[Infra] Add PoC of overfitting judging#140
Merged
Merged
Conversation
Member
|
Do you have a design doc or can you share more info on what this does? |
Contributor
There was a problem hiding this comment.
Pull request overview
This pull request adds a proof-of-concept LLM-based overfitting detection system to the skill-validator infrastructure. The feature analyzes whether evaluation tests genuinely measure skill quality or merely test for memorization of specific skill vocabulary and syntax patterns.
Changes:
- Introduces an
OverfittingJudgeservice that uses LLM prompting to classify rubric items and assertions into categories (outcome/technique/vocabulary for rubrics, broad/narrow for assertions) and computes an overfitting score - Integrates overfitting checks into the validation pipeline as a parallel, non-blocking operation with configurable CLI flags (
--no-overfitting-check,--overfitting-fix) - Extends reporting to display overfitting scores in console output, markdown tables, and dashboard visualizations with severity-based icons and colors
Reviewed changes
Copilot reviewed 8 out of 8 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
| eng/skill-validator/src/Models/Models.cs | Adds OverfittingSeverity enum, assessment record types (RubricOverfitAssessment, AssertionOverfitAssessment), OverfittingResult, and configuration options to SkillVerdict and ValidatorConfig |
| eng/skill-validator/src/Services/OverfittingJudge.cs | New service implementing LLM-based overfitting analysis with prompt engineering, score computation, retry logic, and optional fix generation |
| eng/skill-validator/src/Commands/ValidateCommand.cs | Adds CLI options and integrates overfitting check as a parallel task that doesn't fail the verdict on error |
| eng/skill-validator/src/Services/Reporter.cs | Extends console and markdown output to display overfitting scores with severity-based formatting; adds FormatOverfitCell helper |
| eng/skill-validator/tests/OverfittingJudgeTests.cs | Comprehensive test suite covering score computation, JSON parsing, severity mapping, prompt building, and reporter integration |
| eng/dashboard/generate-benchmark-data.ps1 | Adds overfitting severity and score fields to benchmark entries when moderate or high overfitting is detected |
| eng/dashboard/dashboard.js | Adds overfitting summary cards, chart markers (star symbols), tooltips, and legend notes using shared helper functions |
| eng/skill-validator/OverfittingDetection.md | Comprehensive implementation plan documenting prompt design, architecture, scoring methodology, and integration approach |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Member
Author
I generaly try to avoid checking docs as they can stale ... but here it is a good idea - so added it to the changeset |
Contributor
Skill Validation Results
Model: claude-opus-4.6 | Judge: claude-opus-4.6 |
ViktorHofer
approved these changes
Feb 27, 2026
Member
|
GH shows a few possible null-reference warnings in the build in the pull request review page. New feature? :) |
Member
|
BTW I didn't understand the intent of this score until I finally read the prompt in the judge. I think the eng/skill-validator/README.md describes the different judges. It would be good to provide a simple and understandable description of this judge there. |
Member
Author
Good point! |
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
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.
Motivation
To have visibility into whether the eval tests are providing results that reliably judge skill quality versus just overfit on specific skill features
Sample