Status: complete; both symptoms reproduced in the cooled same-build repeat. The latest user instruction was five additional minutes, then cool and repeat the same build if clean. The earlier proposed hour is canceled. User confirmed WRLD.BLDR LEDs work after the16:13 restart.
K-Mix3/4 recorded16:34:52.525–16:39:52.301, actual299.776seconds. The original live recorder had a3600second limit; a companion stop controller verified wrapper30136/SoX30143 and sent SIGINT at16:39:52.544 per the user’s revised300second request. Return0, stderr0; original raw metadata retained.
There are no long near-silence intervals, short flat candidates or dense periodic episodes. System logs contain zero MAYA transaction errors/restarts, zero-length-transfer reports or ioDrift messages in the recording window.28,104 callbacks stayed48k/512, serious thermal2, low-power0, unmuted, xr2unchanged, with no gaps above20ms, sequence holes or logger misses. DSP median/p99/max4.426/5.150/5.224ms, max arrival gap10.696ms. The MIDI worker was enabled/running in299samples;12,478 additional SysEx handler submissions, sampled depth0–1, no full/invalid/discard counts. Actual LED operation was confirmed by the user for this build. Full result.
Preserved system archive and complete app trace before stopping. Config and patch hashes remained unchanged. SmartGridPID19797 received SIGTERM at16:43:33, and the device API confirmed no running process for its bundle at16:43:35. Earliest repeat attempt is16:48:35 after five minutes passive cooling. No hub/interface/controller/power connection, wireless setting, thermal profile, GPU profile, app source or saved configuration was changed. A first local helper call used an unsupported Python enum argument and failed before sending any signal; correcting it to the API’s integer argument completed the authorized stop. Raw stop evidence is /private/tmp/smartgrid-ipad-passive-cooldown-r1-20260910.retry.json.
Start K-Mix recording BEFORE relaunch, so startup failures are captured. Use the same installed binary and normalDSP/UI/autoplay/MIDI-worker-enabled environment, internal clock, actual48k/512, input0/output4 and same patch. Avoid screenshots. Inspect the initial app trace to verify nominal thermal0 and settled200callbacks at48k/512. If still elevated, preserve that attempt as a thermal probe, stop the app again and allow more passive cooling; do not call an elevated restart a cooled test. No induced thermal condition may fake a nominal state.
Default repeat capture is960seconds including the planned prelaunch lead-in. Preserve and label expected initial silence before playback,470frame/initialization callbacks, and genuine startup USB/callback/analog failures separately. Do not discard startup faults: the earlier hot restart had two failures before its analog recording. During subsequent measured playback, no device queries, screenshots, builds or state changes. Afterward collect archive/app trace and score both sporadic interruptions and periodic corruption. This is one conditional hot/cooled comparison, not proof of a thermal cause from a single pair.
K-Mix recording began16:49:00.456 before launch. SmartGrid began playback about16:49:12, after over five minutes stopped. PID20023, log2026-09-10T16-49-06-780.log; no reinstall or source/config changes. Verification at16:49:18 confirms normalDSP/UI/autoplay, JUCE8.0.15, MIDI worker enabled, increasing SysEx submissions without full/invalid/discard counts, input0/output4, and200consecutive48k/512callbacks with thermal0. No screenshot was taken. All device queries ended with startup verification.
Recorder exec30407, wrapper30495, SoX30502. Expected finish17:05:00.456 for960seconds including prelaunch lead-in. The first46.876seconds contain planned silence0–11.620seconds, then ordinary audio around-30/-32dBFS with no additional long-gap candidate. The first initialization callback was470frames and18.535ms DSP; xr0→1 atcb3 during470-frame startup, with11.948ms arrival gap. There is no gap above20ms in the saved startup trace. These initialization events must remain visible separately from settled playback; do not label the xr change a matched USB dropout without system/analog evidence.
Repeat start metadata. Current raw prefix:/private/tmp/smartgrid-ipad-cooled-r1-20260910. Final analog and system-log comparison pending.
The local-only scan through499.711seconds (16:57:21 checkpoint) found18 long near-silence intervals after playback began, excluding the expected0–11.620second lead-in. Near-silent cores span117–227ms. The first core starts at WAV87.438seconds, approximately16:50:27.894. Visual inspection of three representative stereo waveforms confirms abrupt loss, a flat near-zero interval, and resumed music; analog settling can make the full interruption longer than the thresholded core. This differs from the measured hot intervals, which had no long gaps.
The run continues unchanged through17:05. No iPad queries have occurred since initial nominal-state verification. USB/callback correlation, thermal evolution and periodic corruption remain pending full post-run analysis. Cooling/restart state is associated with recurrence in this pair; it is not yet a demonstrated thermal mechanism. Interim evidence.
The recording finished17:05:00 with960seconds retained, return0 and empty stderr. All21 actual long waveform gaps match21MAYA errors,21driver restarts and21callback stalls. A separate0.608second burst contains29periodic holes alongside29alternate callbacks exceeding the10.667ms budget, without xr increment or a nearby USB failure.18long failures occurred while nominal and3after serious state, so serious is not a guarantee of clean playback. Same config/patch and MIDI worker remained active without queue faults. Full evidence and limits: completed comparison. The authorized continuation/cooldown/repeat is complete; no further experimental dimension was activated.