Daily operation
Guardrail Policy
The versioned, seller-owned limits that bound every recommendation and execution. Nothing the engine does can fall outside them.
A Guardrail Policy is the set of limits you put around the engine. It is versioned, visible, and auditable — and it is enforced in code. The engine cannot propose, let alone execute, a change that falls outside your policy.
What you can control
- Maximum bid increase / decrease per action.
- Maximum budget increase / decrease per action.
- Daily spend movement cap across the whole run.
- Maximum actions per run.
- Protected brand terms, protected campaigns, and protected ASINs — things the engine must never touch.
- No-pause and no-negative lookback windows — protect recent converters from being paused or negated too soon.
- Minimum confidence threshold — ignore recommendations below your bar.
- For auto-approval: which action types are eligible and the confidence threshold they must clear.
How guardrails are applied
Every action in a Manifest is checked against your policy. An action can pass, be clamped to your cap (for example, a proposed bid jump trimmed down to your maximum), or be blocked entirely. Blocked and clamped actions are shown to you with the reason — nothing is silently dropped or silently applied.
Versioning and audit
Each run records which Guardrail Policy version it used. When you change your policy, the new version applies to future runs, and the history is preserved so you can always see which rules were in force for any past execution.
A safe way to experiment
Because Shadow Mode runs the full pipeline without executing, it's the right place to test a new Guardrail Policy. Adjust the limits, run analysis, and see how the Manifest changes — all with zero risk to your account.