STYLE-RMB S16 · Certified robustness, attack surface and eval-awareness

Bias battery predicts RM-Bench Hard ranking

Inconclusive

frozen 2026-07-18 · adjudicated 2026-07-19 · ev:14bc4c666b5b115cf1fd2c86cf209f01 · spec:59c457674b4b4

The frozen predictions

  1. H-style-transfer internal style susceptibility predicts the fleet's RM-Bench Hard ranking
    spearman_biasbattery_vs_rmbench_hard > 0.6 registered p 0.5 not measured Inconclusive
  2. H-style-perm the fleet-level correlation clears its permutation null
    biasbattery_rmbench_perm_p < 0.05 not measured Inconclusive
  3. H-style-baseline the internal battery at least matches the behavioral degradation baseline
    spearman_minus_behavioral_baseline >= 0 not measured Inconclusive
Adjudication figure for STYLE-RMB
STYLE-RMB was not adjudicated this run: missing intermediate: no intermediate 'campaign.bias.battery' for roster_key='armorm' slice='diagnostic-v3-degradation'; the arc that produces it has not run or its shard was not merged.

Subjects and artifacts

  • signals: fleet
  • datasets: rmbench-full , diagnostic-v3
  • frozen spec: campaign-style-rmb.json
  • figure files: light · dark
  • study id: study:campaign-style-rmb@v1#59c45767