Native burst delivery and Joy-Con rumble follow-up¶

Historical record. This preserves the source, tests and limitations recorded at the time; it is not current setup guidance. See the documentation index.

User reports¶

• Test Heavy / Test Light on Switch 2 hardware has persistent slight stutter and crackling, reported on USB and Bluetooth.
• A separate GTA V Enhanced report describes missed closely spaced machine-gun effects with a physical DualSense emulating DualSense, on USB and Bluetooth.

These are distinct paths. The local Switch 2 TestPreview lane does not traverse VIIPER's native DualSense output queue. Fixing one does not establish the other.

Live baseline: Bluetooth Joy-Con 2 (L)¶

On September 9 at 22:56 local time the existing installed RC4.5.4 process remained running, with profile ds, a standalone Bluetooth Joy-Con 2 (L), 100% rumble, no rumble delay, and virtual DualSense output. Neither app was restarted and no saved profile was changed. Test Heavy was started and stopped through the existing UI.

An isolated 25-second observer enabled only the local BTHPORT event 402 v0, keyword 0x8000000000000000. Its parser accepted only structurally complete 36-byte events containing the exact 20-byte rumble-shaped ATT write. It saved waveform hashes, counters, amplitude summaries and timestamps, not raw packets or an ETL. The unique observer session stopped successfully, with zero ETW lost events. Its strict parser passed synthetic valid, malformed, short, non-rumble, neutral and direction-separation tests before capture.

Capture: Desktop/Controller-Diagnostics-2026-09-09/joycon-heavy-rc454-402-20260909-225612.

This sample does not show large host-side stalls, alternating neutral payloads or skipped counters. It is not radio-delivery, motor-motion or perceived-quality proof. The stop was clicked after the accepted capture window; the trace does not establish its physical completion. Earlier event-263 attempts returned no events and provide no write-timing evidence. The user explicitly confirmed that this captured Test Heavy run still crackled/stuttered.

Native DualSense command defects¶

Failing-first broker fixtures use the actual USB output callback and framed feedback writer. A pulse followed by stop, or a trigger update followed by an LED-only command, previously lost the first exact command in a latest-state latch. The broker correction uses its existing bounded ordered queue and reports rejection to USB/IP before committing cumulative state when admission fails. It does not block the USB/IP reader or retire the whole controller. Host retry after explicit overload is not guaranteed.

Separately, actual DS4Windows dispatch-to-physical-owner fixtures reproduce oldest-command eviction at capacity and expiration of an admitted stop/LED command at 20 ms. Downstream retained-head/admission changes now pass 175 focused tests, including actual gamepad-only dispatch, full physical-ring rejection without scalar fallback, lifecycle cancellation, and irrevocable admission even when subsequent diagnostics fail. Exact commands retain their accepted head; PCM and legacy newest-state policies are not converted into an unbounded FIFO.

Source-backed Joy-Con experiments, not a validated fix¶

• Hifi SDL's BLE implementation at local pin d98c5804a9d20b0d96e993741797878c86b8f1e1 uses 10 ms rumble servicing. Its BLE work lives on feat/hidmaestro-filter, not default main.
• Switch2Connect's input-stall report remains an important counterexample to assuming faster rumble is free: reporters describe controls stalling during joined Joy-Con rumble. It does not establish that our controller has the same cause.
• A standalone-only 10 ms candidate preserves Pro/joined Bluetooth at 15 ms, USB at 12 ms, and exact waveform/carrier/amplitude bytes. Nine new cadence cases are included in a 165-test integrated green run. Hardware acceptance remains open; constants do not prove actual timer or radio cadence.
• SDL USB fills one active subframe; Hifi SDL BLE fills one active subframe and two neutral-amplitude subframes. Our three-active-subframe held pattern agrees with Switch2Connect's real held sender at local pin 61ac6642ce12fe7217e38a860b14863b18ca7e28, so this is an implementation difference, not a proven invalid packet. Do not simultaneously change frame count, carriers, strength and timing, or modify incoming PCM/one-shot groups.
• An isolated .NET 8.0.29 one-shot timer check measured 11.4463 ms median for a 10 ms request with successful timeBeginPeriod(1). It also showed long scheduling tails; it is not controller evidence. The exact runtime's private TimerQueue clock uses QueryUnbiasedInterruptTime on this Windows version, not the Environment.TickCount64 discussed in issue dotnet/runtime#122680. Therefore a universal 15.6 ms ceiling was not established.
• The existing 20-report latency average is not an adequate input-tail guard. Opt-in accepted-report gap telemetry now reuses the portable diagnostic mechanism without input-path I/O or allocations. It is disabled in normal production launches and rejects all audio/GATT commands in input-only mode. Its focused runtime/pipe/cancellation/allocation run passed 77 tests.

Combined source validation¶

The first full run caught two source-inspection tests whose extraction markers still named the former void dispatch method. Those markers were corrected to the new bool signature, retaining their assertions and extending the persistent scratch check to retained native admission. The second complete unfiltered run passed 5,007, failed 0, and skipped 11 explicitly gated hardware/cross- runtime integration rows. Allocation assertions remained enabled. Evidence: isolated_results/haptics-candidate-full/full-haptics-candidate-reviewed.trx.

Neither source reproduction is a captured diagnosis of the GTA user's session. No fix described here is included in immutable RC4.5.5. Hardware acceptance and complete portable delivery remain open.