Repository navigation
Test manifest schema compatibility with a bounded same-input probe - #226
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The v17 review reached the configured inference endpoint, but all 19 specialist requests returned the same HTTP 400 signature. Its whole-object target enum is a compatibility hypothesis, not a proven cause. The opt-in target schema now uses nested per-kind anyOf objects with primitive string enums; empty manifests use a valid item schema with zero allowed proposals. Server validation still checks the exact kind/value/evidence-kind binding before spending retrieval quota. Legacy request bytes, profile bytes, limits, and adapter identities remain unchanged.
Adds a main-only, default-off diagnostic that captures one fixed public PR305 task from bare Git objects and sends at most two same-input, zero-retry requests: the actual archived b075 schema and the candidate. It verifies only the target-schema subtree differs, binds both to the same fresh preparation receipt, and uploads only bounded hashes/status/error categories. It never checks out or executes target files, exposes no raw provider body or credentials, and cannot publish reviews. Historical request hashes are comparisons, not eligibility gates.
Validation: 16 root focused tests passed; independent review CLEAR with an actual eligible bare-source preparation and 15 independently run tests. Standard JSON Schema checks cover allowed and cross-kind pairs and the empty case; the existing blocker/quota regression passed. Fake provider tests passed 29 tests plus six subtests. Ruff, YAML/TOML, and diff checks passed. Endpoint compatibility remains unknown until the paired probe.
Reference for the candidate construction: OpenAI Structured Outputs supported schema subset. This does not establish third-party endpoint compatibility.