Define how prose binds to Facts: explicit markup vs. AI-detected links vs. coreference resolution, plus the confidence metadata on each link — the primitive that decides whether prose-as-a-view-over-data (Entry 029) actually works.
Why this matters
Entry 029 makes "prose is a view over the Fact graph" the differentiator; that only holds if there is a concrete, low-friction way for a sentence to reference a Fact. This is a sub-gap hiding inside G-057. Open decisions:
- Binding mechanism — explicit author markup (e.g.
[[fact:…]]), AI-detected links proposed for confirmation, or both; and how that surfaces in the editor (G-015) without feeling like tagging. - Coreference resolution — when prose says "Gandalf" vs. "the Grey Wizard," whether both resolve to one entity automatically (AI, G-046) or manually, and how ambiguity is disambiguated.
- Link confidence — distinguishing a confirmed binding from an inferred, high-confidence one; how that confidence is shown and how it feeds the review queue (G-060) and data quality (G-026).
- Propagation — when a bound Fact changes, how much of the citing prose updates automatically (the cascade Entry 029 promises; mirrors the LoreLine templating in Entry 010).
Related
- Entry 032 — the bridge strategy that surfaces this gap
- Entry 029 — prose-as-a-view-over-data (the thesis this enables)
- Entry 010 — LoreLines templating (a sibling propagation case)
- G-057 — prose-to-Fact authoring UX (this is its core primitive)
- G-046 — AI-proposed mappings (the coreference/extraction engine)
- G-026 — Fact-graph data quality (link-confidence feeds it)
- G-060 — Fact review queue (where proposed links get confirmed)