Writing calibrated confidence statements

Purpose and when to use it

Use confidence statements when moving from records to interpretations and conclusions. Write them separately for materially different claims; one overall label can hide a strong observation and a weak attribution.

Evidence required

Bring together claim-to-source references, method validation, conflicts, candidate alternatives, and coverage limits. Identify any numerical measure and the population and test conditions to which it applies.

Principles and procedure

• Write a bounded proposition and list its direct supporting observations.
• Assess source independence, validation relevance, contradictory evidence, and missing material.
• Use the project vocabulary only with an explanation of the support: observed, corroborated, inferred, attributed, unresolved, or contradicted.
• Use numbers only when an applicable validated method supports their stated meaning; otherwise use a transparent qualitative explanation.

What to document

Keep the reasoning from observations to each conclusion and state what further evidence would change it. Preserve separate confidence assessments where the source and content propositions differ.

Failure modes

An impressive-looking percentage can import benchmark assumptions that do not fit the examined object. A conclusion may also conceal alternatives if the report lists limitations only in an unrelated appendix.

Limits on the conclusion

Qualitative labels are descriptors, not a calibrated numerical scale. Confidence can change with new evidence and does not decide admissibility or identify a person by itself.

Independent checks and corroboration

Seek evidence capable of disproving the preferred interpretation. Independent support matters most when it addresses the actual weak link rather than repeats an already established observation.

How to report the result

Write the finding, its basis, its uncertainty, and its decision boundary together. Avoid very likely or proven without explaining the evidentiary meaning intended.

Related methods

• The complete evaluation method
• Seeking independent corroboration
• Reporting synthetic-evidence findings

Related glossary terms

• confidence statement
• competing hypothesis

How do you write uncertainty clearly?

State the proposition first, then the observations that support it, the independent checks, and the strongest reasonable alternatives. Describe unavailable evidence and how it affects the result. Use plain words tied to these conditions.

For example, say that an acquired file matches a recorded digest while its creator remains unresolved. That is more informative than applying one unexplained confidence percentage to the entire investigation.