Repository navigation
Audit separated Sarvam finalization and preserve failed transport repair - #53
Merged
Merged
Conversation
Sahith59
changed the title
Sahith59
marked this pull request as ready for review
October 11, 2026 19:12
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Native tool histories failed finalization because the caller omitted tool definitions. This adds a separated action/final-answer harness and a bounded repair that retains definitions with
tool_choice="none"; the final stage has no executable backend. Original implementation and failed requests remain replayable.The candidate is not promoted. Setup made 22 calls before stopping on six HTTP400s and preserved one actual pending-approval write in isolated SQLite. The separate four-call repair returned HTTP200 throughout, but base105B returned no final content in either format; Conversations returned the correct confirmed date in both. The larger 288-answer comparison remains unattempted. Strict rubric ambiguities are documented separately from factual/action failures; no general accuracy or memory novelty claim.
Validation: 37 focused tests, Ruff/diff checks, and clean Git export replay of all 26 calls, actual SQLite states, ledgers and all 190 artifact hashes. New known estimate INR0.658995; research cumulative known INR59.345550 / retained INR106.439732. No retries, credentials or real customer data in published evidence. Three independent commits preserve implementation, repair and results. English/default runtime unchanged.