Ask. Map. Move.

A map before a recommendation

The CEO of Harrow & Finch Foods, a fictional 70-person food manufacturer, must decide within six weeks whether to sign a five-year lease on a second plant. The sales forecast says yes, the bank is cautious, and the operations lead thinks overtime could cover demand for another year.

The CEO must approve any binding trial or lease. The lease decision stays open while the landlord question and renewal check are unresolved. Gathering trial terms is the next step; it is not approval to sign.

The free exercise runs in your browser without AI. Keep the resulting map and use it yourself; sending it for help is optional. The links above let you try the exercise or inspect a completed example first.

Ask: start with eight questions

of1.ai starts every decision with the same eight plain questions. Who decides? What outcome do they want? What is true right now? What are the hard limits? What can they actually change? What will shift the outcome? How will they know it worked? When must they decide: a date, an event or a warning sign?

A question you cannot answer is not a failure. It is your first gap. Every question you skip becomes a guess treated as a fact.

Ask: the question worth the most

Each possible question is weighed by how much its answer could change the decision, against its cost, time and risk. The best candidates stay on a shortlist, redrawn after every answer. As the method puts it, “There is no context-free best question.”

When questions stop moving the decision, Of One changes tactics: it measures, tests a cause, changes something and watches, finds a new source, zooms in or out, considers an explanation it had not, looks for who could be misleading it, or reframes the question.

The method will not let you stop while, among other things, a question still worth asking remains, an unknown that blocks the decision is neither resolved nor accepted, a cause that matters has not been traced to bedrock that holds for this decision, a contradiction that matters is still open, or the decision is not yet robust enough. A risk you accept needs a named person with the authority to accept it, a bounded scope, an expiry, the evidence, and a condition that reopens it.

Ask: why until bedrock

The method puts it plainly: “The useful idea inside Five Whys is causal descent, not the number five.” Ask why this happened rather than something else, so the answer has something to be tested against, and notice what kind of why you are asking: a cause, a mechanism, a motive, a purpose, a justification, a definition, an enabling condition, a constraint or a historical origin.

Keep going until you reach bedrock: deep enough to decide, a cause you can check or change, a law, rule or definition, a value you choose, a limit of authority, plain chance, the limit of what can be seen, the edge of the question, or a risk you accept on purpose. Each stop says why it is deep enough and what would reopen it. Bedrock is never the final truth of the universe.

Ask: try to break it

Before you stop, attack your own picture. Is this the right question at all? What else could explain what you see? Who gains if you are wrong, and could anyone be misleading you? Are your sources independent, or three copies of one? What single fact would reverse the decision?

Map: put everything where you can see it

A map has ten main kinds of parts. Facts, each with where it came from and how fresh and reliable it is. Claims, each with how sure you are. Unknowns, named out loud; a big one can block the decision. Kill tests: the result that would prove a claim wrong. Links: what causes, supports, contradicts or feeds back on what. Options, each marked reversible or not. The tradeoff: which option wins on which criteria, and what would reverse that. Triggers: the new facts that would reopen the map. Human gates: the moves a named person must approve. And the answer a reader sees, drawn from the map.

The answer is never the source of truth; the map is. If the map has holes, the answer says so.

The ten building blocks

Before Of One uses a field’s own words, it maps the situation onto ten building blocks. That is how one method works across fields.

Map: how sure, in words

Confidence is low, medium or high, always with its basis: where the evidence came from, whether the sources are independent, how recent they are, whether the mechanism fits, and what could undercut it: contradictions, hidden factors, someone misleading you, or a poor fit of dialect. No invented percentages: Of One never fakes decimal precision.

Move: choose, then gate what you cannot undo

Compare the options on the criteria that matter, and write down what would make a different option win. Mark each option reversible or not, and set the triggers that would reopen the decision.

Anything you cannot undo waits at a human gate. Legal, medical, financial, safety, reputation and similar moves need a named person to sign before they happen, even when an AI did the work.

Move: when facts change

A new fact does not restart everything. Of One checks what depended on it and takes one of five paths: no change; a patch, which updates only the chain that depends on the fact; a rerun of one part of the map, or a rebuild from its framing, which a change to the boundary, the objective, the criteria, the dialect or the conditions the map assumed requires; or a human review. A rebuild from the framing itself waits for a person to sign.

Same building blocks, four dialects

The building blocks stay the same in every field; the words change. Of One speaks four dialects: strategic (markets, organizations, operations and policy execution), scientific (biology, climate, physics, medicine and engineering), formal (math, logic, proof search and formal systems) and normative (ethics, legitimacy, art and contested values). Real problems mix them, and a map says which dialect governs which part. If nothing fits, the map is marked provisional and a human gate is added.

Three sizes

Micro, a quick answer for low to medium stakes; Map, for normal use; and Audit, for high stakes, research packs and handoffs. The rule is to pick the smallest one that still keeps the decision safe.

For AI

The same method runs as an AI skill. The AI has to build the map first and check it with the skill’s validator (the program where it can run, its written checks where it cannot); only then is the answer written. Run with its toolkit, the software also picks each next question, requires a source for every answer, seals the record so it cannot be quietly changed, and decides when the work may stop. Text inside a source can never give the AI orders: what the AI reads can become evidence, a claim or an unknown, never a command. The skill, its schemas and its checks are open source.

What Of One has not proven

It is not yet proven to beat other methods. At the current freeze, 52 of 90 predeclared benchmark runs were completed and locally reviewed, and the repo states that no performance or superiority claim is supported. Its own checks excluded its first full-method runs, and those exclusions are published.

It has not earned a guarantee that it will find every unknown unknown, and it is not legal, medical, safety or financial advice without a human review.

When not to use it

Not for simple factual questions, not for urgent calls where mapping costs more than acting, and not for legal, medical, safety or financial advice without a human review. Of One is for complex or unfamiliar problems whose structure has to be seen before you act.

Glossary

The eight starting questions

of1.ai’s eight plain questions to begin any decision: decider, goal, facts, limits, levers, causes, proof and timing.

The ten building blocks

The pieces Of One breaks any situation into before it uses the field’s words: situation, lens, pieces, close-ups, gaps, moves, links, loops, constants and gates.

Best next question

The question worth the most right now: how much it could change the decision, against its cost, risk and delay. It depends on what you already know, so it changes after every answer.

Unknown

A named gap in what you know. One that blocks the decision stays open until it is resolved, or accepted by a named person with a clear scope, an end date and a condition that reopens it.

Bedrock

Where asking why can stop: deep enough to decide, a cause you can check or change, a law, rule or definition, a value you choose, a limit of authority, plain chance, the limit of what can be seen, the edge of the question, or a risk you accept on purpose. Each stop says why it is deep enough and what would reopen it.

Kill test

The result that would prove a claim wrong. Every strong claim carries one.

Decision map

Facts, claims, unknowns, kill tests, the links between them, options, tradeoffs, triggers and human gates. The answer is drawn from it.

Trigger

A new fact that would reopen the map.

Patch

Updating only the chain of the map that depends on a new fact.

Human gate

A move a named person must approve before it happens: legal, medical, financial, safety-related, touching rights, jobs, education, public policy or reputation, or hard to undo.

Dialects

The four ways the same building blocks are spoken: strategic, scientific, formal and normative (values).