A protocol for learning before scale.

Decision rule

An experiment is complete when it changes the decision, not when the prototype looks finished.

Problem brief

Name the condition that must change before proposing an answer.

• Who experiences the constraint, and in which workflow?
• What workaround exists today, and what does it cost or prevent?
• Why might this question be newly testable now?
• Which facts are known, assumed, or still inaccessible?

Required output: A concise problem frame with a named owner, affected participant, boundary, and reason to learn now.

Falsifiable hypothesis

State the smallest claim that the experiment can support or weaken.

• If we change this condition, what observable behavior should follow?
• Which measurement would count as supporting evidence?
• Which result would break the thesis?
• What remains outside the experiment's scope?

Required output: A testable statement with proof criteria, kill criteria, and explicit limitations.

Experiment protocol

Build the smallest test that can change confidence in the core uncertainty.

• Which participant, system, or environment makes the test relevant?
• What is held constant, and what changes?
• How will observations and counter-evidence be recorded?
• Which safety, privacy, security, or policy constraints govern the test?

Required output: A bounded test with a method, owner, evidence plan, stopping rule, and review date.

Evidence dossier

Separate what happened from what the team hoped would happen.

• What was observed, by whom, and under which conditions?
• Which result was expected, ambiguous, or contradictory?
• Can another observer interpret or reproduce the result?
• Which dependency or limitation changes the conclusion?

Required output: A record of observations, sources, counter-evidence, limitations, and the confidence change they support.

Decision rubric

Choose the decision that the evidence supports. A more optimistic label does not improve the experiment.

Limitations

This protocol is a general operating method. It does not replace domain expertise, research ethics, safety review, security review, legal advice, privacy review, or a test plan designed for the specific environment.

Publication of the method is not evidence that Halceon has delivered a particular engagement, result, venture, investment, or commercial outcome.

Revision history