ETW EventsLost=0 but fewer records arrive: reconcile independent event keys
Zero loss counters do not establish complete capture.
A reference outside the counters
Zero loss counters do not establish complete capture. Compare received events against independently verified evidence for the same scope. A Truthound reference rebuilt from the same ETW consumer would not be independent; first establish how known test requests correspond to provider events. [1, 2, 3]
Requested flags and applied flags
Beats #50966 reports high-load DNS ETW collection yielding fewer events than a separate AMA consumer despite zero loss counters. Its flag comparison distinguishes requested flags from the flags actually applied by the kernel, including INDEPENDENT with a hybrid-shutdown flag. The environment-specific counts are not a universal loss rate or a new reproduction by this article. [1]
Two consumers over the same scope
Use a load test with the same RUN_ID and time window, reconciling source, independent consumer and Beats keys. Hold an apparently normal capture if verified keys are missing. Align provider, keywords and duplicate policies, and collect the flags actually applied. Different consumer filters must be resolved before calling a count difference loss. [1, 3]
Reconcile the keys A~J
The illustrative test independently verifies ten events A~J. All ten landing gives missing count 0. Missing H,I,J gives 3 and fails. Ten different unique keys would not pass merely because the total count matched. The condition covers only the verified reference inventory. [1, 3]
Check prepared key differences
Truthound Core Python SDK 3.3.2 can compare prepared permitted_missing_keys and missing_verified_keys using ColumnPairValidator with relationship="==". The operator calculates key differences and isolates unknown reference scope. The SDK does not change ETW session flags or restore kernel events, and AMA is not guaranteed to be a complete ground truth. [1, 2, 3]
Evidence still needed after a change
Test the INDEPENDENT and hybrid-shutdown combination in isolation and confirm its actual readback. Repeat the same RUN_ID and key reconciliation after a change. Retain counter observations, but do not use their remaining at zero as the sole evidence of complete capture. [1, 3]
Application conditions
Align provider, RUN_ID, time window and duplicate policy, and assess the reference consumer’s own coverage.
Alternatives
Run an isolated flag test with actual readback and independent key reconciliation.
Limitations
The SDK does not change ETW flags or recover kernel events; AMA agreement is not absolute source truth.
Evidence
- Event checked
- Subject checked
- Modified
