Repository navigation
Validate numeric evidence, report completeness and metric bindings - #13
Draft
hudsonaikins wants to merge 4 commits into
Draft
hudsonaikins wants to merge 4 commits into
hudsonaikins wants to merge 4 commits into
Conversation
hudsonaikins
changed the title
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The analyst could report the correct factual answer while citing SQL that returned the value in different units or explicitly associated a report metric with a conflicting value. It could also finalize answered reports with null required metrics.
The runtime now checks scalar value presence, requires unavailable required report metrics to decline, and rejects contradictory explicit metric/value associations in SQL and structured Python JSON/JSON-lines. A value under another metric or in a unit column cannot rescue a contradictory binding. Population rows and SQLite Boolean representations retain their intended behavior. The runtime has no reference-answer or scorer dependency; scoring and tolerances are unchanged.
Validation: 277 tests pass with 15 optional skips; lint/format, Vulture, secrets checks and hosted CI pass on e501bd3. Offline replay rejects all three actual historical failed completions without model calls or grade changes.
Owner decision: the reliability benchmark is frozen. The proposed final 63-response live rerun is shelved and its $7 request withdrawn. This latest source has no live qualification result; no release acceptance is claimed. Historical experiments remain preserved in #9. The next milestone is a bounded real-data artifact demonstration, documented there. This PR remains draft, stacked on #8 and #10.