← back to the AI PM Portal
Find the failures before they find you

HAL Test — Guided Pre-mortem & Inversion

Named after HAL 9000: not because AI systems are bound to fail, but because "what would have to happen for exactly this to occur?" is the most honest question you can ask of any undertaking. Methodological basis: Gary Klein's pre-mortem technique (HBR 2007) and Charlie Munger's inversion principle — plus a Goodhart layer of its own that nobody else asks for. Everything stays in this browser window; nothing is saved or transmitted.

ProjectFailure pathsSafeguardsGoodhart checkResult

1 · What's it about?

Short and concrete — one sentence is enough, but more helps the quality of the next steps.

2 · How does this fail, guaranteed?

For each category: an honest answer to the leading question. Imagine the project has already failed — what was the reason? Multiple entries per category are allowed.

3 · Fold back every failure path

For each failure path collected: one concrete safeguard that defuses it before project start — not "we'll watch out for it," but a real, concrete step.

4 · The Goodhart Layer

The step most pre-mortems skip: can passing this test itself become a failure mode?

Question A

How would a team notice that it has only learned to pass this pre-mortem — without actually fixing the underlying risk?

Question B

Which metric from this project could be optimized without the real outcome improving?

5 · Result

Summary of all failure paths, safeguards, and the Goodhart check — as text to save or share.