The Canon of Contingency

1. Before the story, count what happened.
2. Before the exception, seek the class to which it may belong.
3. The first number is not the truth. It is the place from which honest revision begins.
4. A past event is evidence, not a commandment laid upon the future.
5. One memory speaks loudly. A record lets quieter events return to the room.
6. Where cases are few, confidence shall be small, even when desire is great.
7. Choose comparisons for resemblance, not for the comfort of their conclusion.
8. The wise prior bends toward new evidence without pretending it began there.

Commentary: a prior is an initial probability informed by relevant past cases. It should be explicit enough that another person can question the reference class and recalculate it.

1. A signal earns its name by changing expectation, not by attracting attention.
2. Noise arrives dressed as urgency, coincidence, elegance, and fear.
3. Record the absence of an expected sign, for silence may also update the field.
4. Do not count the same evidence twice because it entered through two doors.
5. Separate what you observed from what you inferred and from what you wished.
6. A precise measurement of the wrong thing remains wrong with greater precision.
7. When a signal appears only after the outcome, it may explain, but it did not forecast.
8. Let every strong signal carry the question: under what conditions would it fail?

Commentary: signals are observations that deserve to alter a probability. Correlated evidence, hindsight, selective memory, and vague definitions can make noise appear stronger than it is.

1. The future is not a road waiting to be found, but a field of paths receiving weight.
2. Name more than one outcome, lest imagination disguise itself as inevitability.
3. A probability is a measure of uncertainty, not a fraction of destiny already written.
4. Near branches may be forecast more clearly than distant branches, yet even the near can break.
5. When conditions change, do not drag the old forecast into the new world.
6. The unlikely is not the impossible, and the likely is not the promised.
7. A branch neglected in thought may still arrive in fact.
8. Hold the map lightly, attend to the terrain, and keep the unmarked horizon visible.

Commentary: branch language protects against false certainty. A good forecast includes alternatives, time horizon, assumptions, and a rule for deciding which branch occurred.

1. To revise is not weakness. It is fidelity to a world that has continued moving.
2. When evidence changes, let the number change before the explanation hardens.
3. Do not hide yesterday's forecast when today reveals its error.
4. To update is not to betray the past. It is to stop demanding that the future obey it.
5. Confess overconfidence in numbers, not in theatrical regret.
6. A model that cannot lose has never entered the court of evidence.
7. Keep the original record, the update, and the reason, so learning has a trace.
8. The last revision is never announced while the question remains open.

Commentary: revision should be timestamped and justified. A forecasting record becomes useful only when old probabilities remain visible beside new ones and final outcomes.

1. The forecast of an unchanged system is not a sentence imposed upon it.
2. Name the lever, the moment, the actor, and the cost before calling hope a plan.
3. When the cue appears, let the chosen action already be waiting.
4. An intention without a path may predict desire more than behavior.
5. Reduce friction where virtue repeatedly fails to cross it.
6. Change the environment when willpower has become an unreliable witness.
7. Record the passive branch and the active branch separately, so agency can be measured.
8. If the intervention fails, examine its design before condemning the person.

Commentary: an intervention is a specified change to the system. It may include an if-then plan, environmental redesign, resources, accountability, or removal of a barrier.

1. A crowd may show a pattern that no member intended.
2. Large numbers quiet some accidents and reveal rates hidden inside individual noise.
3. The person may remain difficult to predict while the population becomes increasingly legible.
4. Institutions remember through rules, incentives, archives, routines, and repeated delay.
5. Small changes may cross a threshold and reorganize the whole system.
6. A forecast announced may enter the system and alter the outcome it describes.
7. Seek the leading signal before the visible event and the pressure before the break.
8. Collective foresight turns history from an archive into an instrument.

Commentary: collective patterns emerge from scale, incentives, networks, thresholds, and feedback. Their usefulness depends on stable definitions, enough observations, and careful separation of signal from coincidence.

1. Every model has a horizon, and every horizon can be moved.
2. Beyond the present model lie variables not yet measured, relations not yet understood, and events too rare to estimate well.
3. The unknown is not an argument against inquiry. It is the next field of inquiry.
4. Count the tail as well as the center, for rare events can redirect the whole system.
5. When the regime changes, yesterday's regularities may become tomorrow's errors.
6. More data can reveal a pattern, but more data of the wrong kind only hardens confusion.
7. More compute can search farther, but it cannot repair a question without a clear outcome.
8. At every horizon write: the forecast is incomplete, and still it may guide action.

Commentary: uncertainty includes missing variables, model error, distribution shift, rare events, and genuine randomness. The boundary of a model is not the end of the method. It marks the next question.