# Validity and Calibration

Coherence uses behavioral measurement and psychometric-style scales. It does not claim clinical psychometrics.

## Construct review

Before publication, answer:

1. What exact behavior/operational condition does the dimension measure?
2. Which observations support it, and which falsify it?
3. What opposite causes can produce similar behavior?
4. Which contexts, languages, cultures, Roles, power relationships, and access needs change interpretation?
5. How many independent observations and contexts are required for each consequence?
6. Does the method distinguish temporary state from persistent pattern?
7. What is the confidence ceiling and abstention rule?
8. Which uses are prohibited?

## Calibration record

For a fixed method/version, compare predicted score/confidence/response with independent or participant-adjudicated outcomes. Report error and agreement, confidence reliability, false intervention, false boundary, missed boundary, abstention quality, appeal/override, and context coverage. Use minimum cohorts and suppress re-identifying slices.

Calibration changes create a new dimension/method version. Existing assessments retain their original versions. A method is suspended from consequential use when it becomes overconfident, confounded, materially biased, under-covered, or unable to distinguish the construct.

Source correction is evidence, not ground truth by fiat; system inference is also not ground truth. Preserve disagreement and lower confidence when adjudication remains unresolved.