{"id":"data2-cohort-retention-math-steps","audience":"analysts","category":"data","track":"data","technique":"verification","level":"advanced","title":"Show cohort retention math step by step","free":false,"task":"Your growth lead wants the signup cohort retention table checked, plus the blended month-1 number and the blended month-2 number for the cohorts that have it. Retention is retained users divided by cohort size, rounded to a whole percent; the blended numbers are weighted by cohort size (total retained over total users) with one decimal. Write one plain-text line per available cohort month like 'Jan M1: 260 / 400 = 65%', then two final lines labeled Weighted M1: and Weighted M2: showing the sums. Under 130 words, no commentary, and skip months that are not available yet.","materialLabel":"Cohort counts (attached to your prompt automatically)","material":"Retention by signup cohort - Tallyroom (retained = still active in that month after signup)\n\nJan cohort: 400 users after removing 25 fraud accounts (the raw count was 425).\n  Month 1: 260 retained\n  Month 2: 180 retained\n  Month 3: 140 retained\nFeb cohort: 500 users.\n  Month 1: 330 retained\n  Month 2: 215 retained\n  Month 3: not available yet\nMar cohort: 300 users.\n  Month 1: 150 retained\n  Month 2: not available yet\n  Month 3: not available yet\n\nNote: the Mar cohort ran a smaller signup promotion, so it is smaller.","rules":[{"type":"plainText"},{"type":"maxWords","max":130}],"mistakes":[{"title":"Averaging the percentages","why":"The AI averages 65, 66 and 50 and gets 60.3% instead of weighting by size.","fix":"Add: \"Weighted by cohort size: total retained divided by total users.\"","checks":["r0"]},{"title":"Hiding the arithmetic","why":"The AI jumps to percentages and some come out wrong.","fix":"Add: \"Write each line like Jan M1: 260 / 400 = 65%.\"","checks":["r1"]},{"title":"Padding with unavailable months","why":"The AI fills Feb M3 and Mar M2 with guesses and adds insights.","fix":"Add: \"Skip months marked not available. No commentary.\"","checks":["r2"]},{"title":"Leaving the format to chance","why":"The AI picks its own length and layout unless told the exact shape.","fix":"Add: \"Plain text, no bold or headings.\"","checks":["plainText"]},{"title":"Trusting the AI to count words","why":"AI is bad at counting. Ask for a total word limit and it often lands over it.","fix":"Cap each part instead, for example \"each line under 15 words.\"","checks":["maxWords"]}],"promptHints":[{"label":"Shows the line shape","patterns":["jan m1: ?260","260 ?/ ?400","like jan"]},{"label":"Asks for a weighted blend","patterns":["weighted","by (cohort )?size","sum of"]},{"label":"Uses the cleaned Jan size","patterns":["400","fraud","425"]},{"label":"Skips missing months","patterns":["not available","skip","only (months )?with data"]},{"label":"Rules out commentary","patterns":["no (notes|commentary)","only these lines","nothing else"]}],"hint":"Haiku should not do silent math. Make it write each division, and tell it which weighting to use.","mission":"Do the retention math for these cohorts.","musts":["Show each division","Weighted month-1 line","Skip unavailable months"],"rubricLabels":["The cohort lines show the division and the right percentages: Jan M1 260 / 400 = 65%, Jan M2 180 / 400 = 45%, Jan M3 140 / 400 = 35%, Feb M1 330 / 500 = 66%, Feb M2 215 / 500 = 43%, Mar M1 150 / 300 = 50%","The last two lines show Weighted M1 as 740 / 1,200 = 61.7% (260 + 330 + 150 over 400 + 500 + 300) and Weighted M2 as 395 / 900 = 43.9% (180 + 215 over 400 + 500), not averages of percentages","It has no lines for months that are not available yet (Feb M3, Mar M2, Mar M3), no raw fraud count of 425, and no commentary or recommendations"]}