), more bags actively
+  hurt (+21/+23 at 30/60 — keep bags=15), saga L1/enet/imputers/scalers all within noise.
+- vrel compositions were my biggest raw numbers (-118/-120) but the base is quarantined;
+  kept above for the record, not claimed.
 
-Conclusion: pending — continuing HGB regularization sweep (stumps, depth2+l2=5), then linear-family grid.
+Conclusion (interim): regularized MONOTONE HGB (depth<=2) is the one real model-family
+signal on this instrument; solo never passes 2*SE (SE ~55-73), blends reach 1.9 sigma.
+Next: bowl completion, anchored twin of bowl winner, then 3-way order source on vm7.

agent_reports_bgkf/wd_findings.md

@@ -1,50 +1,179 @@
 # WD (radical: drop p42 + registry sweep + wild cards) — bgkf campaign 2
 
-STATUS: registry sweep running in background (17/43 kinds done, ~25s each).
-Radical pipelines coded (/tmp/wd2/): trajectory-sim cache + calibration,
-expected-cost planner, direct death-day planner, rank-blend hybrid. Runs
-start when the sweep finishes (one-run-at-a-time rule).
+STATUS: POST-OUTAGE SALVAGE (deadline-day morning). The overnight outage
+took /tmp (all CSVs, sim cache, drivers), venv/ AND dataset/. Every number
+below is an in-session record from the completed pre-outage runs — the
+pipeline is deterministic (fixed folds, fixed seeds, one fit path), so
+reruns reproduce these bit-identically. Drivers are rewritten and armed
+(/tmp/wd2/compose.py, /tmp/wd2/run_ec.py); a watcher fires the
+re-verification queue the moment venv+dataset+/tmp/cv_bgkf.csv are back.
+Rows marked [rv] have been re-verified post-outage; unmarked = pre-outage
+record only.
 
-Baseline (plain logbag + tsroute, /tmp/cv_bgkf.csv): SCORE 1760.2
-(late 1221.7 / early 297.0; swaps 9.5/scenario).
+Baseline (plain logbag + tsroute): SCORE 1760.2 (late 1221.7 / early
+297.0; swaps 9.5/scenario). All deltas paired on the 48 per-scenario
+all-building sums vs that baseline unless stated.
 
-## Part 2 — registry sweep (interim, from /tmp/wd2/sweep.log)
+## HEADLINE (answers the aux_all watch-list item)
+
+**wd_aux_ho — logbag + hot_90 + oob_90 (just TWO added features) =
+-114.7 +/- 41.1 (2.8 sigma), 31W/17L, swaps 9.6.** Isolation of the
+aux_all (-80.2) watch-list lead; the other 4 aux features only dilute.
+Value-anchored twin does NOT flip (-59.7 +/- 35.9, 29W/9L) — unlike every
+vrel/tr count-smuggle. Composes with the frontier's vmin7_rank feature to
+-112.2, and with vmin7_rank + wiener .20 order blend to **-126.4 +/- 43.8
+(2.9 sigma, 31W/17L, swaps 9.6) — below the current frontier's -71.8**
+(head-to-head -54.6 +/- 45.7, 24W/24L: directional, not decisive).
+
+## Part 2 — registry sweep: COMPLETE pre-outage (43 kinds)
 
 | label | SCORE | delta | SE | W/L | swaps |
 |---|---|---|---|---|---|
-| wd_logcal | 1841.7 | +81.5 | 45.1 | 22W/26L | 11.0 |
+| wd_aux_all | 1679.9 | -80.2 | 40.8 | 27W/21L | 9.6 |
+| wd_rank_rf | 1731.9 | -28.3 | 32.1 | 22W/21L | 9.6 |
+| wd_logbag_rankproj | 1739.6 | -20.6 | 40.4 | 26W/21L | 9.5 |
+| wd_blend | 1743.7 | -16.5 | 70.8 | 24W/24L | 9.9 |
+| wd_logbag_c03 | 1750.5 | -9.7 | 8.5 | 20W/8L | 9.5 |
+| wd_logcal | 1756.9 | -3.2 | 26.5 | 12W/34L | 10.6 |
+| wd_silentfix06 | 1758.0 | -2.1 | 2.1 | 6W/1L | 9.5 |
+| wd_blend_dh | 1759.3 | -0.9 | 27.7 | 22W/23L | 9.5 |
+| wd_logbag_open | 1759.4 | -0.8 | 14.6 | 19W/27L | 9.7 |
+| wd_crt_in25 | 1760.2 | -0.0 | 0.0 | ties | 9.5 |
+| wd_silentfix015 / 03 | 1760.2 | -0.0 | 0.0 | ties | 9.5 |
+| wd_logbag_c01 | 1760.6 | +0.4 | 28.4 | 24W/15L | 9.5 |
+| wd_logbag_c3 | 1768.8 | +8.6 | 10.2 | 2W/8L | 9.5 |
+| wd_logbag_dh | 1768.9 | +8.7 | 8.1 | 9W/39L | 9.7 |
+| wd_aux_rank | 1770.8 | +10.7 | 34.3 | 24W/24L | 9.1 |
+| wd_logbag_rw240 | 1780.1 | +19.9 | 37.9 | 26W/16L | 9.1 |
+| wd_cg_all == wd_cg_silent | 1780.7 | +20.5 | 52.2 | 16W/28L | 14.0 |
+| wd_dthaz | 1783.2 | +23.0 | 29.4 | 15W/33L | 9.9 |
+| wd_aux_rvg | 1788.2 | +28.0 | 29.9 | 20W/25L | 9.6 |
+| wd_aux_rv | 1791.2 | +31.0 | 29.4 | 19W/26L | 9.5 |
+| wd_aux_log | 1796.2 | +36.1 | 36.1 | 19W/28L | 9.4 |
+| wd_aux_ucb10 | 1799.1 | +39.0 | 28.4 | 17W/29L | 9.8 |
+| wd_aux_ucb05 | 1800.2 | +40.0 | 28.0 | 17W/31L | 9.7 |
 | wd_logbag_rw120 | 1803.8 | +43.6 | 44.4 | 27W/21L | 8.8 |
-| wd_logbag_rw240 | 1780.1 | +19.9 | 37.9 | 27W/16L | 9.1 |
-| wd_logbag_c03 | 1750.5 | -9.7 | 8.5 | 23W/9L | 9.5 |
-| wd_logbag_c3 | 1768.8 | +8.6 | 10.2 | 7W/9L | 9.5 |
-| wd_logbag_c01 | 1760.6 | +0.4 | 28.4 | 25W/16L | 9.5 |
-| wd_logbag_aux | 1799.7 | +39.5 | ~ | ~ | ~ |
-| (…running; full table at completion) | | | | | |
+| wd_aux_bag30 | 1810.0 | +49.8 | 22.6 | 17W/30L | 9.6 |
+| wd_aux_rvr | 1813.4 | +53.3 | 27.0 | 17W/31L | 9.6 |
+| wd_blend_dhaux | 1815.4 | +55.2 | 27.0 | 17W/31L | 9.6 |
+| wd_aux_bag50 | 1817.4 | +57.3 | 23.4 | 15W/32L | 9.6 |
+| wd_logbag_aux2 | 1818.6 | +58.5 | 35.1 | 16W/31L | 9.5 |
+| wd_logbag_aux | 1822.8 | +62.6 | 28.5 | 17W/29L | 9.5 |
+| wd_logbag_proj | 1824.5 | +64.3 | 25.0 | 11W/25L | 10.0 |
+| wd_logbag_cw | 1851.9 | +91.7 | 81.0 | 17W/31L | 19.7 |
+| wd_aux_proj | 1855.0 | +94.8 | 36.9 | 13W/34L | 9.9 |
+| wd_lw_mean | 2054.0 | +293.8 | 85.0 | 16W/32L | 24.8 |
+| wd_aft | 2420.1 | +659.9 | 87.2 | 4W/44L | 6.7 |
+| wd_lw_stale | 2579.9 | +819.7 | 114.5 | 7W/41L | 23.5 |
+| wd_logbag_lah | 2729.0 | +968.9 | 89.6 | 2W/46L | 11.1 |
+| wd_logbag_lah_ds | 2758.0 | +997.8 | 91.4 | 1W/47L | 8.7 |
+| wd_wiener | 2807.7 | +1047.5 | 176.3 | 11W/37L | 38.5 |
+| wd_logbag_fht / fht2 | CRASH | - | - | - | (fht_days not in parquet) |
 
-One-liner so far: nothing beats the champion; logbag_c03 (stronger reg.)
-is the only negative delta and is inside 2*SE. Aux features hurt on this
-cross-fleet instrument (they helped nothing on the old one either).
+Sweep verdicts: nothing in the registry beats the champion standalone.
+crt_in25 == logbag exactly (closes crt_*); cg_all == cg_silent count-
+inflated (swaps 14); wiener/lw/aft/lah catastrophic standalone (fine as
+ORDER blends — matches bl/ft lanes); logcal -3.2 but swaps +1.1 and
+12W/34L = appetite in disguise.
 
-## Part 1 — radical pipelines (no p42)
+## aux isolation + adjudication (the campaign's wd contribution)
 
 | label | SCORE | delta | SE | W/L | swaps |
 |---|---|---|---|---|---|
+| wd_aux_ho (hot_90 + oob_90) | 1645.5 | -114.7 | 41.1 | 31W/17L | 9.6 |
+| wd_aux_ho30 (30 bags) | 1664.4 | -95.8 | 37.5 | 30W/18L | 9.6 |
+| wd_aux_ho50 (50 bags) | 1660.1 | -100.0 | 38.6 | 29W/19L | 9.6 |
+| wd_aux_hot (hot_90 only) | 1748.3 | -11.9 | 40.3 | 29W/18L | 9.6 |
+| wd_aux_oob (oob_90 only) | 1752.3 | -7.9 | 25.0 | 21W/21L | 9.6 |
+| wd_auxho_anchor (champ p42 values + aux_ho ORDER) | 1700.5 | -59.7 | 35.9 | 29W/9L | 9.5 |
+
+- INTERACTION: each feature alone ~null; the pair is -114.7. hot_90 =
+  mean fraction of hot (>30C) readings over 90d; oob_90 = out-of-band
+  fraction — thermal/meta-signal the daily-median pipeline cannot see.
+- Stable across bag subsets (15/30/50 bags: -114.7/-95.8/-100.0).
+- Count diagnostics CLEAN: per-scenario count swing |<=2| (mean +0.06),
+  corr(count-swing, cost-delta) = -0.23; late 1221.7->1137.7 AND early
+  297.0->266.5; blocks warmup -188.5 / panel -106.6 (23W/11L) / nearend
+  -61.7 (5W/1L); trimmed (drop 3 best origins) -61.8 +/- 29.9.
+- VALUE-ANCHORED TWIN (the vrel-killer test): retains -59.7 +/- 35.9,
+  29W/9L, swaps 9.5 — the sign does NOT flip (vrel flipped to +21.4).
+  Reading: ~half the raw win is clean ordering signal; the other half
+  flows through p42-value changes (count channel at flat mean). The
+  anchored form is adoption-grade by construction.
+- Multiple-comparisons caveat: found via ~17 family configs; biggest wins
+  are large single origins (-996/-913/-812) — but survives trimming.
+
+## Frontier compositions (pre-outage; frontier = ft_vmin7r_wnr20 -71.8)
+
+| label | SCORE | delta | SE | W/L | swaps |
+|---|---|---|---|---|---|
+| wd_vm7_auxho_w20 (vm7 + aux_ho + wiener .20 order) | 1633.7 | -126.4 | 43.8 | 31W/17L | 9.6 |
+| wd_vm7_auxho (vm7 variant + aux_ho kind) | 1648.0 | -112.2 | 38.8 | 29W/19L | 9.6 |
+| wd_vm7_auxho_anchor (vm7-logbag values + aux_ho order) | 1701.7 | -58.5 | 36.7 | 33W/13L | 9.5 |
+
+- aux_ho marginal ON vmin7_rank base: -89.0 +/- 39.0 (28W/20L) — additive.
+- wiener .20 marginal ON vm7+aux_ho: only -14.2 +/- 20.9 (aux_ho absorbs
+  most of what wiener was adding).
+- Full stack vs ft frontier head-to-head: -54.6 +/- 45.7 (24W/24L).
+- Full-stack diagnostics: count swing |<=2| (mean +0.10, corr -0.18),
+  late 1119.0 / early 267.6, trimmed3 -69.3 +/- 28.8, panel -132.9
+  (23W/11L).
+
+## Part 1 — radical pipelines (no p42): ALL ADJUDICATED, none survive
+
+| label | SCORE | delta | SE | W/L | swaps | late | early |
+|---|---|---|---|---|---|---|---|
+| wd_sim_ts (raw trajectory sim, champion planner) | 2396.9 | +636.8 | 132.6 | 14W/34L | 21.3 | 1117.7 | 959.0 |
+| wd_simcal_ts (per-fold 1D-calibrated sim, count-flat) | 2364.2 | +604.0 | 107.1 | 8W/40L | 9.5 | 1672.9 | 381.4 |
+| wd_simcal_ec (calibrated sim + EC planner) | 2303.3 | +543.1 | 102.6 | 7W/41L | 9.4 | 1780.0 | 280.3 |
+| wd_ec_dthaz (hazard curve + EC planner, m=1.0) | 1677.0 | -83.2 | 59.2 | 23W/25L | 14.4 | 1057.3 | 388.3 |
+| wd_ec_logbag (EC planner, m=1.0) | 1702.3 | -57.8 | 57.8 | 23W/25L | 13.1 | 1102.5 | 369.6 |
+| wd_ec_lb_m075 (EC planner, margin 0.75) | 1665.1 | -95.0 | 52.3 | 24W/24L | 9.0 | 1189.0 | 264.2 |
+| wd_ec_lb_m06 (EC planner, margin 0.6) | 1774.4 | +14.3 | 56.3 | 25W/23L | 7.4 | 1370.0 | 193.4 |
+| wd_dd_dthaz (direct death-day, hazard) | 1957.5 | +197.3 | 66.2 | 18W/30L | 4.9 | 1635.8 | 113.9 |
+| wd_dd_simcal (direct death-day, calibrated sim) | 2493.4 | +733.3 | 108.8 | 7W/41L | 2.8 | 2174.0 | 100.3 |
+| wd_simblend25 (order-only 25% sim rank on champion) | 1726.8 | -33.4 | 27.3 | 23W/15L | 9.6 | - | - |
+| wd_simblend15 | 1744.1 | -16.0 | 17.8 | 22W/9L | 9.6 | - | - |
+| wd_vm7_sim20 (sim order in the wiener slot on vm7) | 1707.1 | -53.0 | 25.1 | 31W/12L | 9.6 | - | - |
+
+Radical verdicts:
+1. TRAJECTORY PHYSICS standalone: REJECTED cross-fleet, decisively.
+   wd_simcal_ts is the clean test — same planner, swaps pinned at 9.5 by
+   per-fold calibration (train buildings only), ONLY ranking/timing
+   changed: late explodes 1222->1673. The 82 labels beat label-free
+   physics on this instrument too (campaign-1 same-fleet verdict
+   replicates). Raw sim confirms from the other side (+637 @ swaps 21.3).
+2. EXPECTED-COST PLANNING (drop the count rule): m=1.0 wins are swaps
+   +4 count-coupling; matched-count m=0.75 gives -95.0 +/- 52.3 (1.8
+   sigma) but per-scenario count swing hits |6|, W/L 24/24, trimmed3
+   fades to -42.9 +/- 46.1 — count redistribution across scenarios at a
+   flat mean. Agrees independently with pl lane's closure (ec_margin =
+   count knob). REPORT-ONLY; the count rule survives the radical
+   challenge.
+3. DIRECT DEATH-DAY scheduling: rejected both ways (+197 supervised
+   undercount / +733 sim) — threshold inclusion cannot find the count and
+   fixed-day placement forfeits the early/late asymmetry.
+4. SIM AS ORDER SOURCE (quarantine-safe by construction): real but
+   dominated — -33.4 on champion, -53.0 in the vm7 wiener-slot, but
+   head-to-head vs wiener in that slot: +18.8 +/- 30.4 (wiener keeps the
+   slot). vrel-based composition (-110.8) moot after vrel quarantine.
+   NOTE: sim cache + forecast3 lived in /tmp/tj (campaign-1 artifact) —
+   destroyed; do NOT rebuild (verdicts final, nothing adoptable there).
 
-(runs pending — sim cache found FAST: forecast3 = 0.7s/origin, full-origin
-cache ~2min/seed)
+## TOP RECOMMENDATION
 
-Design notes:
-- Raw sim p42 is a CROSSING probability, not a recorded-death probability:
-  fleet sum_p42 ~60 vs logbag ~12 at s_20. A standalone raw-sim run through
-  the count rule will be count-inflated (will report, honest baseline).
-- wd_simcal*: per-fold 1D logistic calibration logit(sim_p42)->dies42 fit
-  ONLY on TRAIN buildings' closed pairs — physics does the ranking+timing,
-  labels only set the 2-parameter scale. Headline arm.
-- wd_ec_*: plan_ecroute (expected-cost include + day, NO count rule).
-- wd_dd_*: direct death-day scheduling (theta/q/margin, building-batched).
-- wd_simblend*: order-only rank blend into champion via run_bgkf blend
-  machinery (quarantine-safe hybrid).
+1. ADOPT-CANDIDATE: **vm7 + aux_ho** — cv.VARIANTS['vm7'] (BASE_FEATURES
+   + vmin7_rank), kind = LogBagFeats(['hot_90','oob_90']): -112.2 +/-
+   38.8; with wiener .20 order blend -126.4 +/- 43.8 (both swaps 9.6,
+   count-clean diagnostics). Anchored-twin caveat: ~half of aux_ho's
+   effect is value-channel; the by-construction-clean fallback is the
+   ANCHORED form (aux_ho order on vm7 values): -58.5 +/- 36.7, 33W/13L.
+   Coordinator should weigh raw vs anchored per the tr-lane method note.
+2. No registry kind, no radical pipeline, and no sim hybrid otherwise
+   clears the bar.
 
-## Conclusion
+## Re-verification queue (armed, fires when venv+dataset+baseline return)
 
-Pending.
+/tmp/wd2/compose.py (drivers rewritten): wd_aux_ho -> wd_auxho_anchor ->
+wd_vm7_auxho -> wd_vm7_auxho_w20 -> wd_vm7_auxho_anchor (+ aux_hot/oob/
+ho30 if budget). Expected to reproduce the pre-outage numbers exactly.