Live Xbox feedback output policy on ordinary physical targets¶

The Xbox One virtual target is selected independently of physical controller family in the profile selector and OutputSlotManager. Switch 2 feedback enters its canonical HD-rumble session. Other physical targets use XboxOnePhysicalFeedbackSession: conventional targets fold impulse into their side's body motor using max; supported physical DualSense targets keep the four channels separate and use the existing opt-in TriggerLab game-rumble controls. Those are capability projections, not equivalent physical actuators or waveform fidelity claims. Legacy Nintendo's existing fixed-carrier HD writer is unchanged.

Before the September 5 fix, the profile output switch woke only the Switch 2 Xbox feedback session. An ordinary target kept its current rumble state until another game frame or its existing expiry timer arrived. The regression uses the real profile setter, ViiperOutDevice callback, canonical session/pump and ControlService rumble setter, with a recording physical target and manual clock.

The common QueueXboxFeedbackPolicyRefresh entry now wakes the existing Xbox feedback worker for both kinds of target. The ordinary-target request captures the exact physical session, slot, stream generation and publication sequence without waiting behind a state setter. The worker enters existing feedback callback admission and restricts only that current presentation through the same canonical pump. It does not mint a frame, sequence, timestamp, TTL or ACK. Failed neutral state acceptance remains retryable on that worker; expiry and terminal retirement retain their existing behavior.

The current publication identity is exposed before a live output-policy read. Thus a concurrent disable either affects that read or captures that publication for the queued restriction. Re-enabling does not remove a current restriction; a newly accepted game frame can resume, including an identical actuator state. Pending requests retain the newer sequence, and stale slot/session/stream work cannot overwrite or mutate a successor. Profile reloads call the common wake outside the Switch 2-only conversion branch as well.

This does not add an input-rate branch, input polling, another mapper or another feedback worker. The extra request allocation is on profile edits, not per input report. Ordinary target success still proves state-setter acceptance, not a completed HID write. Physical output-disable flags, disconnect behavior, persistent TriggerLab effects, every real source/target route and tactile acceptance require separate hardware evidence. The existing compatibility-matrix packet-builder tests only serialize canonical states; their names do not prove those physical routes.

Switch 2 publication-time follow-up¶

Switch 2 uses the same profile wake but retains its USB/BLE session and sole physical writer. The production Xbox callback now supplies a cached, bounded live-policy reader to that session. After authenticating the incoming frame, the session exposes its new publication revision and then samples policy under its existing gate. A concurrent disable is therefore either seen by that read or queues a restriction for that publication. No new input lock or worker is introduced; the delegate is allocated once per output device, not per report.

A late master-output disable converts Apply to canonical Neutral with unchanged sequence, source/lifetime identity, timestamp and TTL, bypasses profile delay, and clears older delayed effects/release envelopes. An impulse-only disable selects body-only rendering. Re-enabling does not undo restrictions already applied to the current presentation; fresh game feedback remains a new effect. Terminal Stop bypasses the reader entirely. Malformed, foreign and replayed frames are rejected before it. A failed policy read rejects the publication; the caller's existing NACK/retirement path still owns cleanup.

Cold refresh requests preserve newer publication revisions within an exact session/slot/stream. A CAS retry rechecks current identity before it can replace pending work, so a delayed old-owner request cannot erase a successor's wake. Both USB recording-writer and BLE production-callback tests cover the change; they do not establish physical unplug, wireless or tactile acceptance.

Detailed red/green and full-suite results are in VIIPER's dated platform ledger. This source is not part of the live b56 or staged b57 payloads.

Legacy Nintendo writer follow-up¶

The original Switch Pro and Joy-Con writers (not Switch 2) consumed and cached the new motor amplitudes before their interrupt-OUT call. They ignored its boolean result. After a failed zero report, the next pass saw unchanged zero amplitudes and skipped the write, potentially leaving a previous effect active. This affects the ordinary physical rumble mailbox used by Xbox feedback as well as other virtual targets; it is not the reported Switch 2 unplug defect.

Both writers now retain a writer-owned pending bit before preparation/write and clear it only on a successful write return. A failure or exception leaves the latest state eligible on the next existing pass. Each pass merges the current mailbox again; no old failed payload is queued ahead of a newer game effect. Success returns neutral output to its existing idle suppression, while active rumble retains its existing refresh behavior. Packet layout, frequency table, counter advancement, gain and side routing are unchanged.

LegacyNintendoRumbleDeliveryTests runs the real constructors, state mailbox, merge, packet encoder and writer decision, replacing only OS initialization and the final HID write. Thirty cases cover Pro plus left/right Joy-Con with USB/BT report sizes, rejection/exception, uncertain failed nonzero output followed by neutral, newer-state replacement, active refresh, idle suppression and publication during a write. Fifteen failure cases reproduce the old lost-stop behavior. These are simulated write results, not tactile or physical transport evidence.

There is no new worker, timer, loop, allocation or sleep in the writer path. However, these legacy writers still execute synchronously on their input loops using the pre-existing 100ms HID timeout and cancellation-completion wait. Retrying a failed neutral can therefore stall that legacy input loop on another pass. This remains an explicit latency/lifecycle acceptance gap; this fix is not evidence of a nonblocking output path, bounded kernel completion, or terminal flush after the reader has stopped. Switch 2 uses its separate owned writer. Exceptions still propagate as before; the exception test verifies that a caller which continues does not lose the pending state, not that the production input thread recovers from an unhandled HID exception.

DS4 profile-output stop follow-up¶

The DS4 base writer's separate NoOutputData issue is corrected in source: CheckProfileOptions now configures the physical DS4 profile-output policy without rewriting hardware feature bits. Its former ordering disabled the writer before the queued zero could reach it. Immutable input-only capability and a subsequently detected hardware NoOutputData flag both remain authoritative. Profiles cannot clear either limitation. Switch 2's no-HID foundation remains unchanged; other physical device classes retain their existing custom-pipeline configuration.

The DS4 policy uses a revision and the existing short rumble-mailbox lock, not a transport lock. A real toggle clears pending game/preview rumble, and disabled publication cannot be resurrected by re-enabling. The existing DS4 compositor also neutralizes direct haptic-state writes while disabled. A possible nonzero effect submission (including Bluetooth audio-mode controls) requires a final stop through the same sole writer. With no possibly active motors, disabling does not invent a new output probe; in particular, a disabled-at-startup profile does not cause another write to an unproven interface.

The stop bypasses ordinary effect coalescing, sets both motors to zero and omits lightbar/flash validity. USB/Sony-adapter and Bluetooth 0x05/0x11 layouts keep their existing encoding. The audio owner receives the corresponding 0x11 control with F1 validity, preserving audio mode and volume fields and recomputing the Bluetooth CRC. Mode changes while profile output is disabled also clamp motors and omit lightbar validity. No new HID handle, writer, timer, queue or mapping path is introduced.

A rejected/throwing stop is not acknowledged. Direct-lane failures keep their existing disconnect/error policy; if that lifetime continues, the next writer pass remains eligible. Audio-mailbox admission transfers native completion/retry ownership to that mailbox: it is not physical-flush evidence. A late result names only its captured policy revision, and a later enabled nonzero effect always requires another stop on disable. The logical controller's normal input path is not retired merely because profile effects were disabled.

DualShock4ProfileOutputStopTests uses real DS4 construction, mailbox, merge, packet encoding, CRC and audio state/mailbox, substituting only the final writer admission. Its 34 cases cover six output configurations, lost-stop reproduction, failure/throw, preview and stale-frame suppression, rapid toggles, publication during I/O, startup refusal, immutable/dynamic hardware limitations, audio-mode effects, audio-owned retry and the no-HID foundation. Three warmed checks measure zero managed allocation through 20,000 enable/effect/disable/stop cycles, with capture disabled in the fake writer. This is allocation evidence, not a latency measurement. No physical DS4 stop/feedback acceptance is claimed.

The non-audio DS4 writer still uses its existing synchronous HID path and READ_STREAM_TIMEOUT (3,000ms), plus native cancellation retirement. A required final stop can encounter that existing wait or a transport failure. Removing those input-loop stalls and proving terminal physical flush remain separate production gates; this repair does not establish low-latency parity. It does not claim to repair all master-output policy in DualSense/native or original Nintendo custom writers, or the Switch 2 unplug, wireless, game or haptic-delivery gates.