Nº 000003 · APPEND-ONLY · THE LEDGER
Receipts for agent activity.
Every AI inference that touches your notes gets an append-only record — sequenced per agent, checkable at any time, kept as markdown files you own.
The artifact
What a receipt looks like.
- agent
- curtis
- seq
- 0041
- action
- inference · 3 tools read
- verify
- pass — sequence continuous
- seal
- sha256 91af…77c2 · append-only
Verify, not trust
Evidence by structure, not by promise.
Per-agent sequences
Each agent writes to its own monotonic sequence. Delete a receipt and the gap shows. Reorder or duplicate one and verification flags it.
Sharded daily files
Receipts shard into per-day files under Audit/ledger/, and a rolling digest summarizes the ledger — plain markdown, readable in any editor.
Retention off by default
Nothing is deleted unless you opt in. When pruning is opted into, each prune is itself recorded — the gaps it leaves are self-explaining.
The ledger is deliberately non-cryptographic: it makes tampering visible (gaps, duplicates, reordering), not impossible. A hash-chained tier is planned, and it is not promised before it ships.
Stated plainly
Trust is a record you can check.
Curtis Audit is its own Obsidian plugin — agents, other plugins, or a shell one-liner file receipts into the same append-only ledger, and the spec is public for anyone building agents of their own.