Profiling¶

Profiling owns opt-in instrumentation and trace collection used to locate runtime bottlenecks without changing model results or execution policy.

Candidate invariants¶

Rule: Expensive trace collection MUST be disabled by default and bounded by an explicit start and stop lifecycle.

Rule: Profiling MUST NOT change scheduling decisions, generated outputs, or resource ownership beyond documented synchronization overhead.

Rule: Trace events SHOULD identify stage, rank, worker, operation, and request where available without embedding user payloads.

Test disabled overhead, repeated start and stop, cleanup, multi-rank trace naming, and each supported profiler backend.