Repository navigation
Guard gold export and locked-portion scoring for the new sets - #55
Merged
Merged
Conversation
Reviews started before the from-scratch defaults were removed are never exported as gold, cannot be reopened, and are counted as invalid. The six challenge dev protocols whose machine labels were seen are flagged in their gold records, and gold reports also score without them. Scoring a locked test portion against gold requires every test source to be annotated and is refused once a report for that model exists.
professor3333
force-pushed
the
data/gold-scoring-guards
branch
from
October 19, 2026 04:40
593c2ce to
38b2acc
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Prepares the new annotation sets for human gold so the first real export can't silently include invalid or partial data.
MANUAL_VALID_FROM(commit 21446d6): from-scratch reviews started earlier, when the UI preselected a type, label and direction, are skipped by the export (counted asinvalid_reviews), andreopen_review.pyrefuses to reopen them. On the currentdata/annotation.dbthis skips exactly the three known later-domain reviews.challenge/machine_labels_seen.jsonlists the six challenge protocols whose machine labels were seen (all six are the dev portion). Gold records carrymachine_labels_seen; gold reports add scores without them.evaluate_annotation_set.py --split testagainst gold: refused until every test source has gold, and refused once a report for that model exists.dev 0/22, test 0/16).Tests cover the validity rule (including continuing an early review), the flag, and both locked-portion refusals.