The Incident Receipt

A tamper-evident, downloadable JSON audit of exactly what every run did.

Every run produces a tamper-evident Incident Receipt — a downloadable JSON audit of exactly what happened. It’s the verifiable record that the agent did what it claims, so a reviewer can confirm which models, guardrails, and tools fired without trusting my word for it.

It captures:

• Which platform capabilities engaged.
• Every guardrail decision.
• The actions blocked vs executed.
• The fallbacks fired.
• The full event timeline.
• A SHA-256 integrity hash, hash-chained over the event timeline — anyone can recompute it from the timeline.

The hash is computed over the run’s event timeline, so altering any recorded step changes the digest. A reviewer recomputes integrity_sha256 from the timeline and compares — that’s what makes the receipt tamper-evident.

Example

{
  "run_id": "run_8f2c1a90",
  "scenario": "hallucinated_diagnosis",
  "outcome": "resolved",
  "summary": "Backstop caught an ungrounded diagnosis, re-routed to a stronger model, and rolled back the bad deploy. error_rate 0.0.",
  "root_cause": "checkout deployment patched to nginx:9.99-doesnotexist → ImagePullBackOff, zero ready pods",
  "features_engaged": [
    "ai_gateway_virtual_model",
    "priority_fallback_chain",
    "secrets_pii_redaction",
    "quality_gate",
    "llm_as_judge",
    "action_gate",
    "mcp_gateway_linear",
    "custom_infra_mcp",
    "managed_prompt_registry"
  ],
  "guardrail_decisions": [
    { "gate": "quality", "passed": false, "reasons": ["prod-db is not implicated by any metric or log error"] },
    { "gate": "quality", "passed": true, "reasons": ["suspected_resource=checkout is a real service", "confidence 0.91 >= 0.5"] },
    { "gate": "action", "passed": true, "reasons": ["scope != all", "target exists", "matches evidence"] }
  ],
  "actions_blocked": [
    { "tool": "rollback_deploy", "target": "prod-db", "scope": "all", "reason": "protected resource + blast radius" }
  ],
  "actions_executed": [
    { "tool": "rollback_deploy", "target": "checkout", "scope": "deployment", "result": "rolled back to previous ReplicaSet image" }
  ],
  "fallbacks": [
    { "from": "claude-sonnet", "to": "llama-4-maverick", "trigger": "429 rate-limit" }
  ],
  "integrity_sha256": "9b1c7e0d4f3a2c6e8d5b1a0f7c4e2d9b6a3f1e0c8d7b5a4f2e1c0d9b8a7f6e5d"
}