Repository navigation
Conversation
enaboapps
marked this pull request as ready for review
October 3, 2026 09:01
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds opt-in original, newer and combined prediction modes. Combined mode preserves newer suggestions and fills empty slots with normalized legacy suggestions, without mixing scores or personal counts. Existing APIs and default production model remain unchanged. No legacy database is distributed.
Closes #9.
The frozen comparison covers all six existing AAC, general-English and conversational partitions, prefixes zero through four, and 102,481 warmed queries per mode. Every newer suggestion retained its position. At two characters, test top-five accuracy is unchanged for AAC at 85.17%, improves general English 66.40% to 66.44%, and conversation 86.24% to 86.27%. At two characters gains are small because five newer suggestions usually occupy the list. At four characters, general English improves 83.24% to 87.79%, AAC 94.28% to 95.67%, and conversation 95.38% to 96.18%. Combined worst-partition p95 is 6.40 ms; cold loading is 6.6–6.8 seconds. Full metrics and limitations are in docs/legacy-comparison-results.md and .json.
Validation: Rust formatting, Clippy with warnings denied, 25 Rust tests, 14 Python tests and dependency audit passed locally. Hashes verify both source databases and all six partitions; databases are unchanged after evaluation. Historical ranking fixture and synthetic tests cover normalization, boundaries, corrupt data, personal-count exclusion and source immutability. Independent review cleared final head 63d7867 with no actionable findings and verified all 18 reports. All latest-head CI checks passed, including Windows/macOS/Linux checks, dependency audit, production database validation and packaged smoke tests on all three platforms.
No desktop integration, merge, model promotion or release is included.