Bind reasoning-failure risk and receipts to agent authority

[mdheller]

Parent

SocioProphet/sociosphere#271

Purpose

Make reasoning-failure risk affect agent authority: tool grants, memory access, autonomous execution, model route eligibility, and revocation posture.

Scope

Extend agent authority semantics to consume reasoning-failure and TrustOps posture records for:

• exactness risk;
• stale-memory contamination;
• unsupported causal claim risk;
• temporal inconsistency;
• modality contradiction;
• social-pressure/sycophancy risk;
• evaluator-bias risk;
• malicious/obfuscated-agent risk;
• premature termination or agreement without independent evidence;
• failed perturbation robustness gates.

Authority effects

Reasoning-risk posture should support:

• no authority change / record only;
• require review before autonomous execution;
• require deterministic tool verifier;
• restrict memory writeback;
• restrict retrieval hydration;
• restrict side-effecting tools;
• downgrade route eligibility;
• quarantine agent profile;
• revoke or suspend grant;
• restore authority only with passing follow-up receipt and policy approval.

Acceptance criteria

• Agent records can cite governing reasoning-failure receipt ids, TrustOps receipt ids, and guardrail action ids.
• A high-uncertainty or failed perturbation receipt can require human approval before autonomous execution.
• A failed exactness or modality conflict receipt can reduce tool/memory authority until verified.
• Passing follow-up receipts can restore authority only when policy allows it.
• No protocol-local alias can bypass canonical identity authority.