Status: both Maya audio and bidirectional WRLD.BLDR MIDI working at the end, confirmed by the user. Root cause and permanent prevention remain unresolved. No app, firmware, logging profile, or other deployment was performed. The user performed the recovery actions; the agent collected diagnostics over the existing Wi-Fi connection.
Setup: iPad Air 13-inch (M3, Wi-Fi), iPadOS 26.6.1 / build 23G83, MAYA44 USB+ and WRLD.BLDR through the externally powered USB hub. The exact running app binary/hash was not established by this capture; the observed runtime configuration is recorded below. No external analog recording was made.
All event times below are PDT (UTC−07:00). Capture-directory names use UTC and therefore begin 20260929 for this September 28 evening session.
Maya was silent although SmartGrid appeared to play. WRLD.BLDR MIDI feedback initially worked. The user has already established that audio callbacks produce sound: meters animate and unplugging the iPad makes sound emerge from its internal speakers. This investigation did not require re-establishing non-silent sample generation. The user also reports that restarting the app has not fixed either recurring failure.
The same silent-Maya/empty-USB-input-transfer pattern was captured September 24 and September 27. It is not automatically the same cause as the older popping/clicking experiments in this journal.
Prior reports: September 24 findings, charging timeline, and September 27 recurrence. September 27’s two-day archive has uneven retention: USB-audio records begin September 27 around 16:16, so it does not establish continuous two-day USB health.
The first system archive retains actual events from 17:29:32.509 through 17:57:35.998; its earlier synthetic log-class marker is not an event. Maya remains selected at 48 kHz, 512 frames, four inputs/four outputs. Thermal state is 0 and low-power mode is off. Basic Maya and USB2-hub registry properties match the previous recovered baseline. A route, open endpoint, or enumerated USB device is therefore not a sufficient health check.
| Time | Observation |
|---|---|
| 17:50:38.411 | Latest retained pre-wake charging report says Connected → Connected. |
| 17:51:41.300 | SpringBoard records wake source Touch. |
| 17:51:42.065 | SpringBoard selects SystemSoundsAndHaptics. |
| 17:51:42.086 | Maya StartIO begins. |
| 17:51:42.357 | First excessive-zero-length-packet warning. |
| 17:51:42.720 | First zero-length input-transfer batch. |
| 17:53:37.327 | Existing SmartGrid process 4409 exits under installcoordinationd. This diagnostic session did not perform that installation. |
| 17:53:39.305 | Current SmartGrid process 4761 launches; persisted app log starts at 17:53:46. |
| 17:53:41.349–17:57:35.996 | USB input underflows, continuing to capture end. |
Counts: 14 excessive-empty-packet warnings, 29,026 empty input-transfer messages, 21,533 USB underflows, and zero pipe-stalled messages in the initial archive. Driver creation records associate these objects with Maya. The explicit errors are on its INPUT transfer path; they corroborate USB malfunction but are not measurements of analog OUTPUT or missing output payloads. Output silence is user-observed.
The failure predates this app launch, but an earlier SmartGrid process existed; do not write that the failure preceded all app execution. No new external-power connection was found around the observed onset. The snapshot reports external power connected, battery 100%, fully charged, and not actively charging. Charger telemetry does not establish voltage delivered to Maya, and missing logs do not rule out an earlier electrical event.
A full sysdiagnose was initiated remotely once through the existing CoreDevice developer tunnel at 18:01:37 PDT, using advertised feature com.apple.coredevice.feature.capturesysdiagnose. The response was ready at 18:03:41; all 722,934,494 bytes were downloaded over Wi-Fi. Received size matched the expected size, gzip -t passed, and the tar index was read successfully. This verifies that computer-triggered sysdiagnose works for this paired iPad; no button sequence, app deployment, or Audio Glitch Trace profile was required for this general capture.
The bundle contains USB, IOService, power, device-tree and port registry text dumps, retained system logs, power logs and a stacks report. Component limits: the XML-registry command and Type-C-retimer collector failed; coreaudio reporting did not meet its collection conditions. These are recorded subsystem limitations inside an otherwise complete, validated archive.
Audio Glitch Trace was not installed. This sysdiagnose is not an Audio Glitch Trace capture. The previously inspected Apple Audio Glitch Trace instructions URL redirected to developer sign-in; no profile was downloaded or installed during this investigation.
The host used the repository’s pinned pymobiledevice3 environment and shared connect_developer() connector, then DiagnosticsServiceService.capture_sysdiagnose(False) and its returned file generator. This is a verified service recipe, not a newly added ipad_logs.py command. The advertised feature and capture times are preserved in capture.json.
A bounded live system stream also succeeded: 270,986 events / 194,820,459 bytes, host interval 18:06:11–18:16:12 PDT, with device timestamps and host-receipt timestamps retained. Kernel records were collected, not just app logs. These are iPad unified/system diagnostics rather than a claim to have obtained a standalone dmesg buffer or raw USB bus packets.
| Intervention, in order | Result |
|---|---|
| Settings → Developer → Reset Media Services, with USB/power unchanged | Audio services and usbaudiod restarted, but Maya did not recover. Audio callbacks then stopped; controls independent of audio remained responsive. |
| Force-quit/reopen SmartGrid after that reset, with USB/power unchanged | UI/meters/callbacks resumed at 48 kHz/512; Maya remained silent and driver empty-transfer/underflow errors continued. |
| Disconnect/reconnect iPad-to-hub cable | Maya audio returned. All hub devices re-enumerated. iPad → WRLD.BLDR MIDI feedback broke; WRLD.BLDR → iPad still worked. |
| Disconnect/reconnect WRLD.BLDR’s own USB cable, leaving iPad/hub/Maya connected | MIDI feedback returned and Maya audio remained working. User confirmed both working. |
Correction of an intermediate assumption: an isolated Maya-cable reconnect was requested, but the user subsequently clarified that they had unplugged the iPad-to-hub cable. Logs agree: the entire upstream hub terminated and all devices re-enumerated. Maya-only recovery remains untested. A capture directory named 20260929T011352.155831Z-after-maya-only-reconnect-working was created before this clarification; its label is incorrect. It actually contains the state after the iPad-to-hub reconnect, with Maya working and WRLD.BLDR feedback stalled.
The media reset changed usbaudiod PID 438 → 5005 and audiomxd 106 → 5002, while SmartGrid remained PID 4761. SmartGrid received media-services-lost/reset notifications. The new Maya driver repeatedly attempted StartIO at 44.1 kHz, received empty packets and increased lock delay through 500 ms. At approximately 18:07:58, SmartGrid logged AUIOClient_StartIO failed (-66681); the installed SDK names -66681 kAudioQueueErr_CannotStart. Fewer later errors were caused by stopped I/O, not recovery. A fresh SmartGrid process 5010 launched at 18:09:44.661 and resumed callbacks, but silence and transfer errors persisted. Maya, WRLD and hub targeted registry data remained identical across the software reset, including USB session identities. Thus restarting both the app and the USB audio daemon was insufficient in this incident.
The final archive records the physical sequence:
preferred_rate=192000 while actual session/device audio remains 48,000 Hz / 512 frames; the preference value is not evidence of an active 192 kHz route.WRLD’s final-window totals are 11,479 OUT endpoint 0x01 stalls, each status 0xe0005000, zero bytes transferred; one IN 0x82 stall and four control 0x00 stalls occur around its physical removal. Do not confuse those removal-time errors with the sustained one-way output failure. Final registry comparison changes only WRLD’s sessionID; Maya and hub properties are unchanged by the successful WRLD-only reconnect.
The app logged both WRLD ports disconnected at 18:13:01 and reopened both at 18:13:10. Reconnect logic did run. At 18:13:54 its MIDI worker was running, queues were empty, submission counts had increased, and errors=0, while the kernel was continuously reporting stalled WRLD OUT transfers. Those counters measure API submissions rather than delivery acknowledgements. Input continuing to work does not establish output health. Endpoint availability and open handles are insufficient for detecting this state.
No reliable in-app silent-output detector or software recovery method was demonstrated. A hardware acknowledgement/round-trip protocol could potentially detect lost MIDI feedback if WRLD supports one; actual audio-path feedback/loopback would provide stronger output verification than route or callback status. These remain design candidates, not implemented fixes. MIDIRestart() was previously identified as an untested MIDI-only experiment; it was not called here. No supported in-app physical USB reset was demonstrated. The Apple-documented media-services reset was tested and failed to clear Maya.
The evidence establishes persistent USB-stream/endpoint failure with apparently healthy app routing/submission, plus recovery after USB device reconnection. It does not identify Maya firmware, WRLD firmware, hub/cable/power behavior or the iPad’s lower-level USB stack as the initiating cause, nor rule out application activity as a trigger. There is no demonstrated permanent fix.
Next time, preserve the failure before changing anything, then test Maya’s own cable only while hub power, iPad and WRLD connections stay fixed; document the exact physical action. This isolates the reset more narrowly than the verified iPad-to-hub recovery. Separately compare audio restart with and without sleep after an established working baseline, and compare power-first versus iPad-first connection order with a working-before-power-change baseline. A successful single run does not establish prevention. App callback/meters need not be re-proven as the first step.
The reset-induced stopped-callback state is a separate app recovery-handling issue worth investigating, without assuming that fixing it would cure the original USB silence. Do not deploy an experiment without the user’s authorization.
Diagnostic root: /Users/joyo/Documents/SmartGridOne/diagnostics/.
| Evidence | Path beneath diagnostic root |
|---|---|
| September 28 findings, focused excerpts and coverage metadata | 20260928-maya-live-failure/ |
| Initial failed app / system archive / device state | 20260929T005705.147174Z-app/, 20260929T005733.955585Z-system.logarchive/, 20260929T005738.775395Z-failed-device-state/ |
| Full failed-state sysdiagnose | 20260929T010136.392593Z-failed-sysdiagnose/sysdiagnose_2026.09.28_18-01-37-0700_iPhone-OS_iPad_23G83.tar.gz |
| Live recovery stream | 20260929T010611.485739Z-live-recovery/ |
| Media-reset app / system archive | 20260929T010845.994811Z-app/, 20260929T011002.272752Z-system.logarchive/ |
| Reset + fresh-launch still-silent state / app | 20260929T011057.536667Z-after-media-reset-still-silent/, 20260929T011101.776881Z-app/ |
| Hub-reconnect state, misleading historical label explained above | 20260929T011352.155831Z-after-maya-only-reconnect-working/ |
| Maya recovered / MIDI stalled app and system archive | 20260929T011434.311593Z-app/, 20260929T011429.852773Z-system.logarchive/ |
| Final working system / device state / app | 20260929T011632.684223Z-system.logarchive/, 20260929T011642.264648Z-both-audio-midi-working/, 20260929T011809.520961Z-app/ |
System archive retained bounds: initial 17:29:32.509–17:57:35.998; reset 17:55:02.002–18:10:03.996; reconnect 18:04:29.000–18:14:30.994; final 18:11:32.004–18:16:33.939. Coverage describes retained events, not guaranteed completeness. Live-stream and archive subsecond timestamps differ slightly; use the historical archive for the final physical-action timeline.
One final app download failed with an empty error string and remains 20260929T011637.735371Z-app.partial; it was not repaired or represented as complete. A fresh invocation succeeded after the concurrent system capture completed. The exact failure cause was not established.
References: Apple media-services reset documentation, CoreMIDI MIDIRestart, Audio Glitch Trace instructions URL.