Documentation

The goal object

Everything starts here. A goal carries an objective, a horizon, a budget, a target metric, hard limits, markets, products, a segment, an attribution model and a financial objective. Nothing downstream ranks against a goal that has not been written down, and changing the attribution model on a goal changes how credit is distributed everywhere that goal is used.

The four truth layers

Network truth is what the ad platform reports about itself. Attributed truth is independent journey attribution under a rule you chose. CRM truth is what your sales team did with the lead. Financial truth is collected revenue, refunds and margin. They are never averaged, because the distance between them is usually the finding.

• Network: spend, delivery, platform-attributed conversions
• Attributed: which touches preceded the outcome
• CRM: qualified, progressed, won or rejected
• Financial: collected, refunded, recognised, margin

Campaign intent

The canonical internal object that separates what you asked for from how each network implements it. Adapters translate one intent into the structures Google, Meta, TikTok, LinkedIn and Microsoft each accept. Network APIs change; the intent does not.

The Campaign Constitution

Hard account limits evaluated before any action executes: spend ceilings, daily movement limits, max CAC, minimum margin, protected campaigns, market restrictions, approval requirements and minimum evidence thresholds. A rule always beats a recommendation score.

Decisions and actions

A decision records what was proposed, the evidence it rested on, the metric snapshot at the time, the confidence, the expected impact and every constitution check it passed. An action records what actually executed against which external object, with the before state kept so it can be reversed.

Decision clocks

Safety checks run in minutes, pacing hourly, optimisation daily, learning over days and strategy weekly. Fast data does not mean fast campaign changes — a budget moved every minute never leaves the learning phase, so optimisation waits for the evidence threshold even when the dashboard updated seconds ago.