Auto-sync: 2026-06-30 03:55:26

results/README.md

@@ -68,6 +68,8 @@ Use these summaries for the current paper tables:
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_grid035040045_safe_margin0p20_noopbonus0p03_summary.md`,
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_summary.md`,
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_summary.md`,
+	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_oraclek8trace_summary.md`,
+	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_nmchallenger0p02_summary.md`,
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_srcscorebonus0p02_summary.md`,
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_grid035040045_safe_margin0p20_noopbonus0p03_srcscorebonus0p02_summary.md`,
 	  `h16_policy_ckpt_near_miss_policy_bc5_bestpt_retrieval_residual_k4_composemasked_dropnmnoop_l2comp002_grid035040045_safe_margin0p20_noopbonus0p03_summary.md`,
@@ -124,24 +126,30 @@ Completed deployment-clean diagnostics:
 
 Completed residual transport jobs:
 
-Last checked: `2026-06-29 21:42 UTC`. The K4 masked composed type-consensus
-sweep completed and produced a new clean best, 35.59%, when the exact
+Last checked: `2026-06-30 03:45 UTC`. The K4 masked composed type-consensus
+sweep produced the previous clean best, 35.59%, when the exact
 `residual_near_miss+residual_no_op` composite is additionally masked. Pure
 masked composition reached 35.30%, and masked composition with the exact typed
 no-op prior reached 35.54%. Raw selected candidate types show no random-negative
 or wrong-direction composite leak. The paper table/paired analysis now use the
-exact compatibility-mask row as `best_clean_key`. Component-wise composite
+near-miss challenger row as `best_clean_key`. Component-wise composite
 priors are negative (35.36%), and a train-source score prior on the exact
-compatibility chart reaches 35.48%, below the 35.59% top row. Adding the same
+compatibility chart reaches 35.48%, below the previous 35.59% compatibility row. Adding the same
 source-score prior to the typed no-op exact-mask row reaches 35.54%, also below
 the top row. The mask-only exact compatibility ablation completed at 35.48%,
 isolating the compatibility mask from typed-prior tuning.
 The singleton near-miss revival diagnostic completed: exact
-`residual_near_miss` bonuses 0.01 and 0.02 both tie the 35.59% top row and
+`residual_near_miss` bonuses 0.01 and 0.02 both tie the previous 35.59% compatibility row and
 slightly raise mean progress to 57.10%, but do not improve success.
 Composite-only L2 trust penalties reduce
 action MSE but do not improve success: standalone L2 `0.02` ties 35.54%, L2
 `0.05` reaches 35.48%, and exact compatibility + L2 `0.02` reaches 35.54%.
+The unique-action K=8 candidate-oracle diagnostic is complete: the archived
+non-unique run remains `_nonunique`, while the deduplicated branch trace reaches
+43.07% candidate-oracle success with mean best branch rank 2.85. The resulting
+trace-motivated near-miss challenger gate (`residual_near_miss`, margin 0.02)
+is deployment-clean and reaches 36.00% mean success with 57.42% progress, the
+current clean best.
 
 - Code supports `--retrieval-residual-scale`, defaulting to `1.0` so the
   completed 32.12% result is unchanged.
@@ -293,7 +301,8 @@ action MSE but do not improve success: standalone L2 `0.02` ties 35.54%, L2
 - Jobs `14915009`/`14915013`/`14915014` completed the exact compatibility-mask
   GPU array, summary, and rebuild. Dropping only `near_miss+no_op` raises the
   clean row to 35.59% with seeds 35.30%/34.78%/36.70%, progress 57.07%, and
-  action MSE 0.406. This is the current best clean deployment diagnostic.
+  action MSE 0.406. This is the previous best clean deployment diagnostic before
+  the near-miss challenger gate.
 - Job `14915146` completed the CPU smoke for exact compatibility masking plus
   composite-only L2 trust penalty `0.02`.
 - Jobs `14915213`/`14915217`/`14915251` completed the exact compatibility + L2
@@ -382,9 +391,13 @@ action MSE but do not improve success: standalone L2 `0.02` ties 35.54%, L2
 - The evaluator now pads/slices residual horizons to the policy horizon
   defensively and deduplicates candidate-oracle actions before oracle selection;
   invalid padding branches cannot win. Jobs `14953513`/`14953522`/`14953524`
-  rerun the unique-action K=8 candidate-oracle diagnostic.
+  completed the unique-action K=8 candidate-oracle diagnostic at 43.07%.
+  Jobs `14953960`/`14953961` completed the branch-trace rerun used for selector
+  calibration.
+- Jobs `14954280`/`14954281` completed the trace-motivated near-miss challenger
+  gate clean deployment run at 36.00%, the current clean best.
 - Jobs `14936131`/`14936132`/`14936133` completed the compatible-chart
-  abstention test at margin `0.10`: it reaches 34.67%, below the 35.59% top row.
+  abstention test at margin `0.10`: it reaches 34.67%, below the 36.00% top row.
   This rules out a naive lower-margin fix for the apparent oracle headroom.
 
 Paper table automation: