Blueprint kit: test suite v2.4

37 scenarios sent through kit v1.4's real prompts: the original 20, 12 harder ones, and 5 for phone apps. Each answer is graded by a separate AI call against a pass rule written before testing. Security reviews are scored against hidden answer keys for 4 flawed apps, and two clean apps check for false alarms. A scenario only passes if every run passes.

This uses your own Claude usage. 37 tests at 3 runs each is about 220 calls and can take an hour or two. Start with 1 run, or one group, to check everything works.
–scenarios passed every run
–individual runs passed
–hidden flaws found (all runs)
–false alarms on the clean app

Results to publish

Everything from this session: every raw answer, every grade and the kit version. Copy it into the tests repository's results folder, or paste it back into the chat.

Prompt text © 2026 Dwayne Brown. It's part of the Blueprint Kit: you can run these tests and quote the results, but you can't copy or redistribute the prompts. Read the license · Get the kit