This is what the audits have been taught. Every rule is a correction from a real audit, compiled into that area's prompt at the start of the next run — there is no publish step, so a rule saved now is in force on the next audit.
Admins run audits, edit them and approve them. Standard users read finished audits and the evidence behind them, which is what a team member needs to send one to a student.
Every audit is six calls to the model: five areas, then a synthesis pass that assembles the document. Each has its own set of rules. The tag tells you which set your change will teach, so a fix to bio wording cannot quietly change how checkout pages get scored.
| Tag | Wrote that section | A change there teaches |
|---|---|---|
| synthesis | The final pass that assembles the document | How audits are written — roadmap phrasing, first impressions, wins, next steps |
| bio | The Instagram bio call | How bios are reviewed and rewritten |
| landing | The landing and sales page call | How pages are judged — headlines, proof, what renders |
| checkout | The checkout call | How checkout pages are judged |
| funnel | The offer, funnel and pricing call | How offers and prices are judged |
| hooks | The hooks call | How video hooks are written |
| per area | Several calls at once | Each row separately — the landing summary teaches landing, the bio summary teaches bio |
| evidence | Nothing — these are the captures | Nothing. Read-only, because evidence is what we saw rather than something we wrote |
There is a seventh set, core, injected into all six calls — tone, plain language, and the rule that a claim about a page must rest on a screenshot. Nothing here is tagged core on purpose: a correction that broad changes every area at once, so it should be a deliberate choice made when you turn feedback into a rule.