AI API receipt · sgl-project/sglang
sgl-project/sglang#43331 · upstream bug reproduced by running the issue's own reproduction
Badgr cannot tell this response is wrong
With --linear-attn-decode-backend cutedsl the answer to "capital of France" came back as "The. beha -t"; the default triton backend answered correctly.
garbled text is still non-empty text with no schema to check; Badgr cannot tell nonsense from an answer. If the caller supplies the expected value (text contains "Paris"), Badgr rejects it: response text is missing expected content. 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. beha -t" tool_calls: none
Written by Badgr’s receipt ledger and read back from GET /v1/receipts/replay-sglang-43331-cutedsl-gdn-decode-garbage.
{
"request_id": "replay-sglang-43331-cutedsl-gdn-decode-garbage",
"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": "1791566159.53506"
}Recorded 2026-10-10 in Badgr’s local development environment. Replay cases and verdicts are pinned by backend/test_ai_api_replay_cases.py.
← All AI API receipts