feat(scoring): transport-agnostic CI grounding#35
Merged
Conversation
The disjoint 50/30/20 three-bucket AQ penalized users for old behavior, anchoring scores down even after genuine improvement. Switch to 65% recent-30d + 35% full-window cumulative blend: recent dominates while the full window (which includes recent activity) stabilizes without anchoring. Verified on real data: 94 Elite vs 83 Advanced under the old disjoint scheme.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Context Intelligence only counted MCP knowledge calls as grounding, making the axis blind to WebFetch/WebSearch (98% knowledge — validated against 144 real calls), knowledge skills (deep-research 1312 calls org-wide, claude-api 107, sdd-explore 659), CLI tools (codegraph explore, graphify, gh issue/pr view), and non-Claude skill paths (attributionSkill, Agent subagent_type).
This caused real scoring distortion: Nico scored 2.9/20 CI (5.9% coverage) despite heavy contextual tool usage, because 96% of MCP calls were browser automation and all non-MCP knowledge channels were invisible. Craft dropped from ~95 to 78, AQ from ~93 to 88.
Grounding now arms from 6 channels instead of 2, all idempotent per-session. WebFetch excludes localhost to avoid counting app-testing as knowledge. KNOWLEDGE_SKILL_NEEDLES follows the PLAN_SKILL_NEEDLES pattern and is extensible via GNOMON_KNOWLEDGE_SKILL_NEEDLES env var. Generic reference tools (man, pydoc, tldr) deliberately excluded — CI measures project-contextual grounding, not general programming reference.