Repository navigation
Connect receipt reporting and prepare the Sarvam source comparison - #56
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 ordinary Sarvam workflow previously relied on the model to label execution status. This version derives application status from trusted intent and real receipts while preserving raw model answers, dates and clarification flags. The finalizer also answers the original question, enabling source-request questions rather than only current-date reports.
The new live integration check completed 12 episodes / 30 API requests: application reports and stored states 12/12, approved writes 2/2, strict tasks 11/12. Raw reports remain 8/12; four status errors and one unnecessary safely rejected closed-booking attempt are preserved. The prospective application gate passes; the old failed gate and historical artifacts are unchanged. This is an integration result, not a general model-accuracy or memory gain.
Prepared, but did not execute, a frozen 288-episode comparison from 72 English/Hindi/Telugu inputs (24 related scenarios, eight known families), two fixed models and two backend-enforcement arms. A prospectively frozen attribution supplement distinguishes truthful reporting of an unsafe write from a factual error while retaining original desired-task grades. The execution wrapper binds that supplement before entering the existing source/gate/budget checks. Cases are assistant-authored, correlated, and lack independent native review; server snapshots and action arguments are supplied.
Validation: 129 focused tests; Ruff and diff checks; full offline reconstruction of all 30 live requests, actual SQLite state, reporting and charges; historical replay unchanged. A clean Git export passes the same 129 tests and verifies all 201 evidence-file hashes, exact setup replay and both prepared evaluation contracts without network or credential access. The archive contains immutable raw requests/receipts, databases, ledger, protocol, source and separate labels. No comparison calls or inference remain active.
New usage estimate INR 2.990055; native-research retained accounting INR 114.815436 within the existing INR 6,000 authorization. The larger comparison remains the next phase. Three independent commits preserve implementation, evaluation correction and evidence separately.