Status: backlog — captured, not started. Doc branch docs/suggestion-ranking-idea.
A word-level ranking prior that biases the suggestion strip and live-converge toward higher-ranked words (higher rank → higher chance of being chosen). Complementary to, and separate from, the touch-level adaptive-typing feature (#ADAPTIVE).
Full design: docs/SUGGESTION_RANKING.md.
Summary
- Single re-rank hook in
Suggest.kt#getSuggestedWordsForBatchInput covers both the strip and live-converge (live-converge reuses that pipeline and takes getWord(0)).
- Rank source options: amplify dictionary frequency / user-history boost / explicit weight table / learned — TBD.
- Guardrails: deterministic boost (not random), capped so it only breaks near-ties and never overrides a clearly better shape match, additive bonus to stay threshold-safe.
- Lower-risk than the touch-level feature because it never changes the literal letters shown.
Suggested to build after adaptive typing (which establishes the shared learning/persistence plumbing).
Status: backlog — captured, not started. Doc branch
docs/suggestion-ranking-idea.A word-level ranking prior that biases the suggestion strip and live-converge toward higher-ranked words (higher rank → higher chance of being chosen). Complementary to, and separate from, the touch-level adaptive-typing feature (#ADAPTIVE).
Full design:
docs/SUGGESTION_RANKING.md.Summary
Suggest.kt#getSuggestedWordsForBatchInputcovers both the strip and live-converge (live-converge reuses that pipeline and takesgetWord(0)).Suggested to build after adaptive typing (which establishes the shared learning/persistence plumbing).