ERR-RB2 S16 · Certified robustness, attack surface and eval-awareness
Error anticipation on RewardBench 2
Refuted K-err fired
frozen 2026-07-18 · adjudicated 2026-07-19 · ev:6bf316c059230ff4ff99b118b1664631 · spec:c8ac36e72adcc
The frozen predictions
- H-err-qwen internal features beat the exposed seen-pair margin at predicting held-out RB2 completion errors on Qwen3-8B
err_auroc_delta_qwen8b > 0.03registered p 0.6 err_auroc_delta_qwen8b Value -0.0009738428193937221 Subject ERR-RB2 Threshold > 0.03 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631Refuted - H-err-llama the same holds on Llama-8B (cross-family)
err_auroc_delta_llama8b > 0.03err_auroc_delta_llama8b Value -0.003470736845503808 Subject ERR-RB2 Threshold > 0.03 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631Refuted
K-err fired on err_auroc_delta_min_family <= 0
, observed -0.003471
: internals add nothing over the seen-pair margin the RM already exposes when predicting held-out errors, on at least one family. A fired kill criterion is a registered negative
result, not a failure of the machinery.
Registered companions
-
err_auroc_internal_qwen8berr_auroc_internal_qwen8b Value 0.8580564710073929 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_margin_qwen8berr_auroc_margin_qwen8b Value 0.8590303138267866 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_delta_qwen8b_ci_lowerr_auroc_delta_qwen8b_ci_low Value -0.004634508227426182 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_delta_qwen8b_ci_higherr_auroc_delta_qwen8b_ci_high Value 0.0023281954874443313 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_internal_llama8berr_auroc_internal_llama8b Value 0.8728096018338405 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_margin_llama8berr_auroc_margin_llama8b Value 0.8762803386793443 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_delta_llama8b_ci_lowerr_auroc_delta_llama8b_ci_low Value -0.005095063654071338 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_delta_llama8b_ci_higherr_auroc_delta_llama8b_ci_high Value -0.0018241777525145248 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631 -
err_auroc_delta_min_familyerr_auroc_delta_min_family Value -0.003470736845503808 Subject ERR-RB2 Trust Adjudicated Evidenceev:6bf316c059230ff4ff99b118b1664631