ATLAS-VCE ·

Value-convergence excess across the fleet beats the capability-matched null

Refuted K-vce fired

frozen 2026-07-18 · adjudicated 2026-07-19 · ev:0c80d9086ebb0ad9704b88a8eefb01fa · spec:2008bee7d02b9

The frozen predictions

  1. H-vce value-convergence excess on real reward-model pairs beats the capability-matched random-utility null
    real_vce > 0 registered p 0.5 real_vce Value -0.06987390356875109 Subject ATLAS-VCE Threshold > 0 Trust Adjudicated Evidence ev:0c80d9086ebb0ad9704b88a8eefb01fa Refuted
  2. H-vce-p the convergent-pair alignment exceeds the random-utility null at p < 0.05
    reward_convergent_p_value < 0.05 reward_convergent_p_value Subject ATLAS-VCE Threshold < 0.05 Trust Adjudicated Evidence ev:0c80d9086ebb0ad9704b88a8eefb01fa Confirmed

K-vce fired on real_vce <= 0 , observed -0.06987 : convergence is fully explained by shared capability structure. A fired kill criterion is a registered negative result, not a failure of the machinery.

Adjudication figure for ATLAS-VCE
ATLAS-VCE fired its kill criterion (K-vce); the registered negative result.

Registered companions

Subjects and artifacts

  • signals: fleet
  • datasets: rb2-full , rb2-helpfulness-512
  • frozen spec: campaign-atlas-vce.json
  • figure files: light · dark
  • study id: study:campaign-atlas-vce@v1#2008bee7