Skip to content

Compare claims action

🔍 A Research action (#155). Given a note, surfaces which other notes' claims agree with or contradict its claims, over the knowledge model (#145). Deterministic and offline — no text-similarity AI.

How comparison works

  • Agreement — another note asserts the identical claim once normalized (trim, lower-case, collapse whitespace, strip surrounding quotes and trailing punctuation).
  • Contradiction — either:
  • Structural — the two notes are joined by a contradicts semantic relation (#147, either direction); the paired claims are flagged regardless of text.
  • Textual negation — the same proposition asserted with opposite polarity, over a fixed, closed bilingual marker set (not · no · never · n't · nunca · jamás · tampoco).

!!! note "Limitation" The textual rule catches clean polarity flips where the rest of the sentence is otherwise identical (e.g. "coffee is healthy" vs "coffee is not healthy"). Richer grammatical negation relies on the structural contradicts relation — declare it with the create semantic relation action for reliable results.

Options

  • Result property — where the comparison is written (default claimComparison).
  • Write to — Frontmatter or Context (also exposed as {{property}}).
  • Source note (optional) — the note to compare; empty ⇒ the note being built.

Result

A { agreeing: [...], contradicting: [...] } object — each a deduped, path-sorted list of [[wikilinks]] to the matching notes — plus a Notice with the two counts. Empty sets are a valid result. If the index isn't ready, the action safely no-ops.

Capabilities

File-system read + the disclosed result-property write. No network, no AI.