From dry-run to live in 14 days.

Magistry's Brain has four modes, set on the Brain page: Off, Dry-run, Approval-only and Autopilot. Dry-run is safe; live is where results come from. The two weeks between them are how you earn the confidence to let reversible moves run on their own. This is the ramp we recommend.

The four modes

• Off: no analysis cycles run and no AI credits are used. Your campaigns keep running as they are.
• Dry-run: the Brain logs what it would do so you can judge the quality of its thinking. Nothing is dispatched, not even a move you approve.
• Approval-only: the Brain proposes moves across channels and nothing acts until you approve it. Each approved move is dispatched for real and its outcome is measured.
• Autopilot: reversible moves run on their own, within the autonomy level, domains, confidence floor and daily ad cap you set. Irreversible actions always wait for you.

Week one: dry-run and read

1. 1Day 1: connect the store and the accounts you want managed, then set the Brain to Dry-run.
2. 2Days 2 to 6: read the proposals every day. Would a good operator make these moves? Note the ones that are wrong, and why.
3. 3Days 3 to 6: adjust settings where a proposal shows a rule that is too strict or too loose.
4. 4Day 7: weekly review. When most proposals look obvious and a few look boring, move to Approval-only.

Forecast accuracy is scored even in Dry-run, so the Brain starts building a track record before it spends anything. The CS agent and the paid-search agent have their own setup wizards, which offer to start in dry-run: the agent classifies, drafts and decides, but never sends, refunds or cancels. Review what it would have done, adjust its caps, then switch it to Approval-only yourself.

Week two: approve, then hand over

In Approval-only every move waits for you, in the dashboard or in the mobile app. Approve the moves you agree with; each one dispatches and is measured. By the end of the week you are judging measured outcomes as well as proposals.

Switch to Autopilot when those outcomes hold up. Start at the Conservative autonomy level, where only small, very high-confidence reversible moves approve themselves, and widen to Moderate or Aggressive as the record grows. Choose which domains autopilot may act in, and set a daily cap for ads.

The ceilings that stay on

Autopilot runs inside per-store ceilings, whatever the autonomy level. These are the defaults:

store-wide dispatches    : 20 per day
price changes            : 10 per day
promotions               : 5 per day
catalog moves            : 10 per day
theme presets            : 1 per day
campaign budget changes  : 50 per day
campaign pauses          : 20 per day
CS refunds               : 25 per day
products moved to draft  : 35 per week

The weekly catalog limits (Max drafts / week, Max new products / week and Max concurrent optimisations) are editable under Settings, in Worker config. Leave the daily ceilings alone until you have a few weeks of clean outcomes. The flip itself should feel anticlimactic, and that is the point of the two weeks.