Sample · Illustrative

The HUMMBL Evidence Packet

A portable bundle of runtime governance receipts — signed delegation, append-only event log, stop-state record, and cost governor decisions. It travels outside any vendor dashboard, into an audit, a customer review, or a regulator's hands.

What is an evidence packet?

A dashboard can tell you that an agent ran. It cannot tell you who authorized the agent, what it was allowed to touch, what it actually touched, or what evidence survived the run.

An evidence packet is the difference. It is a file (or set of files) that documents what happened during an agent run — and it does not depend on HUMMBL's dashboard being open, on a vendor staying in business, or on a log not having rotated.

In plain language: a dashboard is a window. An evidence packet is a receipt. Windows close. Receipts survive.

What's inside a sample packet

1. Signed delegation

Who authorized the agent, for what scope, for how long

An HMAC-SHA256 signed delegation token recording the authorizer, the scope of authority, the chain depth, and the expiry. This is the answer to "who allowed the agent to do that?"

token_id: dct-2026-06-24-a1b2c3 authorizer: operator-steward scope: ["github:read", "linear:write"] chain_depth: 2 issued_at: 2026-06-24T14:00:00Z expires_at: 2026-06-24T18:00:00Z signature: hmac-sha256:7f3e...a91c
2. Append-only event log

What the agent actually touched, in order

An append-only log of agent actions — timestamped (UTC), attributed to a canonical identity, never edited or deleted. This is the answer to "what actually happened?"

2026-06-24T14:02:11Z agent=autonomous-agent action=github.list_issues repo=hummbl-dev/founder-mode 2026-06-24T14:03:45Z agent=autonomous-agent action=linear.create_issue team=ENG 2026-06-24T14:05:02Z agent=autonomous-agent action=bus.post type=STATUS
3. Stop-state record

Whether any kill switch or circuit breaker fired

The kill-switch state at run start and end, plus any circuit breaker transitions. This is the answer to "did anything stop the agent, and when?"

kill_switch: DISENGAGED (start) -> DISENGAGED (end) circuit_breaker: github_adapter CLOSED (start) -> CLOSED (end) circuit_breaker: linear_adapter CLOSED (start) -> HALF_OPEN (end, 1 transient error)
4. Cost governor decisions

What the run cost, and whether any budget bound fired

The cost governor's spend record for the run — budget ceiling, actual spend, and any halt decisions. This is the answer to "what did it cost, and did anything bind the spend?"

budget_ceiling: $2.00 (daily) actual_spend: $0.34 (this run) halt_decisions: none governor_state: WITHIN_BUDGET

Why this matters

When an audit, a customer security review, or a post-incident review lands, two questions fail first:

  • Who authorized the agent to do that?
  • Where is the receipt?

HUMMBL exists to make both questions answerable — not with a screenshot, but with a portable evidence packet that survives outside any vendor's UI.

This page shows an illustrative sample built from HUMMBL's own internal production primitives. It is not a customer deliverable. No fabricated customers.

For Operators

Find your governance gaps

If your current evidence is a dashboard screenshot, it will not survive an audit. A readiness audit shows you what should exist when your agent acts and what is missing today.

Book an architecture review →
For Builders

Produce your own receipts

The evidence packet format is not a black box. It is built from receipts your agent stack can produce locally. Install the primitives and produce your first evidence receipt this week.

Read the builder docs →

Make governance survive contact with action

A dashboard is a window. An evidence packet is a receipt. Start producing yours.