assumption-audit

Assumption Audit

Every plan is a stack of beliefs the founder has stopped noticing they hold. This skill writes them down as statements that can be proved false, throws out the ones that do not matter, and orders what remains by hours to test.

The ordering is the whole trick. Ranked by importance, the biggest assumption goes first, it costs six weeks to test, and the founder therefore tests nothing and launches. Ranked by cost, something gets tested on Tuesday.

When to use

Early — while the plan is still cheap to change. Before bet-sizing commits a cap, and ideally before quarterly-planning writes the bet into goals.md.

Also when a plan has been described three times and the description keeps changing shape while the conclusion stays fixed. That is a plan built downward from a conclusion, and its assumptions were never examined.

Inputs

Read first — house rule 1:

• The plan, in whatever form exists
• metrics.md — the recorded numbers, which decide what is evidence and what is belief
• clients/ and pipeline.md — the same, for anything the plan assumes about demand
• decisions/ — assumptions the founder has already tested and forgotten they tested

Beliefs

• The assumptions that kill a plan are the ones the founder does not experience as assumptions. Anything they can state as a belief has already been examined once. The load-bearing ones are stated as facts, in passing, in the first paragraph — and they never appear on a list the founder writes unaided.
• If striking what is already recorded does not remove at least half the list, the list is not assumptions, it is the plan retyped one sentence at a time. Twenty items is a transcript. What survives is usually three, and the collapse from twenty to three is the audit.
• A statement that cannot be false is not a modest assumption; it is not an assumption at all. "The market wants this" survives every possible world, which is exactly why it feels solid and exactly why it is worth nothing.
• Reluctance is the ranking. Whichever assumption the founder is least willing to test is load-bearing, whatever they marked it — and the reasons the test would be unfair, unrepresentative, or premature are a more reliable signal than the mark, because they took effort to produce.