Repository navigation
Publish evaluation-backed adversarial control claims #1008
Description
Publish the adversarial-control results established by #1007 for its exact realized profiles and backend configurations.
Acceptance criteria
- Link every claim to completed evaluation, runtime conformance, and concrete backend realization evidence. Bind the same contract/profile, mechanism/provider, policy, apparatus, attacker/monitor, memory, effect, and backend configuration throughout.
- Distinguish capability declarations, runtime enforcement, backend realization, experimental results, and mathematical assurance. Permitted in-world influence is not automatically a control failure; backend integrity is a separate boundary.
- Report safety, usefulness, costs, uncertainty, limitations, and unsupported conditions. Do not infer universal robustness, mechanism parity, backend equivalence, opacity, or bisimulation from the evaluation.
- Update ASR-536 completion evidence, assurance records, lineage, and public evaluation guidance. Document revised control semantics, migrations, and supported behavior #1019 owns control reference/migration documentation; Publish reproduced participant-crossing bisimulation documentation #976 owns publication of the separate finite-bisimulation result.
- Keep Hub OpenRAE/lilrae#36's evaluation evidence and Hub OpenRAE/lilrae#37's public guidance consistent with the published results.
Prerequisite: #1007, including the completed realization evidence for every claimed backend.
Activity
[-]Publish evidenced adversarial participant-control claims[/-][+]Deferred: Publish evidenced adversarial participant-control claims[/+][-]Deferred: Publish evidenced adversarial participant-control claims[/-][+]Publish evidenced adversarial participant-control claims[/+][-]Publish evidenced adversarial participant-control claims[/-][+]Publish evaluation-backed adversarial control claims[/+]