Run the full 13-step audit

SSE integrity

Check SSE stream integrity

Inspect SSE structure and streamed model identity for substitution or corruption.

How this check works

A stream integrity check looks at SSE events on an OpenAI-compatible API. It checks event shape and whether the streamed model name still matches the request.

  1. 01Probe

    Open an SSE stream for a known request.

  2. 02Inspect

    Check event types, usage, and the streamed model name.

  3. 03Conclude

    Flag substitution or corruption when the stream no longer matches the request.

What this check does

LMSpeed reads the SSE stream and compares event types, usage, and model identity with the expected OpenAI-compatible pattern.

Why it matters

A broken or rewritten stream can hide a model swap, drop thinking signatures, or corrupt usage data while the final text still looks fine.

Final text can look fine after the SSE stream was rewritten. This check looks at event types, usage, and the streamed model name. Use it if your client reads the stream, not only the last message.

How to read the result
Final text can look fine after the stream was rewritten. A model-name mismatch in SSE is a stronger identity signal than the last paragraph.
When to run this check
Use this if your client reads the stream, not only the final message. Pair it with identity checks for chat apps that show the model name.

Recent reports

Public reports that included this check.

FAQ

What is OpenAI-compatible API stream integrity?

It means the SSE events stay well-formed and the streamed model identity still matches the request.

Why inspect the stream instead of the final text?

The final text can look complete after the stream was rewritten. Event types and model names are where substitution often shows up.

Does this check every streaming client?

It checks the SSE surface this endpoint returns. A different client wrapper can still change how events are consumed.

Other audit checks