Summary
The live trace-import tests default to a 7-day lookback. When the connected project has no traces in that window, they fail on an empty-data assertion instead of skipping, which is indistinguishable from a product regression.
Evidence
Against a project whose newest observation was 23 days old:
AssertionError: langfuse live response returned no records.
Set LANGFUSE_SINCE to a window with known trace data.
assert []
All three data-dependent tests failed this way. Re-running with LANGFUSE_SINCE=45d moved them past the data check and surfaced the real defect underneath (#102).
Location
tests/integration/test_live_trace_import_contracts.py:72
return os.environ.get(f"{provider.upper()}_SINCE") or "7d"Fix
Skip with an explicit reason when the lookback window returns no records, rather than asserting. An absent fixture in the connected project is an environment condition, not a failure. Optionally widen the default window.
Why it matters
An empty window masks real regressions. The first live run of this suite reported three failures, only two of which were genuine, and the genuine cause stayed hidden behind the empty-data assertion until the window was widened.
Summary
The live trace-import tests default to a 7-day lookback. When the connected project has no traces in that window, they fail on an empty-data assertion instead of skipping, which is indistinguishable from a product regression.
Evidence
Against a project whose newest observation was 23 days old:
All three data-dependent tests failed this way. Re-running with
LANGFUSE_SINCE=45dmoved them past the data check and surfaced the real defect underneath (#102).Location
tests/integration/test_live_trace_import_contracts.py:72Fix
Skip with an explicit reason when the lookback window returns no records, rather than asserting. An absent fixture in the connected project is an environment condition, not a failure. Optionally widen the default window.
Why it matters
An empty window masks real regressions. The first live run of this suite reported three failures, only two of which were genuine, and the genuine cause stayed hidden behind the empty-data assertion until the window was widened.