My numbered prompt hit all 10 rules. It still dropped a warning.

Sources and assumptions

• Anthropic, Prompting best practices (Claude Platform Docs), section 'Be clear and direct': Verbatim: 'Provide instructions as sequential steps using numbered lists or bullet points when the order or completeness of steps matters.' Also: 'Be specific about the desired output format and constraints.' (checked 2026-10-05)
• OpenAI API docs, Prompt guidance (model: GPT-5.5): Verbatim: 'GPT-5.5 interprets prompts in a literal and thorough manner, enabling specific, descriptive instructions when the product requires them.' Also: 'Describe the expected outcome, success criteria, allowed side effects, evidence rules, and output shape. Avoid step-by-step process guidance unless the exact path matters.' (checked 2026-10-05)

Assumptions:

• All sample notices are fictional (Dana ext 214, Priya Shah ext 309 and Marcus are invented names); no real personal data.
• All runs used GPT-5.5 via the OpenAI Chat Completions API with default settings (no system message, default reasoning effort and temperature), on 2026-10-05 US time (ran_at timestamps 2026-10-06 UTC). The result is from the API, not the ChatGPT app.
• Claude Sonnet 5.5 runs were attempted (test ids paragraph-1 and numbered-1), but the local claude runner failed in this research session and produced no output. The post must not claim any Claude result.
• Small sample: 2 runs per prompt version (10 runs in total). Results can vary between runs and models, so the slides should say 'in this test', not 'always'.
• The 'numbered list' here is a list of requirements for the output (success criteria), not process steps. That fits both vendors' guidance.
• Scoring by hand: short notice (6 requirements: exact date/times, subject line, under 100 words, parking left out, 'What to do:' line, ends with contact + extension): all 4 runs met 6/6. Move notice (10 requirements: subject under 8 words, every date/time exact, under 120 words, no parking, no apology, exactly 3 pack bullets, 'Deadline:' line, IT handles monitors/docks, signed 'Facilities Team', ends with contact + extension): all 4 runs met 10/10. The longest output (move-paragraph-1) is about 94 words.
• The dropped warning: only move-numbered-2 left out the original 'Do NOT pack your monitor or docking station'. move-numbered-1 and both paragraph runs kept a 'Do not pack' line, even though the paragraph prompt also only said 'make sure people know IT handles monitors and docking stations'. So the cause shown is narrow wording read literally, not the numbered format.