{"id":"hr-manager-message-policy-check-decision","audience":"managers","category":"people-work","track":"people","level":"advanced","technique":"verification","title":"Check a manager's draft against the policy before it goes out","free":false,"task":"Review the manager's draft reply to an employee against the policy excerpt. Output exactly three lines: line 1 starts \"Verdict:\" and is one of Send as is, Edit first, Hold for HR. Line 2 starts \"Issues:\" and lists each mismatch with the policy section number in under 40 words (or \"None\"). Line 3 starts \"Fix:\" with one sentence. Plain text. You are checking consistency only; HR makes the final call.","materialLabel":"Policy excerpt and draft (attached to your prompt automatically)","material":"POLICY EXCERPT\n7.1 Bereavement leave: Up to 3 paid days for immediate family (spouse, partner, child, parent, sibling). Up to 1 paid day for other relatives. Days must be taken within 14 days of the date of death.\n7.2 Employees may request extra unpaid time; requests go to the People Team, not the manager.\n7.3 No documentation is required for the first 3 days.\n\nEMPLOYEE (Sam Idowu, Nov 4): \"My grandmother died on Oct 30. Can I take some days off?\"\n\nMANAGER DRAFT (Kim Alvarez): \"So sorry Sam. Take the rest of this week and next week too, whatever you need, I'll approve it. Please send me the death certificate when you can. If you need more time we can sort it out between us.\"\n\nNote: Kim is a kind manager and Sam is a strong performer; neither fact matters for the policy check.","rules":[{"type":"maxWords","max":110},{"type":"plainText"}],"mistakes":[{"title":"Approving the draft","why":"The AI says the kind message is fine.","fix":"Say: \"Check each statement against the numbered policy.\"","checks":["r2"]},{"title":"No section numbers","why":"The AI describes problems vaguely.","fix":"Say: \"Cite the policy section number for each mismatch.\"","checks":["r1"]},{"title":"Extra commentary","why":"The AI writes a long review with praise.","fix":"Say: \"Exactly three lines: Verdict:, Issues:, Fix:.\"","checks":["r0"]},{"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"]},{"title":"Markdown","why":"The AI bolds words or adds headings.","fix":"Say: \"Plain text, no bold, no headings.\"","checks":["plainText"]}],"promptHints":[{"label":"Three labeled lines","patterns":["verdict:","issues:","fix:"]},{"label":"Allowed verdicts","patterns":["send as is","edit first","hold for hr"]},{"label":"Section numbers","patterns":["section number","7\\.\\d"]},{"label":"Policy only","patterns":["policy","consistency"]},{"label":"Word cap","patterns":["under \\d+ words","under 40"]}],"hint":"Give the three line labels, the fixed verdict options and ask for section numbers.","mission":"Decide whether a manager's draft reply matches the policy.","musts":["Verdict from three options","Cite policy section numbers","Consistency check only"],"rubricLabels":["Has exactly three lines starting Verdict:, Issues: and Fix:, and the verdict is exactly one of the three allowed options","Issues cites the mismatches with section numbers: grandmother is outside 7.1 immediate family so up to 1 paid day, no documentation needed under 7.3, and extra unpaid time goes to the People Team under 7.2","The verdict is Edit first or Hold for HR (not Send as is), and the output makes no comment on Kim's or Sam's performance or character"]}