The audit ledger

Every action — permitted or blocked — writes an entry. Each entry’s hash covers the full payload plus the previous entry’s hash, so the ledger forms a chain that cannot be silently tampered with: altering any entry breaks every hash after it.

What an entry contains

Field	Description
session_id	Agent DNA ID (sender)
classified_act	Action type (ACTION_PAYMENT_INITIATE, ACTION_AGENT_MESSAGE, etc.)
resource	Target resource or agent
status	SUCCESS · BLOCKED_POLICY · FAILED_RUNTIME · PENDING_CONFIRMATION
raw_intent	Original intent string or JSON payload
error	Reason for block/failure (empty on success)
prev_hash	SHA-256 of previous entry (chain link)
hash	SHA-256 of this entry

Verifying the chain

Verify the full chain at any time:

curl -H "X-Imara-Token: $TOKEN" http://localhost:7070/ledger/verify

Related surfaces:

Method	Path	Description
GET	/ledger	Query entries (limit, offset, status, session)
GET	/ledger/chain-head	Current chain head hash
GET	/ledger/stream	SSE real-time stream
GET	/ledger/mesh	Agent-to-agent traffic graph (window=1h|6h|24h|7d|all)

Getting the audit trail out

The ledger is designed so regulators and SIEMs never depend on the box it was written on:

• Cloud WORM replication — push-only replica to S3 / GCS / Azure / MinIO with Object Lock (SEC 17a-4(f) posture). Regulators read the bucket directly with read-only credentials; the kernel is not in the loop. Optionally routed through Kafka/RedPanda with the audit-consumer binary doing the final WORM write.
• Snapshots — periodic compressed (zstd) BoltDB backups through a crash-safe staging area to S3, with optional AES-256-GCM encryption.
• Webhooks — POST every commit (filterable by status) to Splunk, Slack, or any HTTP collector.
• Regulator metadata index — real-time BoltDB sidecar backing the multi-tenant fleet view (/audit-meta/fleet).
• Traces + metrics — policy VERDICT spans at /v1/traces, OTLP/HTTP JSON ingest at /v1/otel/v1/traces for agent-reported spans, Prometheus metrics at /metrics/prometheus.