Evidence · DiggingBeagle record
Falsification-Gated Self-Evolution challenges memory hypotheses before persistence and reduces harmful updates
Inspectable evidence for Falsification-gated agent memory reduced harmful accepted updates in controlled trials. The source describes the specific setting, measurement and limits summarized by its associated Claim.
- Source role
- primary disclosure
Evidence record
Inspectable evidence for Falsification-gated agent memory reduced harmful accepted updates in controlled trials. The source describes the specific setting, measurement and limits summarized by its associated Claim.
Claim-level citations (1)
- supportsFalsification-gated agent memory reduced harmful accepted updates in controlled trials: FGSE reports 3.6% harmful accepted agent-memory updates with falsification probes, compared with 10.6% in the ablation without those probes.
Section 6.3 ablation, harmful-update and post-commit-violation columns