This is the complete pack behind my ChatGPT vs Copilot steering committee test: everything you need to run the same test on Claude, Gemini, or whatever your work gives you.
The project, Kestrel, is a fictional 3,800-user document migration built to be as close to real life as I can publish. Ten accuracy traps are seeded across the files: a budget figure corrected four hours after it was stated, a stale cutover date, a disputed RAG status, a contract deadline that lives mainly in one email.
What's in the pack (H2)
The exact prompt I used, the ten source files, the locked committee deck template, the scoring sheet, and the answer key to all ten traps.
How to run it (H2)
Then do the one check that matters most: look for anything in the output that appears in none of the ten files. A name, a date, a job title. A tool signed into your work account can pull details from your profile into your deliverables without warning you.
Read the full test and scores:
- Give your AI tool all ten files and the prompt, verbatim.
- Ask for the status report first, then have it populate the deck template from that report. The chain is the point: anything wrong early stays wrong.
- Score with the sheet before you open the answer key.
Subscribers get the template
Enter the email you use for the Cliffinkent newsletter. If it is confirmed, the worksheet opens here.
No account. No download portal. Just a subscriber check.