Repository navigation
Preserve task coverage and uncertainty in concise responses - #11
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Short questions could suppress necessary verification, and the one-topic/confirmation rules could interrupt work that was already authorized. Version 0.5.0 matches length to task complexity, preserves material uncertainty, covers all requested topics, and continues routine authorized fixes while retaining consequential clarifications and destructive-action confirmation.
Adds a bilingual response suite with 36 actual captured answers across no rules, 0.4.0, and 0.5.0, plus independent blinded semantic review and an offline scorer. Quality is separate from keyword indicators and character counts; source, answer, and reviewed-input fingerprints reject stale grades. CI checks the recorded candidate and its match to the shipped canonical rules. The Codex package is regenerated, README host sections are preserved, and the Claude hook banner is shortened to keep full output below 10 KB.
In this small pilot, semantic passes were 12/12 without concise, 8/12 with 0.4.0, and 12/12 with 0.5.0. The revised batch is 17.4% longer than 0.4.0 and 41.0% shorter than no rules. This is one batch per condition with an agent reviewer from the same model family; it does not establish general or cross-model improvement. Raw answers, review evidence, and limitations are in evals/PILOT.md.
Validation:
Authored with OpenAI Codex. One implementation commit.