Experiment formalisation

Formalise only when a named consumer needs a question the current records cannot answer. Default to a bounded annotation layer over authoritative typed records.

Work from the affected contract

1. Run formalisation triage: name the consumer and Must questions, existing answers, cost, owner and stop condition. An evidenced decision not to formalise is a successful outcome.
2. Inventory current records and draft queries before choosing terms. Verify vocabulary against versioned primary sources; distinguish verified, unverified, refuted and locally minted terms.
3. Map roles, factors, controls, pairing, measurement/absence and lineage without changing runtime truth. For this annotation workflow, typed contracts remain authoritative and a sync check detects drift.
4. Use Tier 1 unless stable grammar, a real query consumer, maintenance ownership and unmet Tier-1 questions justify graph/reasoner machinery.
5. Validate query positive/negative controls, shape/invariant applicability and synchronization with actual emitted records. Record source revisions, gaps and unsupported mappings.

Scope and completion

Deliver competency questions, verified mapping, extension/gap register, necessary context/queries and check evidence. Do not add ontology machinery to solve a contract-enforcement or statistical-inference defect. Never convert an unresolved vocabulary claim into an authoritative ontology assertion.

Use the user’s existing intent and authorization. Ask only for missing information that materially changes the result; use additional reviewers when they address a concrete uncertainty. Treat unavailable checks as gaps rather than successful verification.

Focused references