Board: md lane closed, no survivor - logbag(C=1,b=15) locally optimal; monotone HGB genuine but subsumed by frontier (+1.0 as 3rd source), shelf fallback only

agent_reports_bgkf/FRONTIER.md

@@ -46,10 +46,21 @@ Also dead: z-scores (weaker vrel), building relatives, dynamics ranks.
 - wiener FHT order blend w=0.20: additive on champion (-50.3), vrel (-34.1
   increment), logbag_rank (-40.7 increment). Compose it onto ANY base:
   blend=('wiener', 0.20).
-- md's depth-1/monotone HGB blend (hgbd1l25) on vrel: current leader; needs
-  wiener composition + count-matched check.
+- md's monotone depth<=2 HGB: genuine quarantine-clean order source in
+  isolation (anchored -44.7 +/- 38.6) but SUBSUMED on the frontier stack
+  (+1.0 +/- 24.9 as third source on vm7+wiener.20). SHELF FALLBACK ONLY,
+  to be used if the wiener source is ever quarantined. (The old
+  hgbd1l25-on-vrel leader row is void - vrel base quarantined.)
 
 ## Closed lanes
+- MODEL FAMILIES (md, complete 08-23, no survivor): keep champion
+  logbag(C=1, bags=15) - locally optimal (C=0.3 -9.7 n.s.; MORE BAGS HURT
+  +21/+23 at 30/60; scalers/penalties/imputers noise). Monotone HGB is
+  the family's one real ordering signal (bowl bottom depth2/l2=1 -86.4,
+  no solo config > 2SE; anchored twin holds -44.7) but adds +1.0 +/- 24.9
+  as third source on vm7+wiener.20 - subsumed, shelf fallback only.
+  Unconstrained gbm +49.3 (reproduced). Frontier reference independently
+  reproduced exact (1688.4, -71.8 +/- 23.5).
 - PLANNER (pl, complete, no survivor): routing/scheduling axis dead (+/-1 pt
   across NN ordering/restarts/split_days/chunking); risk-scaled placement
   sub-bar (-10.8 +/- 8.0, 2 lucky origins carry it; composes on vrel at