What survives?
Fields whose meaning remains available at the destination.
Compatibility intelligence · exact bounded mappings
Tracevity binds exact source and destination format versions, then separates documented semantics, its own interpretation, and reproducible local fixture evidence.
Direct answer
A destination can parse a trace while losing principal identity, authority, event relationships, privacy controls, or evidence needed to verify external state. Each mapping below is directional and bounded to the versions, assertions, and synthetic fixtures it names.
Four answers
Fields whose meaning remains available at the destination.
Normalization, partial preservation, ambiguity, and source semantics without represented destination evidence.
Facts that still require identity, authorization, destination, or state evidence.
Primary documentation, bounded Tracevity interpretation, and linked reproducible local evaluation runs stay separate.
Published bounded mappings
Direction matters. A finding in one direction does not imply the reverse mapping, another version pair, backend acceptance, or behavior outside the named fixtures.
Directional mapping
Representative GenAI agent, model, and tool attributes survive local HTTP/protobuf serialization, while principal, approval, destination state, and complete-capture claims remain outside the representation.
Exercised August 29, 2026 · 7 explicit semantic assertions
Read the evidenceDirectional mapping
OpenInference kinds and attributes remain ordinary OTLP span attributes, but parseability does not repair broken correlation or prove a complete capture.
Exercised August 29, 2026 · 7 explicit semantic assertions
Read the evidenceDirectional mapping
MLflow's actual local exporter carries native mlflow.* model, tool, input, output, and trace structure through OTLP, but those vendor semantics are not automatically portable to a generic consumer.
Exercised August 29, 2026 · 6 explicit semantic assertions
Read the evidenceDirectional mapping
The documented flag maps selected MLflow model, usage, tool, name, and kind semantics into gen_ai.* while stripping most mlflow.* attributes; the transformation is useful but not lossless.
Exercised August 29, 2026 · 7 explicit semantic assertions
Read the evidenceThe published fixtures are synthetic and provider-free. Their local generation and validation establish only the named fixture output—not collector or backend acceptance, production completeness, external settlement, or hidden model reasoning.