spantruth
Does your LLM telemetry tell the truth?
$ uvx spantruth doctor
Ghost span
the call never reaches a backend
-100%
Double span
one call reports usage on two spans
+100%
Cache-blind tokens
cached tokens left out of the count
-30%