Regulator replay · deterministic reconstruction

JSONL audit trail replay

Every audit line can be replayed into the six stages that produced it. Pick a transaction, then walk IntentRouter → RAG → Graph/Memory → PolicyEngine → Q-LOCK → Audit and see exactly which fact moved the verdict.

$ python -m fde_toolkit aiqos --export audit.jsonl

Transactions

120 of 500

Replay · TXN-SYN-20000

6 stages · 427 ms reconstructed

IntentRouter

ROUTEactor router · 9 ms

Input: raw request · domain DATA_CENTER
Output: infra.telemetry_review → Operations agent

intent	infra.telemetry_review
agent	Operations agent
model path	SLM specialist
confidence	0.91

Classify before spending a token; specialist routes are cheaper and faster than the frontier fallback.

stage 1 / 6

Stored audit record

the JSONL line this replay was reconstructed from

{
  "actor": "agent",
  "event": "infra.telemetry",
  "transaction": "TXN-SYN-20000",
  "model": "risk-model-v12",
  "confidence": 0.83,
  "policy": "AML-POLICY-v18",
  "q_lock": "ALLOW",
  "reason": [
    "none"
  ],
  "human_review": false
}