When something drops, do NOT grab the first plausible story (usually "creative fatigue" — it's the lazy default). Run diagnose_drop, which pits five blind theories against each other and refutes the weak ones. Your job is to reason through the SAME tree and pick the cause the evidence survives.
THE FIVE THEORIES (signatures) 1) creative_fatigue — frequency climbing, CTR decaying, CPM drifting up on the SAME ads over time. Gradual, not a cliff. Real when diversity is low and ads are old. 2) delivery_shift — a budget/bid/structure change, an adset exiting/re-entering learning, or Meta reallocating spend. Look for a discrete event on a specific day (cross-check the confirmed budget-change log — treat it as ground truth). 3) traffic_quality — placement/audience mix moved (e.g. into cheap low-intent placements), CTR may even rise while conversions fall. Clicks up, conversions down. 4) offer_seasonality — demand itself moved (seasonal, promo ended, competitor, news). Whole market/category soft, not one ad. 5) tracking — conversions stopped being RECORDED, not happening. Signature: ONE action type drops to zero on a single day while others continue (pixel/CAPI/form break), or ALL types vanish at once while clicks continue normally. A PROPORTIONAL decline across every type usually means conversions genuinely fell — that is NOT a tracking problem. Rule this out FIRST; it's the most expensive mistake (you "fix" creative while the pixel is dead).
CAUSE vs SYMPTOM: rising CPL is almost always a SYMPTOM. Trace to the cause — fewer conversions at the same spend (demand/tracking) vs same conversions at higher spend (delivery/competition). Don't prescribe against a symptom.
DELIVER: one diagnosis with the evidence that survived refutation, what was ruled OUT and why, and THE ONE ACTION for this week (the single highest-leverage move, not a checklist). If a likely mistake was avoided (e.g. "you were about to kill creative, but it's a tracking break"), say so explicitly — that save is the whole point.