Every change keeps its own receipt.
Each executed action writes one immutable record: what was done, which actor did it — a named person, or the Growth Brain under a named mode and scope grant — the evidence it relied on, the state before, the state after, the policy check result and the rollback path. Records are append-only, so a correction is a new entry referencing the old one rather than an overwrite. Filter by channel, campaign, actor, action type or date, then open an entry to see the figures exactly as the deterministic services returned them at that moment. Blocked and rejected actions are logged too, so the record also shows what the system chose not to do.
Before-state is stored, not inferred
The record keeps the actual prior values — budget, optimisation target, audience configuration, status — captured at write time. Rollback replays those stored values instead of reconstructing them, which still works after the network's own change history has aged out. Reverting a week-old change is therefore the same operation as reverting the last one.
The actor is a specific name
Entries never say 'automation'. They name the person, or the mode and the scope grant that permitted the write, along with the Constitution version in force. When a change is questioned in a Monday review, the question has an address.
Evidence is frozen with the entry
Metrics attached to a record are the values the attribution service, budget engine and business metrics service returned when the action ran, with their windows and the attribution rule named. A later restatement of a figure, or a change of attribution rule, does not quietly rewrite why a decision was made. Credited revenue stays labelled as credited, kept separate from any experiment result stored alongside it.
In this capability
- Append-only entries
- Stored before-state
- Policy result per action
- Per-entry rollback
Autopilot
The rest of Autopilot
Recommend Mode
The Growth Brain drafts each action with its evidence attached. Nothing reaches a network until a person approves it.
Guarded Mode
Writes execute inside the Constitution. Anything that breaches a limit stops and escalates, naming the clause that blocked it.
Autonomous Mode
The loop runs continuously inside hard limits. People review exceptions: escalations, rollbacks and limits repeatedly hit.
Campaign Constitution
One account-level object: spend ceilings, max CAC, minimum margin, protected campaigns, approvals and evidence thresholds.
Emergency Controls
Pause automation, pause a channel, pause all spend, revert the last change, lock a campaign. Each is scoped and logged.
One goal. Every channel. Real business truth.
Connect your ad accounts, say what a good customer is worth, and let the Growth Brain work the objective inside limits you set.
No ad spend commitment · Connect one channel to start · [email protected]