Thermal mechanisms after native tracing — September 10
Discussion update only. No new device query, condition, build or experiment was started. The native trace report and thermal comparison provide the measurements. The foundational internal-recording observation must remain part of all subsequent reasoning.
Observation versus mechanism
The three sequential comparisons contain 22 nominal versus 6 induced-Serious matched transaction/restart failures over approximately nine minutes per state; the third pair is 8 versus 5. This motivates a thermal-policy hypothesis but does not establish a causal rate effect. Session age, bursts and missing same-session nominal reversal remain confounded. Inducing Serious is a policy intervention, not a measurement of hub/MAYA temperature or proof of electrical heating. Apple documents thermal simulation without physically warming the device: WWDC19.
For the captured native event, completed app processing and prompt audio/observed USB-thread dispatch weaken immediate CPU starvation. UI scheduled fraction, worker wake frequencies and observed priorities barely changed between the short nominal/Serious windows. Those measurements do not cover CPU frequency, every driver operation, GPU/DMA bursts, input supply current, or MAYA’s supply waveform.
Remaining explanations, conditional on a real policy effect
- Reduced or less variable electrical demand. Thermal policy can change processing/display/radio behavior and charging. Lower iPad demand could change load on shared power circuitry; fewer or smaller load transients could improve a marginal USB electrical path. This is a hypothesis, not a measured demand reduction. Battery filling was already off in both measured states of trials 1 and 3; many failures also occurred with PD absent. Those observations weaken the battery-filling and PD-renegotiation variants, but do not measure shared-supply stability. The simple no-PD experiment has already been done and is not a new proposed exclusion. Actual CPU scheduled fraction is not electrical power.
- Changed host-side timing or concurrency. Thermal policy can change cluster activity/frequency and background work. A changed interleaving, hardware-transfer service pattern or driver/controller state could make a latent timing bug less likely without any late app callback. This remains possible, but the specific claim that our ready audio thread was starved conflicts with the captured trace. App wake counts do not measure all OS, GPU, memory/DMA or controller activity. We have no documented Serious-specific USB priority or clock policy and must not invent one.
- Physical temperature dependence. An oscillator, regulator or other component could behave differently with temperature, but the induced-state result, if replicated causally, would favor policy-mediated effects over a requirement that the external hub/MAYA physically warm. No external component temperature was measured. CPU operating frequency must not be conflated with the USB frame clock or MAYA sample clock.
- Time/session-state confounding. The apparent benefit could reflect burstiness or evolution after startup/restarts. This remains an alternative to all causal thermal mechanisms, particularly given the third pair and lack of a measured reversal. It is not a reason to disregard the association, and the user has accepted proceeding without making a new order-reversal trial a prerequisite.
Sources and discriminating evidence
Apple describes thermal-dependent CPU-cluster activation/frequency: CPU scheduling Tech Talk. Apple describes possible charging, display, radio and performance reductions: temperature management. TI explains voltage undershoot/overshoot following changes in load: load-transient response. These establish mechanism plausibility, not Satechi circuit details or a diagnosed fault.
To distinguish the leading branches, the most direct missing observations are synchronized supply voltage at MAYA / input demand and the lowest available USB transfer-completion/timing information before errors. A basic slow power reading cannot exclude a brief rail disturbance. A independently powered or replacement-hub comparison changes additional variables unless the data path is preserved; it must be described accordingly. No such new test is running.