v0.2.0

Changelog

• f4ddf7a Add a Honeycomb export path to the demo
• d27f256 Add a loss count, capture LangChain, and check values that only look conformant
• 7bdcddc Capture Braintrust, which turned out not to need a key at all
• 4f86d8e Capture the control case: OpenTelemetry's own instrumentation
• e6fc636 Measure what a span costs, and fuzz the codec that reads the network
• c64b6ec Show the Honeycomb integration, and what it looks like without it
• 8b3ebf9 Show the queries normalization makes possible, in Honeycomb
• 1a51f91 Watch the three things this repository does not control

v0.1.0

Changelog

• 99acd48 Add Braintrust, which asks a question the conventions half answer
• 1f57f0c Add LiteLLM, and tighten detection where it was too eager
• 66aabf5 Add the fallback, and keep it from diluting the dialects it backs up
• 31fc955 Build a real Collector with the processor in it, and run a span through
• b3c1fc4 Capture LiteLLM, which was winning its own span by a coin flip
• 9f41814 Capture OpenInference, which was losing the provider it was handed
• d872192 Capture a real OpenLLMetry span, and fix what it caught
• 373c40c Capture the Vercel AI SDK, and teach the sink two framings it did not know
• e08d52a Explain the losses the table only counts
• e0e8c5d Explain the moving target, and make CI hold the docs to it
• 7a4b09b Generate the conformance table instead of writing it
• ecd824b Join the two halves and prove it on fixtures
• 6ce6768 Rewrite the README around what it does, why, and how
• 12e82e2 Run the whole trip in containers: agent, normalizer, Jaeger
• a15c5b9 Say in the README what the captures actually found
• 45678df Ship the CLI, and let it say which version of itself it is
• f154c33 Show the span, rather than describing it
• 1ac0fc2 Teach the normalizer the dialect that is already half-conformant
• 2ce2460 Translate GenAI spans into a schema you can actually pin
• 314c56c Wrap the normalizer as a Collector processor