A few low-risk papercuts, each independent, please file PRs one-by-one:
is_llm_span only counts a span if gen_ai.request.model is set; also classify by gen_ai.operation.name
- unknown trace format silently defaults to ADK; return an explicit "unrecognized" result instead
gen_ai.system.message events are dropped everywhere, so system instructions never reach spans or the UI
- timestamp-unit guessing in
incremental_processor.py produces garbage for microsecond inputs
- invocation dedup by exact user-text equality collapses legitimately identical prompts
A few low-risk papercuts, each independent, please file PRs one-by-one:
is_llm_spanonly counts a span ifgen_ai.request.model is set; also classify bygen_ai.operation.namegen_ai.system.messageevents are dropped everywhere, so system instructions never reach spans or the UIincremental_processor.pyproduces garbage for microsecond inputs