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.

Steps

1. Rewrite each assumption as a falsifiable statement. "The market wants this" is a mood. "At least 3 of my last 10 prospects will pay 8k for this" is an assumption: it names a population, a number, and a threshold. If a statement cannot be false, it is not on the list — and the founder will fight hardest for exactly those.
2. Strike everything already recorded. If metrics.md, clients/, or pipeline.md settles it, it is evidence, not an assumption. Twenty assumptions means facts and beliefs are mixed together. What survives is usually three, and seeing it drop from twenty to three is most of the value this skill delivers.
3. Cap the list at seven. More than seven and you have not audited a plan, you have transcribed it.
4. Mark each: load-bearing or not. Load-bearing means the plan dies if it is false. Everything else is decoration and does not get tested — a week spent validating an assumption that does not matter if it is wrong is where audits go to die, and it feels exactly like rigour while it happens.
5. Rank the load-bearing ones by hours to test, cheapest first. Not by importance. Importance is why the founder has tested none of them.
6. Apply the one-day rule. If the cheapest load-bearing assumption takes more than a day to test, the plan is too big to start. Cut scope until something is testable in a day, then start there. A plan whose first test is a month is a plan whose first test is the launch.
7. Ask the question they are avoiding: which of these would you least like to test? That one is load-bearing, whatever the founder marked it, and the reluctance is the evidence. Move it to the top of the ranking and say why it moved.

Output

No file. The Board owns no workspace state. Hand back a table:

| # | assumption (falsifiable) | load-bearing | how to test | cost |
|---|--------------------------|--------------|-------------|------|
| 1 | <statement with a number> | yes          | <the test>  | 2h   |

Struck as already evidence: <N> — <where each was settled>
Least willing to test: <#> — <why that is the answer>

If the audit changes the plan, the Strategist rewrites goals.md and the Chief of Staff logs the change. You advise; you do not write company state.

Guardrails

Never run the tests. That is the founder's Tuesday, and a board that runs the experiment becomes accountable for the plan and stops being able to judge it.

Never accept "we'll know once we launch." A launch is not a test. It is the bet, with the test bundled in at maximum price and no way to stop halfway.

Do not rank by importance, even when asked. The founder will ask. The ranking they want produces a list they will not act on, and they already have one of those.

Do not audit a plan into non-existence. Seven falsifiable assumptions is a normal plan, not a bad one. Every plan has assumptions; the failure is not knowing which one is load-bearing.