An unworn internal-short bench capture on 7 October exposed a recording boundary: one final EEG packet was preserved in the raw notification file but was not decoded into the sample table during shutdown. This is bench evidence about the recorder, not a participant result.
Recording summaries now expose recovered raw notifications. An otherwise successful run with such a tail is marked as completed with anomalies. Raw-only packets are not counted as decoded samples or live evidence. Older captures need independent reconciliation between raw packets, sample rows, and decoder counters; the original records remain unchanged.
October timing diagnostics also clarified the reported sample rate. The native parser constructs timestamps inside a packet using the advertised rate. An estimate of exactly 250 Hz can therefore persist even when packet timestamps contain a gap. The website now labels 250 Hz as nominal rather than an independently measured ADC cadence. Packet-level timing checks remain useful, but independent sample-clock verification is still open.
Offline tools compare packet-boundary structure, movement-sensor timing, and alternative channel representations while retaining source hashes and the original samples. Shared-noise cancellation can also cancel shared neural activity. Floating-input, resistor, and internal-short measurements cannot establish successful intent decoding or a usable worn EEG noise floor.
The current work improves what a recording report can justify. It does not establish reliable live control, solve the physical source of interference, or promote experimental firmware or denoising into a validated participant system.