Trust & Risk · Audit Trail
Who did what, on whose authority, with which evidence
An agent action without a trail is not an action we allow. Each entry carries the actor, the policy version, the evidence and the human who authorised it — exportable as a packet a regulator can read without us in the room.
Entries today
612k
Retention
7 years
regulated surfaces
Actions without evidence
0
Regulator exports this year
4
all served from this trail
Search the trail
Filter by actor, action or evidence reference
| When | Actor | Action | Evidence |
|---|---|---|---|
| today 11:42 | @compliance-guard | Vetoed a creative claiming assured returns | Clause 4(2)(f), draft v3, reviewer Ritika Shah |
| today 10:18 | Ritika Shah | Approved a research note for publication | Sources 6/6 verified, verdict id cv-88412 |
| today 09:55 | @spend-optimiser | Proposed ₹4.2L reallocation | Held for Nikhil Verma, money gate active |
| yesterday 19:30 | @retention-sentinel | Sent 2,140 save messages | Template rs-14, consent verified per member |
| yesterday 16:04 | Meera Iyer | Promoted @support-triage to act-with-approval | Eval 88.1 for 14 days, escalation 4.2% |
What an entry must contain
The schema is enforced at write time
Actor
An agent handle or a named human. Never 'the system'.
Authority
The autonomy level and, for gated actions, the approving human.
Policy version
The exact prompt and guardrail version in force at that second.
Evidence
Passage ids, clause references, verdict ids or a consent token.
Reversal path
How the action can be undone, and by whom.
Run id
Links to the full trace and the gateway usage line.