Repository navigation
Preserve source coverage and add bounded adjudication correction - #224
Merged
Merged
Conversation
On overflow, place each whole source unit in the first earlier batch with enough escaped-JSON capacity. Cache escaped sizes during packing to avoid repeated payload scans. Preserve source units, hunk ownership, citations and request admission. Continue to fail closed when review coverage is incomplete.
Reuse spare source batch capacity and cache escaped byte sizes. Keep the exact next-fit inventory when the packed source count exceeds admission or the provider selects five batches. Allow one correction shared by adjudication parsing and the two recoverable semantic rules. Validate both logical calls against immutable original inputs, retaining the original deadline and shared provider caps. Keep invalid output fail closed and preserve usage accounting. Update the partial-coverage fixture and historical calibration tests. Permit recalibration of a populated baseline when the evaluator contract changes, after verifying prior provenance and retaining current-main, first-attempt execution with ten immutable sample slots. Report sample uploads from their actual outcomes and set the CLI version to 0.9.21.
Assert the exact byte-cap cause and typed local admission classification, zero HTTP requests, and zero reserved attempts. Check that the formatted review error preserves the precise cause.
Retain the validator cause when adjudication fails. Update integration mocks to accept only the fixed correction suffix and assert exact request phases, immutable inputs and known usage receipts. Validation: E2E suites with default and no default features, relevant adjudication and admission tests, formatting and all-target Clippy.
Classify unknown candidate IDs at the strict validator so they use the correction attempt shared by syntax and semantic errors. Preserve strict identity checks and residual admission. Decode legacy and phased adjudication records with closed shape and phase checks. Exercise exact request roles, attempt counts, usage and fail-closed publication with local mocks.
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.
Large diffs can exhaust the source batch allowance despite unused capacity, and correctable adjudication output can terminate a review without a valid result.
Reuse spare source batch capacity with cached escaped sizes, falling back to the exact original layout when source batches exceed admission or provider selection is limited to five batches. Allow one shared correction for malformed adjudication JSON, unsupported refutations, or missing required causal references, preserving complete validation against immutable inputs within the original deadline and remaining budget. Update the coverage fixture and benchmark tests, and allow populated-baseline recalibration when the evaluator contract changes while retaining execution from
main, first attempts, ten immutable sample slots, and provenance checks. Set the CLI version to0.9.21.Focused offline packing and adjudication checks passed, and the corrected mock benchmark passed 70/70 cases.