Twenty questions for any decision: the eight bones, the best next question, and when to stop

To make a hard decision with twenty questions, answer eight plain questions first (who decides, what they want, what is true, what cannot change, what can, what moves the result, how you will know, and when you must decide), then keep asking whichever open question could change your decision most for what it costs, and stop only when no question left is worth asking.

That is the Of One method in one sentence. Its public steps are Ask, Map, Move: ask the right questions, map what you find, make the best move. Until September 2026, of1.ai called these steps Prism, Map and Forge, and called the starting state “the fog.” The method did not change; only the names did.

This guide covers the Ask step: the eight bones, the best next question, and the stop rule. It also shows the same eight bones filled in for three very different decisions, so you can see that the shape stays the same while the words change.

Why twenty questions can find one thing in a million

Twenty Questions is an old parlor game. One player picks something, and the others ask up to twenty yes-or-no questions to guess it. Wikipedia traces it to at least the 1780s and notes that it became a radio panel show in 1946.

The game works because of halving. If each question cuts what is left in half, twenty questions can tell apart 2^20 things, which is 1,048,576. The first question leaves about 524,288 candidates. The tenth leaves about 1,000. The twentieth leaves one.

Information theory gives this a unit. A yes-or-no question whose two answers are equally likely carries one bit, the same amount of uncertainty as a fair coin flip. Twenty good questions, twenty bits, about a million possibilities narrowed to one.

Real decisions borrow the idea, not the number. Four things are different:

Answers are rarely a clean yes or no. “How many customers would switch?” has a range for an answer, and the range may be fuzzy.
Questions do not split things evenly. Some cut the possibilities by 90 percent. Some barely move them.
Questions cost different amounts. Asking a colleague takes a minute. Running a trial takes a month.
You rarely need to find one exact thing. You need to know enough to choose well, and enough to see what would change your mind.

So the lesson is not “ask twenty questions.” It is “ask the question that could change your decision most, for what it costs, and let each answer decide the next question.” The Of One repo is blunt about this: its question engine has a maximum number of rounds, but that limit is a point where a human reviews the work, “not an epistemic claim that a fixed number of questions is sufficient.”

The eight bones

Before you can ask a good question, you need a frame to ask it against. Of One’s public frame is eight plain questions. Together they are the eight bones, the skeleton every decision shares.

#	Bone	The plain question
1	Decider	Who makes the decision?
2	Goal	What outcome do they want?
3	Facts	What is true right now?
4	Limits	What are the hard limits?
5	Levers	What can they actually change?
6	Causes	What will shift the outcome?
7	Proof	How will you know it worked?
8	Timing	When must they decide: a date, an event or a warning sign?

The full method behind of1.ai has twelve framing questions and ten building blocks, and the eight bones are its plain-language entry point. You do not need the full set to start. You need an answer, or an honest “I don’t know,” for each of the eight.

The “I don’t know” answers matter most. In Of One, a gap is not something to paper over. It becomes a named unknown, and every unknown is a candidate for the next question.

Same bones, three fields

Of One’s premise is that every field shares this skeleton: the words change, the geometry does not. The repo builds that in with one skeleton and four dialects (strategic, scientific, formal, and normative), so a market, a lab result, a proof and a contested value all get mapped onto the same parts.

Here are three illustrative decisions, invented for this guide, with the same eight bones filled in.

Bone	A founder’s first hire	A scientist’s failed experiment	A city’s flood plan
Decider	The founder, who signs the offer	The lab lead, who decides whether to rerun or change the design	The city council, on the engineer’s advice
Goal	Ship the product’s next version without the founder writing all the code	Know whether the effect is real or was never there	Fewer homes flooded in the next big storm, at a cost voters accept
Facts	Six months of cash; two customers asking for features	The treated samples showed no difference from the controls	Which streets flooded in the last storm; what the current drains can carry
Limits	Cannot pay above a set salary; must start within 60 days	Cannot get more of the same sample batch; grant ends in a year	State rules on building in the floodplain; a fixed capital budget
Levers	Role scope, contractor versus employee, equity offered	Dose, sample size, measurement method, protocol	Raise a levee, buy out homes, enlarge drains, warn earlier
Causes	Hiring speed, candidate skill, how much the founder still has to review	A real null result, a bad reagent, too few samples, a measurement too coarse	Rainfall, river level, paving that sends water into streets
Proof	The next version ships on time with the founder coding less than half the time	A rerun with a positive control that works	Fewer flooded homes in the next comparable storm, per a count agreed in advance
Timing	Decide before the two customers’ renewal dates	Decide before the next grant report	Decide before the next budget vote, or sooner if a storm season forecast turns bad

Read across any row and the words have nothing in common. Read down any column and the shape is identical: someone decides, they want something, some things are true, some cannot move, some can, causes link them, proof tells you it worked, and time runs.

That shared shape is why the same method can work in all three rooms. The founder’s weakest bone might be Causes (why is shipping slow?). The scientist’s is almost certainly Causes too (why no effect?). The council’s weakest might be Proof, because nobody has agreed how to count success. Each weak bone becomes a question.

The best next question

Once the bones are filled in, you will have more open questions than time. Of One’s rule for choosing is simple to say: ask the question that could change your decision most, for its cost.

The repo’s question engine scores each candidate question on what it is likely to tell you and what it is likely to change about the decision, minus its cost, risk, delay and overlap with questions already asked. It also counts what a question makes possible later: some questions are worth asking only because they open the way to a better one.

Two lines from the repo are worth keeping close:

“There is no context-free best question.” The best question depends on what you already know.
The most informative question is not always the most useful one. A question can teach you a lot about something that would not change your choice.

The engine can compute these scores with probabilities. When the probabilities would be made up, the method switches to plain qualitative judgments instead of inventing numbers. On paper, low, medium and high work well.

Worked example (illustrative): the founder’s hire. Four candidate questions, scored by hand:

Candidate question	Could the answer change the decision?	Cost	Verdict
What salary do similar engineers ask for locally?	Low: the founder already has a hard ceiling	Low	Skip for now
Is shipping slow because of coding time, or because the founder reviews everything?	High: if it is review, a hire will not fix it	Low: one week of tracking time	Ask first
Would the two customers wait three more months?	Medium	Low: two phone calls	Ask second
Would a contractor do as well as an employee?	High	Medium: a paid two-week trial	Ask after the first answer

The second question wins because it could flip the whole decision and costs almost nothing. After it is answered, rescore the rest. If the answer is “review is the bottleneck,” the contractor question may drop to nothing, and a new question appears: how can the founder review less?

When to stop

Most people stop asking when they feel sure. Of One refuses to let feeling decide. Its stop gate blocks a finished answer while, among other things:

a question with positive value for its cost is still open;
an unknown that blocks the decision has not been resolved or formally accepted;
more information could still materially change the decision;
the required challenge checks have not been passed (see Kill tests).

Accepting a gap is allowed, but only on the record: a named person, a bounded scope, an expiry date and the conditions that would reopen it.

In practice, stop when no question left would change what you do or is worth its cost, the challenge checks are done, and every blocking unknown is resolved or accepted on the record. When you stop, write down what you still do not know and what new fact would make you look again.

Do it today

Write the decision in one sentence, with a date.
Answer the eight bones. Mark each answer sure, fairly sure, guessing, or do not know.
Turn every “guessing” and “do not know” into a question.
For each question, ask: could the answer change what I do? What does it cost to find out?
Ask the top one. Update the bones. Rescore.
Stop when no remaining question is worth its cost, the challenge checks are done and every blocking unknown is resolved or accepted; write down what would reopen the decision.

The /ask/ tool walks you through these steps one bone at a time and shows the best next question for your biggest gap. The full method is on the method page. Next, learn what to do with the answers in What goes on a decision map, and how to take a “why” all the way down in Why until bedrock.