the-repeat-purchase-check
Ask the product's actual consumption cycle before diagnosing timing. A category inference is a guess, and the whole diagnosis inverts on it: a reorder prompt that is late for a 30-day consumable is early for a 90-day one, and recommending the wrong direction is worse than recommending nothing. Ask how long a unit lasts in normal use, or derive it from the observed gap between first and second orders among customers who did reorder. Where neither is available, say the timing finding is unavailable and diagnose only the parts that do not depend on it.
The Repeat Purchase Check
Take the flows a customer sees after their first order and find the specific gap, timing or message, that's stopping the second one.
Input integrity. Run the checks in
references/data-input-integrity.mdbefore computing anything, and report what they found. Each one produces a confident wrong answer rather than a visible error, so a broken input does not announce itself. Recent cohorts have not had time to repeat, so they always look worse. Do not read an immature cohort as a decline. Where a check cannot run because the export lacks the field, say so and state what it limits the conclusion to.