Subscribe

Lab Tests

TESTED
Lab Tests

We run ChatGPT, Claude and Copilot on fictional finance files with known answers, then publish the prompts, the evidence and what went wrong.

Known answersEvery file has an answer key.
Exact promptsCopy them and rerun.
Failures shownNot just the wins.
DatedRerun when tools change.
Illustration of an unmatched bank items list with caught traps ticked, a magnifying glass over two amounts, a transposed-digits pair marked suspected, and the headline 6 planted traps, which AI caught them, with ChatGPT 6/6 and Claude 6/6
Stress testChatGPTClaude

40 unmatched bank items, 6 planted traps: what did the AI catch?

We gave ChatGPT and Claude the same fictional bank file with a known answer key. Both caught all 6 planted traps and marked nothing as cleared.

Tie: 6 of 6 eachOctober 23, 2026
Illustration of a coded transactions list with review labels, two deposit rows wrongly coded to 4000 Fee revenue under a magnifying glass, a Stripe payout that should go to 1050 Stripe clearing, and scores of 40 of 60 for ChatGPT and Claude
Stress testChatGPTClaude

60 bank transactions, 20 judgment calls: did the AI guess or flag?

Tie: 40 of 60 eachOctober 25, 2026
Get each new Lab Test by email

One email when a test is published, with its test files.