Skip to content

test: add reviewer v0.4 antagonist specs#7

Closed
laynepenney wants to merge 1 commit into
mainfrom
sentinel/reviewer-v0.4-antagonists
Closed

test: add reviewer v0.4 antagonist specs#7
laynepenney wants to merge 1 commit into
mainfrom
sentinel/reviewer-v0.4-antagonists

Conversation

@laynepenney
Copy link
Copy Markdown
Member

Summary

  • add Sprint C reviewer-framework v0.4 antagonist specs across reviewer, judge bridge, and suggestion-engine surfaces
  • keep the pack OSS-safe by modeling temporal-like ambiguity geometrically rather than copying Conversa corpus material
  • validate plugin seams for custom Predicate/Reviewer subclasses and chain composition

Spec Status

This PR is intentionally red-by-design.

Targeted slice result:

  • python -m pytest -q tests/unit/test_reviewer.py tests/unit/test_judge.py tests/unit/test_suggestion_engine.py
  • 104 passed, 3 failed
  • ruff check tests/unit/test_reviewer.py tests/unit/test_judge.py tests/unit/test_suggestion_engine.py passes

Intended red seams:

  1. JudgingReviewer does not clamp out-of-range adapter scores at the bridge
  2. JudgingReviewer does not clamp negative adapter scores at the bridge
  3. HallucinationSignalRule does not handle score=None safely

Boundary

  • no Conversa corpus copied into OSS
  • eval#1 category-scoping remains a separate slice and is not folded into this PR

laynepenney added a commit that referenced this pull request May 8, 2026
…gainst None

JudgingReviewer now clamps adapter scores to [0.0, 1.0] at the bridge
layer, preventing out-of-range scores from propagating through reviewer
chains. HallucinationSignalRule now skips verdicts with score=None
instead of raising TypeError.

Closes #7

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@laynepenney
Copy link
Copy Markdown
Member Author

Closing this red-by-design spec PR in favor of #9. The contract served its purpose, but should not take a knowingly failing test PR pre-v1.0. #9 carries the same spec tests plus the implementation, and the full slice is green there.

@laynepenney laynepenney closed this May 8, 2026
@laynepenney
Copy link
Copy Markdown
Member Author

Closing this red-by-design spec PR in favor of #9. The contract served its purpose, but main should not take a knowingly failing test PR pre-v1.0. #9 carries the same spec tests plus the implementation, and the full slice is green there.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant