feat(scoring): dual-window blend and expanded skill recognition#33
Merged
Conversation
Improvements take months to reflect in AQ because all sessions in the window carry equal weight. This adds a 30-day rolling window (65% weight) blended with the full window (35% weight) so recent behavior dominates while history stays as soft context. Also shrinks the default mirdash upload window from 6 to 3 months, expands review skill detection to cover review-* prefixed skills and adversarial protocols (judgment-day, jd-judge-*), and counts SDD task planning skills (sdd-tasks, sdd-ff) toward the Discipline axis alongside TaskCreate.
Guard all dual-window blend logic behind RECENCY_BLEND_ENABLED (default True). When False, scoring falls back to the previous single-window behavior with no recent accumulators or AQ blending. Regenerated scoring vectors after the expanded task/review skill detection.
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The scoring window was too large (6 months, flat weighting) so improvements took months to show up. A user who starts grounding every session would see their CI score diluted by months of ungrounded history.
This adds a dual-window blend: 65% weight on an exact 30-day rolling window plus 35% on the full analysis window. A dedicated accumulator tracks the recent window during the same file pass, so there's no approximation from calendar-month buckets. The default mirdash upload window also shrinks from 6 to 3 months.
Review skill detection now covers
review-*prefixed skills (review-readability, review-risk, etc.) and adversarial protocols (judgment-day, jd-judge-*). The Discipline axis counts SDD task planning skills (sdd-tasks, sdd-ff) alongside TaskCreate/TaskUpdate.