Fresh-seed BIB-CCS comparison

Open track · seeds 29, 31, 37, 41, 43 · 12 agents · horizon 40

Fresh-seed BIB-CCS score by SQC checkpoint The BIB specialist scores 0.8479 but loses general action accuracy. Uniform replay scores 0.8033. Cell-aware seed 42 scores 0.7824, the production default scores 0.7741, and cell-aware seed 73 scores 0.7695. 0.750.780.810.840.86 BIB specialist.8479 Uniform replay, w16.8033 Cell-aware, seed 42.7824 Production default.7741 Cell-aware, seed 73.7695
Production defaultCell-aware adapterReplay tradeoffBIB specialist

Cell-aware replay: multi-seed confirmation

The general-action guardrail replicated; the external BIB gain did not

Fresh BIB-CCS scores across three cell-aware training seeds Seed 17 rolled back to the production default at 0.7741. Seed 42 scored 0.7824. Seed 73 scored 0.7695. The mean was 0.7753 with standard deviation 0.0054. .765.770.775.780.785 Seed 17 · rollback.7741 Seed 42 · epoch 32.7824 Seed 73 · epoch 40.7695 Mean .7753 · SD .0054 · range .7695–.7824

Evidence decision

Latest classification of what the experiments establish

General action preservation
Demonstrated
75.31% mean exact action; 75.00% unseen; 66% worst-cell floor across training seeds.
Robust external BIB gain
Not demonstrated
Mean gain was .0013, smaller than cross-seed variation; only seed 42 improved.
Production selection
Balanced checkpoint retained
The BIB specialist and replay adapters remain experimental checkpoints.

Checkpoint tradeoffs

CheckpointFresh BIB-CCSV1.2 exact actionStatus
Balanced base.774175.25%Production default
BIB action-only.847944.04%Experimental specialist
Uniform replay, w16.803373.87%Experimental tradeoff
Cell-aware KD-256, 3-seed mean.775375.31%Preserved action; BIB gain not robust

BIB-CCS v1 open-track artifacts updated 13 July 2026. Scores describe synthetic benchmark evidence and do not by themselves establish construct, ecological, or operational validation.