Docs · Guides

From a first fit to a recurring decision

Shadow test, capped pilot, then a scheduled decision that reuses its economics and keeps costs predictable.

1 · Shadow test

Run P34 next to your current process for a few decision cycles. Act as you normally would, record what P34 would have bought, and compare both against what actually sold. Nothing is at risk but the fits.

2 · Capped pilot

Act on P34's plans inside a fixed budget — a share of your weekly buying — with a person approving each plan. Log every outcome: it is next cycle's history.

3 · A recurring decision

• Schedule it: one fit per decision cycle, with the current menu at T = 0 and the new sales appended.
• Keep the description and schema stable: repeat fits then reuse the compiled economics. Change the description only when the economics change.
• Pin the model: move to a new version deliberately, after comparing on your own data.
• Watch the ledger: GET /account/ledger gives the cost of every cycle; top up ahead of it.

4 · Scale with controls

Raise caps as realized results — not predicted ones — earn it. For a company-wide rollout with approval queues, audit and a governed pilot, see Enterprise.