Skip to main content

Question Evolution Loop

What question would change a consequential decision—and what action would let reality answer it?

The Question Evolution Loop turns questioning into a learning controller. A question is valuable when it changes a consequential decision. It becomes wiser only when action returns evidence that improves, rewrites, or retires the next question.

BETTER QUESTION → BETTER DREAM → TESTED ACTION → BETTER ENGINEERING STANDARD

The question improves the imagined setpoint—the Dream. Engineering makes it testable. Comparable evidence may then raise the shared standard. A compelling Dream alone does not authorize a standard.

Use the loop after Diagnose a Problem when the visible friction still needs causal work. Use it directly when you already have a named beneficiary, setpoint, and pending decision but no measured variance yet.

Two Modes​

ModeStart only whenHonest immediate claim
ReactiveA measured, typed variance has been accepted by the human who owns the decisionThe variance changed a decision and a correction is checking
ProactiveA setpoint, beneficiary, and consequential pending decision are explicitUncertainty changed a decision and an experiment is pending

If neither entry condition is present, return to Purpose or name the missing decision context. Do not fill the gap with a question bank.

The Method​

ORIENT → FORMULATE → INCUBATE → CAPTURE → REFINE → COMMIT → TEST → EVOLVE OR KILL

1. Orient​

Name the mode, beneficiary, setpoint, pending decision, and the variance or uncertainty. If the problem is still only a symptom, diagnose it first.

2. Formulate​

Use this meditation arc to expose assumptions before selecting the controller question:

  1. Reality: What happened, and what did I infer?
  2. Purpose: Who should benefit, and what does good mean for them?
  3. Story or model: What belief makes the current question seem right?
  4. Upstream cause: What earlier decision, incentive, or boundary could produce this state?
  5. Agency: What can I influence, authorize, test, or stop?
  6. Consequence: What changes if the belief is true, false, or incomplete?
  7. Evolution: What should the next loop ask sooner?

This arc may produce many candidates. Select one controller question using four tests:

  • relevance to the beneficiary and setpoint;
  • expected leverage on the pending decision;
  • discriminating evidence or a falsifier;
  • the fastest safe, reversible learning opportunity.

3. Incubate​

Formulate one question and avoid forcing an answer. For an overnight practice, write it before sleep and introduce no new material before morning capture. When the decision cannot wait, use a short same-session pause: step away, breathe, walk, or sit without input long enough for the first answer to loosen.

Incubation is a mechanism for surfacing assumptions. It is not evidence.

4. Capture​

Record raw insight before messages, research, meetings, or edits introduce a fresh frame. Preserve the exact language, including contradictions and uncertainty. Then name the assumption that changed.

5. Refine​

Rewrite the question until its answer can change the pending decision. Remove embedded answers, false choices, compound clauses, and requests for validation. State what evidence would distinguish the leading interpretations.

6. Commit​

Write the decision that changes because of the question. A question that changes no decision is not ready for the controller: rewrite it or retire it.

7. Test​

Choose the smallest safe, reversible experiment that can produce the discriminating evidence. Freeze its proof signal, review point, and kill signal before action. The Decision Optimisation Engine can carry the decision into action, review, precedent, and later recall.

8. Evolve or Kill​

At review, compare the observation with the frozen expectation through Performance and Standards. Then:

  • retain the question when it changed action and remains useful;
  • rewrite it when evidence exposed a better assumption or discriminator;
  • retire it when it repeatedly produces no changed decision;
  • raise the setpoint only with human authority and supporting evidence.

Evolution owns the wider lifecycle: the surviving lesson must change the next intention, question, experiment, action, or control.

Question Controller Card​

Copy this card before incubation. Complete the action and proof fields before calling the loop closed.

Mode: reactive | proactive
Setpoint:
Beneficiary:
Pending decision:
Variance or uncertainty:
Controller question:
Assumption:
Evidence/falsifier:
Decision changed:
Next experiment:
Proof signal:
Review point:
Kill signal:
Next setpoint/question:
Disposition: retain | rewrite | retire | checking | no-change

Evidence Boundary​

Insight and a changed action are immediate receipts. A completed card, meditation, or launched experiment does not prove improvement. Use checking while awaiting the declared review point. Claim improved only when later comparable evidence meets the frozen meaningful difference.

Questions that repeatedly fail to change a decision consume attention without returning learning. Rewrite or retire them; do not preserve them because they sound profound.

Human authority retains values, consent, protected constraints, irreversible commitments, variance acceptance, and durable setpoint changes.

Failure Modes​

  • Question bank instead of control: volume hides the missing decision.
  • Meditation as proof: insight is treated as evidence of external improvement.
  • Answer forcing: incubation becomes rehearsal of the preferred answer.
  • Infinite deferral: questioning postpones a safe reversible commitment.
  • Post-hoc proof: the success signal moves after the experiment.
  • Immortal question: repeated no-action output is kept instead of rewritten or retired.

Context​

Questions​

Next question: What evidence could change this decision through the smallest safe action?

  • Which assumption changed during incubation?
  • At the review point, did this question change action, earn a rewrite, or deserve retirement?

Changes my mind: repeated use produces thoughtful controller cards but does not change a consequential decision, retire weak questions, or improve later comparable evidence.