ADR 0010 — Absent data stays absent

Status: Accepted

Context

OBSERVED: the current normaliser substitutes created/updated with the literal "2026-07-22", researchArea with "General Research", status with "draft", authorAgent with "unknown", version with "0.1".

Decision

No defaulting. Absent values are None in the model and null in output. Derived readings (e.g. StatusClass) are published alongside the verbatim source value, never instead of it.

Evidence

OBSERVED: 735 of 756 records carry the fabricated date and 742 of 756 the default research area. Any timeline or facet built on them is misleading, and a consumer cannot distinguish a real value from a substituted one.

Alternatives

Keep defaults for convenience — rejected: it makes the output untrustworthy, which is fatal for a research record.

Consequences

Views must handle missing data — e.g. "recent activity" is omitted from slice 1 rather than faked (I.2).

Reversibility

Low cost.