Repository navigation
Release 2.0.3: opt-in transparent sub-page scoring for dataset v2.02 - #296
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What changed
Prepare CLI 2.0.3. Datasets can opt into
metadata.sub_page_handling: "transparent-v1"; text and formatting scoring then ignore only<sub_page>container tags on both sides. Old datasets retain existing behavior. Dataset v2.02 requires this version; compare scores using the same dataset and CLI revisions.Why
Providers do not need to reproduce GT-only embedded-page wrappers. Without this handling, otherwise identical text loses 5–10 points. Header/footer matching from merged #294 is unchanged. Release notes also cover the intervening form-scoring and Metric Lab changes.
Testing
4,159 parser-metric tests passed, 1 skipped. New regression cases cover nested wrappers, retained footer content, wrappers in provider output, legacy default behavior, and rejection of unknown contracts. Ruff checks pass. Three new opt-in cases fail without the fix and pass with it; the legacy case stays green. Replay: 508 text GT self-checks, 592 header/footer assertions, and six wrapper-omission checks all pass. Table-only eligibility remains unchanged.