Skip to content

Agent Roles

James Nikolas Putradi edited this page Nov 20, 2026 · 1 revision

Agent roles

There is no human law author. This page is the practical view of Appendix J; the appendix is normative.

The roles

Role Writes Must not read Scored on
Conductor task graph, manifests, locks, merges — protocol integrity
LAA model.bend, LAWS.bend, witnesses, abs.yaml, props_locked.sv, manifest oracle, seeded bugs, PROOF_ref.bend, agent traces law discrimination (LA3-LA6)
RPA PROOF_ref.bend + effort log oracle, traces provability within the effort cap
OAA oracle TB, oracle SVA, golden RTL, seeded bugs, checklist every LAA artifact oracle recall on the bug corpus
PA props_c2.sv the Bend laws, the oracle parity within 5 pp
AA LA-gate evidence, adversarial review oracle plaintext defects found
IA the RTL and PROOF.bend under test every protected path the arm's visible gates (and nothing else)

How independence is enforced, not promised

  1. Separate Freebuff runs. Different session, different workspace, no shared conversation. Roles communicate through committed files and issue links only.
  2. Read-only mounts. The policy proxy mounts protected paths read-only and logs every read attempt.
  3. Different model families where the compute budget allows, recorded per artifact. A same-family LAA/OAA pair is refused by the provenance check unless declared.
  4. Different objectives. LAA is rewarded for discrimination, OAA for recall, AA for findings. No role is scored on another role's metric.
  5. Hash locks. Once locked, a law file changes only through an amendment.
  6. Evidence, not sidecars. The policy-proxy audit log is the proof the barrier held; provenance.yaml is a claim about it.

The failure this invites

A law author and an implementation agent from the same model family can agree on the same blind spot: the laws omit what that family tends to omit, the agent passes easily, and C3 looks strong while measuring little.

Four defences, all pre-registered:

  • Mutation pressure (LA3, LA4): a blind-spot law set fails to kill plausible bugs.
  • Independent oracle recall: OAA's kill rate is measured against the same corpus.
  • Adversarial auditor: AA proposes implementations that satisfy every law and violate spec.md.
  • Ablation A7 + LAW-FAR: same-family versus cross-family FAR becomes a moderator, not noise.

Ambiguity, resolved deterministically

If LAA and OAA read the spec differently:

  1. both readings are published in ambiguities.yaml;
  2. the reading stricter on the design becomes the law; the other becomes an oracle-only requirement;
  3. if the readings are behaviourally incompatible, the task is quarantined — never silently resolved.

What a human still does

Release approval, ethics, external correspondence, and adjudication of an unresolved blocker. A human may spot-check 3% of locked law sets for quality. No human writes law content.

Clone this wiki locally