Problem
The composite ranking system has a storyBoost specifically designed to surface "the active story/manuscript" above auto-saved conversation chatter (per the module's own docstring: relevance alone rewards frequency, which buries quiet-but-true memory like an active creative work under repeated conversational echoes). But it only applies to results matching storyTerms — which defaults to an empty array. With no terms configured, storyBoost never applies to anything, for anyone, regardless of how relevant it would be to a user who's actively writing something.
This is a specific, well-targeted feature (not a generic toggle) that appears aimed exactly at users doing sustained creative writing alongside conversational use — and it's silently inert unless someone already knows to hand-populate storyTerms themselves.
What we could do
At minimum, document this clearly with a concrete example of good storyTerms values (e.g. a manuscript's title, character names, or a project codename). More usefully: since the plugin already knows what memory sources exist, consider whether storyTerms could be seeded semi-automatically — e.g. by detecting a large, frequently-updated single-source document from local file sync and suggesting/adding its distinguishing terms automatically.
How we'd test it
Set up a corpus with real creative-writing content interleaved with ordinary conversation chatter. Compare search/auto-context results with storyTerms empty (today's default) versus populated with real terms from that work, and confirm the boosted config actually surfaces the creative content over chatter in cases where relevance alone would have buried it.
Problem
The composite ranking system has a
storyBoostspecifically designed to surface "the active story/manuscript" above auto-saved conversation chatter (per the module's own docstring: relevance alone rewards frequency, which buries quiet-but-true memory like an active creative work under repeated conversational echoes). But it only applies to results matchingstoryTerms— which defaults to an empty array. With no terms configured,storyBoostnever applies to anything, for anyone, regardless of how relevant it would be to a user who's actively writing something.This is a specific, well-targeted feature (not a generic toggle) that appears aimed exactly at users doing sustained creative writing alongside conversational use — and it's silently inert unless someone already knows to hand-populate
storyTermsthemselves.What we could do
At minimum, document this clearly with a concrete example of good
storyTermsvalues (e.g. a manuscript's title, character names, or a project codename). More usefully: since the plugin already knows what memory sources exist, consider whetherstoryTermscould be seeded semi-automatically — e.g. by detecting a large, frequently-updated single-source document from local file sync and suggesting/adding its distinguishing terms automatically.How we'd test it
Set up a corpus with real creative-writing content interleaved with ordinary conversation chatter. Compare search/auto-context results with
storyTermsempty (today's default) versus populated with real terms from that work, and confirm the boosted config actually surfaces the creative content over chatter in cases where relevance alone would have buried it.