All times are September 13, 2026, Pacific Daylight Time. These are two observational SmartGridOne runs on the iPad through the USB hub with MAYA44 USB+ and WRLD.BLDR. They were not controlled A/B trials, and the retained evidence does not by itself establish why the failure rate changed.
real_time=0 at process start; this log alone does not establish the I/O thread’s sleep interval or exact installed binary hash.0xe00002ed. Each matched an AUAInputTransferManager status error and a CoreAudio I/O restart, followed by a 250 ms USB-audio lock delay. No matching MAYA output-endpoint transaction error was found in this cluster.long_gaps rose from 0 to 13, and xruns from 2 to 15, during the same cluster. Logged last-gap durations were about 317–327 ms. The 2 initial xruns and 2 format mutes preceded the failure cluster; do not count them as cluster failures. Instrumented processing overruns remained at 1 after startup.symptomsd power snapshots reported external power connected throughout the failure cluster. At 16:19:55 the battery was shown as 100% and battery-charging=1; by the next 20-second snapshot at 16:20:15, battery-charging=0, and it remained 0 through at least 16:39. The first three MAYA errors occurred while the charging flag was 1; the remaining ten occurred after it switched off. These flags establish reported battery charging state, not charging current, PD watts, or MAYA supply voltage. Power snapshots.recording-2026-09-13T161755.270.wav (about 12.53 GB). Its 4 KiB header was saved locally; the full recording was not copied or waveform-analyzed for this note. The earlier user listening observation that internal recordings lack the audible glitch remains separate evidence.Evidence: app timing extract, MAYA system episode, full app log, and system archive. The recording header is /private/tmp/sgo-20260913-1515-1645/recording-2026-09-13T161755.270.header.bin. Raw artifacts under /private/tmp are outside Git and may later be garbage-collected.
0xe00002ed transaction-status errors. The app had one 45.674 ms long_gap and 2 xruns by 19:31:30; neither counter increased again through its final diagnostic snapshot at 21:14:21. Processing overruns stayed at 1 and format mutes at 0.preferred_rate=45100 in minute diagnostics while the actual session/device rate remained 48,000; that preferred-value discrepancy is an observation, not evidence of an active 45.1 kHz route.Evidence: app log, USB-subsystem extract, end-transition extract, and system archive. The archive viewer warned of a wall-clock adjustment when queried with --end; timestamps above were cross-checked against app logs and the separate end-transition query. This was a log-based check, not an analog-output recording, so it cannot exclude audible artifacts without USB errors or app counter changes.
The afternoon episode reproduces the established failure chain: MAYA input transfer error → driver I/O restart/lock delay → roughly 320 ms app callback gap. The evening run shows that the same iPad/MAYA topology can remain free of that logged chain for about 1 hour 43 minutes after startup, even though initial enumeration was noisy. The difference is real within the captured logs, but neither run isolates its cause. In particular, startup aborts and the user-quit restart in the evening should not be grouped with spontaneous mid-session 0xe00002ed failures.