Concepts

Evidence health

Before you judge anything, check whether the evidence can bear the comparison. Evidence health reports gaps and judges nothing.

Start every review with evidence health. It answers one question: is the evidence fit for the comparison you are about to make? It judges nothing. Each check names what it does to an analysis and what would close it, and an agent should quote its warnings before it quotes a number.

What it checks

  • Sync freshness and the warehouse coverage of the window. Days before the warehouse starts read as zero, which means "not held", not "no spend".
  • Ads with spend the history has not pulled. They are counted and never judged.
  • Past uploads missing from the library.
  • The share of spend on creatives with no Values. Untagged spend cannot be grouped by angle, persona or format.
  • Videos read from their poster only. No scene-level claim is supported for those, so a hook read from a single frame rests on that frame alone.
  • Draft Values on winners. A draft Value already counts as evidence, but a person has not approved it.
  • The goal each funnel is judged on, and how many creatives could be judged.
  • Outcomes Meta cannot see: qualified leads, revenue, margin.

Funnels, goals and verdicts

Ads Manager's "Results" column is a different event on every row, so Win & Scale never compares on it.

  • A Funnel is where an ad sends people: instant form, website, messaging, profile or traffic. Every verdict is made inside one funnel.
  • A Goal is the event a funnel is judged on, the deepest one it produced in volume. For history verdicts the order is purchase, then add to cart, then website leads; form leads for instant forms; conversations for messaging; link clicks for profile and traffic. Live ads are judged on the goal the operator set for the campaign, else for the account, else the event the ad set was bought for. Set it with meta_set_thresholds (PATCH /api/v1/thresholds).
  • A Verdict is winner, average, loser or unproven: one creative in one funnel, measured on its goal against same-market peers.

Purchases and ROAS are blank where the purchase pixel did not fire, and a report with no purchases shows no CPA or ROAS rather than a confident wrong number.

Rules that follow

  • Low delivery is inconclusive, never a loss. An ad that has not spent enough to expect results has no verdict.
  • Hook rate is not the goal. Hook rate (3-second views over impressions) comes from the daily warehouse, and a funnel is never judged on it.
  • Everything is observational. Meta chose who saw each ad, so say that a trait was associated with results, and use a Test to confirm the ones that matter.

Quote the warnings, then quote the numbers. "Spend on creatives with no Values is high" changes how much a persona or angle rollup can be trusted.

On this page