Audit K16 LCB dominance diagnostic

workspace/reports/draft_audit_ctt.md

@@ -141,6 +141,18 @@ diagnostic, but it is still hand-built RGB statistics rather than learned
 object-centric CTT geometry. Object-layout-only reaches 0.3125, combined
 RGB-stat/object-layout ties 0.3264, and RGB-stat success-weighted reaches
 0.3333, so deterministic object-layout features are not yet the missing method.
+The score-source conformal LCB fallback now has a K=16 train-calibrated Part-G
+artifact as well:
+`runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test`. It calibrates
+only on train-split K=16 measured rollout rows and evaluates once on held-out
+test rows. The auto threshold covers 0.1319 of held-out rows and reaches
+0.2778 selected success, below the 0.2917 base action; the fixed tau0 rerun
+`runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0` falls back
+0.8889 of the time and only matches the base at 0.2917. Both runs have complete
+action-bound safety labels, unsafe execution rate 0.0000, and pairwise causal
+calibration ECE 0.1633 over 12,434 within-chart measured pairs. Therefore the
+plain score-source LCB rule is a useful safety/reporting artifact, but a
+negative selector result rather than a reliable causal dominance certificate.
 The nonlinear selector path previously accepted chart-compat feature sets
 without loading selector chart maps, making that diagnostic too weak. It now
 loads the same deployment-visible selector chart maps as the ridge selector and
@@ -350,6 +362,12 @@ Current artifact evidence:
   success selector gap 0.2431, with pairwise causal calibration ECE 0.0150
   over 12,434 held-out candidate pairs. It uses train-only target--source
   RGB-stat chart compatibility and remains a selector diagnostic.
+- `runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test` and
+  `runs/ctt_base_context_obs_dominance_envclip_k16_train_to_test_tau0` contain
+  the score-source conformal LCB dominance diagnostics for the same K=16 rows.
+  They now report pairwise causal calibration ECE 0.1633, complete
+  action-bound safety label coverage, unsafe execution rate 0.0000, and
+  selected success 0.2778--0.2917, so they are negative selector diagnostics.
 - `runs/ctt_base_context_obs_nonlinear_dominance_chartcompat_obs_*_envclip_k16_train_to_test`
   and `runs/ctt_base_context_obs_nonlinear_dominance_chartcompat_obs_*_reg_envclip_k16_train_to_test`
   contain the fixed nonlinear chart-compat diagnostics. They now log