Skip to content

Evidence map (compound thinking)

Experimental. The ambitious end-state: ask "what do you think about X?" and get your system's answer — not a generic model's.

The evidence map is a transparent synthesis of the active note, reconstructed only from your graph: which ideas support it, which contradict it, what evidence exists, and what gaps remain. Every row links to the exact source note. It is not "an AI's opinion" — it's a view of your knowledge, with no invented content.

Opening it

Run "Show evidence map" from the command palette, or click Open next to Evidence map in Settings → ZettelFlow → Zettelkasten toolkit. It follows the active note and updates (debounced) as you switch notes or edit.

The four buckets

For the active note, buildEvidenceMap composes existing graph primitives:

  • Supports — the notes connected to it by a supports relation (#147), in either direction (the ideas backing the position).
  • Contradicts — its contradicts partners (the #153 findContradictions, in + out).

  • Evidence — the sourced claims (#148) on the note and its supporting notes: each entry is a real claim grounded to a real note and a real source ([[link]] or a citation). Only sourced claims appear here — that's the "no unsourced claims" rule.

    Where the claims come from now (#582). Until #561 this map was empty in every vault, because a claim could only be written by hand-editing YAML — and the first version of that door wrote the claim without its source, which landed every new claim straight in Gaps below. The claim door now offers the source on the same line, so an idea arrives here grounded rather than thin.

  • Gaps — where the position is thin: the note's unsourced claims (claims with no evidence) and its open questions (the #153 findUnansweredQuestions).

Every emitted note path exists in the model; the map is deterministic, read-only and never invents content. An unindexed note yields an empty map.

AI is out of scope (for now)

This ships as the deterministic, AI-free evidence map — it works entirely offline. Grounded natural-language generation over this map (the "write me the synthesis" step) is a deliberate follow-up that will compose with the optional AI actions; the map itself never needs it.

Architecture

buildEvidenceMap(model, path)                    (pure, Obsidian-free, unit-tested)
  → { focus, supports, contradicts, evidence: [{ note, claim, source }],
      gaps: { unsourcedClaims, openQuestions } }
  reuses findContradictions (#153), findUnansweredQuestions (#153), the #148 claims/sources

EvidenceMapView (ItemView) + EvidenceMapComponent (show-evidence-map command, no hotkey)
  reads the KnowledgeIndex model for the active note → buildEvidenceMap
  four sections, every note row opens via workspace.openLinkText; writes nothing