AI API receipt · sgl-project/sglang
sgl-project/sglang#43146 · upstream bug reproduced by running the issue's own reproduction
Badgr cannot tell this response is wrong
For a minimax-m3 model the toggle raised ValueError on one setting and was a silent no-op on the other.
a request that errors is retried elsewhere, but a silent no-op still returns valid text; Badgr cannot see which reasoning mode ran. The receipt therefore records success. This is a limit of what the contract can prove, shown here so it is not mistaken for protection.
The output the bug produces, taken from the issue’s reproduction and fed to Badgr’s validators. Not a live provider call.
text: "The answer is 4." tool_calls: none
Written by Badgr’s receipt ledger and read back from GET /v1/receipts/replay-sglang-43146-minimax-m3-reasoning-toggle.
{
"request_id": "replay-sglang-43146-minimax-m3-reasoning-toggle",
"status": "success",
"endpoint": "/v1/chat/completions",
"stream": false,
"failure_stage": null,
"error_code": null,
"error_message": null,
"retryable": null,
"provider_mode": "replay",
"routing_decision": "replayed_upstream_output",
"created_at": "1791524709.944198"
}Recorded 2026-10-09 in Badgr’s local development environment. Replay cases and verdicts are pinned by backend/test_ai_api_replay_cases.py.
← All AI API receipts