Obligatory-attribute misses are machine-readable gap signals in knowledge graphs

knowledge-graphgap-detectionknowledge-managementepistemicsPKM

In any knowledge graph, entities belonging to a class share predicates that all members of that class are expected to carry. When an instance lacks an expected predicate, that absence is a detectable incompleteness signal — not an ambiguous open-world unknown, but a machine-readable structural flag. Razniewski et al. describe this as the "obligatory attributes" mechanism: the method involves analyzing "the ratio of instances of the class that have the predicate" and checking "if this ratio changes when we go into subclasses," revealing incompleteness when entities lack expected properties.

The mechanism sidesteps the open-world assumption problem. In classical knowledge-base semantics, a missing triple might mean the entity lacks the property or that the property exists but isn't recorded. Obligatory-attribute analysis resolves this ambiguity statistically: if 95% of entities in a class carry a predicate, the 5% that don't are more likely to be incomplete records than exceptional members. (Audit clarification 2026-09-11: the 95%/5% sentence is this note's illustration, not the paper's criterion. Razniewski et al. define an obligatory attribute as "a predicate that all instances of the class must have in the real world at the current point in time" and test for it by whether the ratio of instances carrying the predicate stays the same across subclasses and intersecting classes — "If the ratio changes, then a theorem tells us (under a number of assumptions) that the predicate cannot be obligatory." Once a predicate is obligatory, an instance lacking it "is necessarily incomplete on that attribute" — a categorical inference, not a probabilistic one. Reported performance on YAGO and Wikidata: "a precision of 80% at a recall of 40% over all predicates.")

Applied to SeekVault: every claim-note belongs to the class claim-note, which carries obligatory predicates — source_url, source_tier, source_quote, and date_created. Any note in 30-notes/ missing one of these fields is, by this formalism, a machine-readable gap signal. The vault's YAML frontmatter spec (§7 of the operating spec) functions as the class definition; schema violations are incompleteness markers detectable by a simple lint pass without requiring any judgment about content.

This makes structural gap detection cheap: a single pass over 30-notes/*.md for missing frontmatter keys reveals a list of notes needing sourcing attention. No LLM reasoning required. The obligatory-attribute method is suited to the first pass in a gap-detection sequence, with more expensive methods — behavioral analysis (claim-query-failure-clustering-as-gap-signal), LLM corpus scanning (claim-llm-explicit-implicit-gap-detection) — handling the gaps it doesn't catch.

See question-gap-detection for the broader detection problem this mechanism partially answers. Razniewski et al. describe additional completeness tools in the same paper (cardinality assertions, No-Change Assumption, text-extraction-based recall) that could yield further vault-applicable mechanisms on a dedicated run.