{"id":"fpa-model-cell-audit-labels","audience":"fpa","category":"finance-work","track":"finance","technique":"examples","level":"intermediate","title":"Label five model formulas as Hardcode, Wrong range, Sign error or OK","free":false,"task":"You're reviewing a forecast model before it goes to the CFO. Write a prompt for exactly 5 lines, one per cell in order, formatted \"C12: Label - reason\" where Label is exactly Hardcode, Wrong range, Sign error or OK. Each reason is under 15 words and names the evidence. Judge only from the audit notes. Don't suggest a fix formula. Plain text, no intro.","materialLabel":"Model audit notes (attached to your prompt automatically)","material":"MODEL AUDIT NOTES - FY27 revenue model, tab 'Forecast' (cell, formula as written, what the model sheet shows)\n\nC12  =SUM(C2:C10)\n     Rows 2 to 11 are all revenue lines (11 lines). Row 12 is the total.\nD5   =C5*1.04\n     The growth assumption of 4% is in Assumptions!B3. D5 is the only cell that uses 1.04 directly.\nE20  =E18-E19\n     E18 is Gross sales. E19 is Refunds, stored as a positive number; refunds should reduce sales, so subtracting them is the intended treatment.\nF7   =F5+F6\n     F5 is Gross sales. F6 is Discounts, stored as a positive number, and they should reduce revenue.\nG3   =AVERAGE(G1:G12)\n     G1 to G12 are the twelve monthly ARR values for the year.\n\nNote from the model owner: 'All of these were checked last quarter.'","rules":[{"type":"plainText"},{"type":"numberedItems","count":5},{"type":"maxWords","max":130}],"promptHints":[{"label":"Gives the line format","patterns":["c12: label","1\\."]},{"label":"Restricts labels","patterns":["hardcode.*wrong range.*sign error"]},{"label":"Asks for evidence","patterns":["reason","evidence"]},{"label":"Bans fixes","patterns":["(do not|don'?t).*(formula|fix)"]},{"label":"Caps length","patterns":["\\d+\\s*words","under"]}],"mistakes":[{"title":"Trusting the owner","why":"The AI labels everything OK because it was 'checked'.","fix":"Add: \"Ignore the owner's claim. Judge from the notes. Exactly 5 numbered plain-text lines.\"","checks":["r0","numberedItems","plainText"]},{"title":"Free-form labels","why":"The AI writes 'Possible error' or 'Fine'.","fix":"Say: \"Label is exactly Hardcode, Wrong range, Sign error or OK.\"","checks":["r2"]},{"title":"Missing the sign issue","why":"The AI assumes adding discounts is fine.","fix":"Add: \"Reasons name the evidence, such as stored as positive numbers.\"","checks":["r1"]},{"title":"Writing fixes","why":"The AI adds replacement formulas.","fix":"Add: \"Do not write corrected formulas.\"","checks":["r2"]},{"title":"Trusting the AI to count words","why":"AI is bad at counting and runs past word limits.","fix":"Ask for fewer words than the real cap and keep each line short.","checks":["maxWords"]}],"hint":"A model audit needs exact labels and evidence. Give the four labels and ban fixes and trust in the owner.","rubricLabels":["Five lines in order C12, D5, E20, F7, G3 with labels Wrong range, Hardcode, OK, Sign error and OK","Reasons name the evidence (row 11 omitted from the sum; 4% typed into the formula instead of Assumptions!B3; refunds are subtracted; discounts are added; twelve monthly values averaged)","Every label is exactly one of the four allowed labels and no line proposes a replacement formula"]}