DualSense Edge fidelity audit¶

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

Goal¶

Complete evidence-backed physical DualSense Edge support and VIIPER Edge emulation without regressing ordinary DualSense, Nintendo, or other virtual outputs. Work in isolated Git worktrees, add regression tests before fixes, review integration against the current destination, and preserve unrelated work.

This is not a claim of perfect compatibility or physical hardware validation. The side conversation cannot use the app's persistent-goal mechanism; this file is its resumable objective and evidence ledger.

Source snapshots¶

• DS4Windows: ca9bf655f1fbdea3d68df26089ee5de06ff4098b.
• VIIPER: 85b6d4800d29741103747ea77f654c7e9b8d8b67.
• Isolated branch in both repositories: fix/dualsense-edge-fidelity.
• Destination worktrees have unrelated in-progress changes; do not stage, overwrite, discard, or commit those changes.

Acceptance checklist¶

• [x] Independently verify USB/BT input offsets, extra-button masks, motion, touch, trigger status, Edge profile status, battery and audio status.
• [x] Verify Edge descriptors and feature-report behavior; fix confirmed mismatches without advertising unsupported writes as successful.
• [x] Verify native trigger, rumble, LED, audio and haptics output fields and transport bounds, including compatibility with existing V5 clients.
• [x] Audit physical Edge capability detection, hardware-profile interaction, trigger stops, remapping/Special Actions and reconnect ownership.
• [x] Add and execute targeted regressions and wider relevant test suites.
• [x] Review both diffs, integrate only validated changes, and test integration.
• [x] Document hardware-dependent or unsupported behavior explicitly.

Checked items above mean source/test audit, not a physical validation pass.

Baseline findings (corrections recorded below)¶

1. Fn/paddle/mute masks agree with SDL and HIDMaestro in a static exhaustive byte-mask comparison. This is not an application or hardware test.
2. VIIPER's Edge descriptor uses base-controller lengths for output 0x02 and feature 0xF2. Captured Edge descriptors use 63 and 52 payload bytes.
3. Edge feature reads are not implemented by the advertised generic feature handler; generic feature writes currently succeed without applying state.
4. Virtual USB input byte 54 replaces all physical connection/status flags with 0x08. Audio-present/mute flags need a deliberate virtual-route policy.
5. Physical improved-rumble capability detection differs from SDL; profile application subsequently overrides the detected value. Determine actual behavior before changing it.
6. Output-selection help text incorrectly says adaptive triggers need future work. Edge-only controls currently use the mapping list, not artwork targets.

Independent references¶

• SDL: https://github.com/libsdl-org/SDL/blob/main/src/joystick/hidapi/SDL_hidapi_ps5.c
• HIDMaestro: https://github.com/hifihedgehog/HIDMaestro/blob/master/profiles/sony/dualsense-edge.json
• Captured Edge descriptor: https://github.com/ShadowBlip/InputPlumber/blob/main/src/drivers/dualsense/report_descriptor.rs
• Output/input fields: https://github.com/SpecialKO/XInput_HID/blob/master/dualsense.cpp
• Hardware constraints: https://www.playstation.com/en-au/support/hardware/set-up-edge-pc/

Reference implementations are cross-checks, not unquestioned authorities. Their field names can disagree. Preserve verified physical wire values and use source attribution/license review for any copied implementation.

Implemented corrections (2026-09-23)¶

• Edge USB report 0x02 now advertises 63 payload bytes, and 0xF2 advertises
• Standard DualSense stays at 47 and 15 respectively. Descriptor cloning remains isolated between personas and audio/gamepad variants.
• Virtual USB input retains the physical headphone-present, microphone-present and microphone-muted flags (byte 54 bits 0..2), while independently declaring a USB transport. Physical transport/unknown upper bits are not propagated.
• Unsupported Edge onboard-profile feature reads/writes now receive a real USB STALL through the existing transactional EP0 contract. A device-handler handled=false alone was insufficient: the generic HID fallback returned success. Tests verify the wire response and subsequent normal enumeration. This does NOT implement profile storage or forward writes to hardware.
• Improved rumble follows the independent SDL capability rule: Edge supported independently of its different firmware version series; ordinary DualSense needs known firmware >=0x0224, or retains the default if firmware is unknown. Applying a profile can no longer bypass a known unsupported capability; legacy rumble remains available.
• Physical firmware/calibration reads honor HID success and the requested ID. Bluetooth reads also require the Sony 0xA3 CRC and have five bounded attempts. The shared existing CRC implementation avoids depending on application-wide fast-table initialization. Calibration accepts only nonzero gyro speed and all six nonzero axis ranges; failure preserves the last good calibration.
• Output-selection help now describes supported feedback accurately and calls out the lack of virtual onboard-profile editing.

Verified report contracts¶

Offsets here include the report ID. Bluetooth 0x31 adds one byte relative to the normalized USB 0x01 input body.

Test evidence¶

• Final full .NET suite: 6,854 passed, 12 skipped, 6,866 total (Release x64). Skipped tests are not counted as proof.
• Targeted new physical capability/button/feature/calibration cases: 29 passed.
• VIIPER go test ./... -count=1: passed all packages, including ordinary DualSense, DualShock 4, Nintendo, Xbox, USB and API transport tests.
• VIIPER go test -race ./device/dualsense ./internal/server/usb -count=1: passed. This checks concurrent claim lifetime and existing streaming behavior.
• New Edge transaction tests cover forged/stale/duplicate claims, cancellation, delivery failure, bounded exhaustion/recovery and concurrent completion.
• No apps restarted, driver changes, physical profile writes, firmware updates, or hardware haptics tests were performed by this audit.

Remaining acceptance gates — goal not complete¶

1. Windows read-only PnP enumeration found no present physical Edge. USB and Bluetooth hardware checks are still required: input, both Fn/paddle pairs, touch/motion, both trigger-stop positions, native/adaptive feedback, PCM, headset status/audio/microphone, multiple stored profiles and reconnects.
2. Virtual onboard profiles (features 0x60..0x65, 0x68, 0x70..0x7B and feature 0x80's profile-snapshot subcommand 0x70) are not implemented. They now fail honestly. A software profile store and verified transaction semantics would be separate implementation work, not arbitrary writes to the user's physical controller.
3. Firmware feature 0x20 remains synthetic; this is not a captured full Edge firmware identity. HIDMaestro #47 confirms a difference in firmware series but does not supply a fully verified physical response. Do not invent one.
4. Feature 0xF2 payload semantics remain unverified. Known nonzero output extensions include Edge configuration previews and profile controls, now explicitly rejected; they are not padding and are not implemented by the common game-feedback path. Correct descriptor sizes do not prove those features work. Do not change the established V5 wire ABI to speculate.
5. USB input bytes 56..63 are a physical authentication tag. They cannot simply be copied after altering virtual state. No authentication claim is made.
6. Physical hardware profiles can remap/attenuate before DS4Windows receives input or delivers effects. Sony documents that medium/short trigger stops disable adaptive effects. These are hardware behavior, not missing bits.
7. Existing UI Charging derives from USB link status rather than the complete battery charge-state enum. Raw battery status remains intact in emulation; a UI policy change needs separate behavior checks before altering it.

Additional reference snapshots and provenance¶

• SDL c71abd08605b8bb7078372307a93274725c99fe0.
• HIDMaestro 9df50410230c11b410f43909ede0e5fc8b23d15b.
• https://github.com/daidr/dualsense-tester at f6e6247fd66ada9c8b63f3ba62c6d72945c0da53: Edge profile/trigger-stop status and feature transaction cross-check (MIT).
• https://github.com/Kurobac/edgemap at eb1d96e69c000d3423424b62ccfc4206df4053b3: USB/BT and firmware/calibration report cross-check (GPL-3.0).
• https://github.com/hifihedgehog/HIDMaestro/issues/47: explicit limitations of synthetic Edge firmware responses.

No reference source files were copied wholesale. Changes implement verified protocol facts in the existing mapping/transport stack and reuse this repository's CRC and transaction contracts. Existing third-party notices stay unchanged.

Integration evidence¶

• Source checkpoints: DS4Windows 0e2d332, VIIPER 11f0dc1.
• Integrated only the audited files into DS4Windows-rc463-mouse and VIIPER-haptics-reference. File blob hashes matched the isolated checkpoints. Unrelated staged release assets were unchanged by integration.
• Destination DS4Windows full suite: 7,022 passed, 12 skipped, 7,034 total. Evidence: artifacts/edge-audit-integration/results/edge-integration-final.trx.
• Destination VIIPER: full go test ./... -count=1 and the DualSense/USB -race suites passed after source integration.
• Integration testing initially used output under the other worktree. Five source-inspection tests consequently read that older checkout, not the current source. Moving isolated output inside the destination checkout resolved all five; no product fix or shared build-output deletion was needed.
• The main thread's release commit advanced independently to DS4Windows 65d4fde; VIIPER was at 2ea6829. Edge changes are separate from that release work. This audit does not rebuild bundled release payloads, install binaries, push commits, or publish a release.
• Remaining acceptance gates above are still open. A hardware-availability question was sent because no physical Edge was detected.

Final RC4.6.6 independent review¶

• Rechecked the pinned SDL, edgemap and dualsense-tester sources against the integrated physical/virtual input, calibration and shared output paths. Existing Fn/paddle masks, motion/touch positions, 11-byte trigger blocks, USB/BT normalization, model-specific status bytes and descriptor counts agree with those source contracts. This remains source/test evidence, not a new physical Edge capture.
• Closed the remaining USB firmware-read validation gap: HID success alone could previously accept another report ID as firmware bytes. USB firmware and calibration now both require their requested ID; Bluetooth CRC and its existing bounded retries are unchanged. No per-frame work or delay is added.
• Cross-persona native game feedback is retained in both directions. Edge configuration authorization is rejected when either the virtual source or physical recipient is Edge: a base virtual identity does not change how an actual Edge interprets USB byte 39/41 bit 7. Rejection precedes native writes and compatibility fallback. Base-to-base reports retain their uninterpreted high bits. Media-only PCM retains its independent lane and last safe state.
• The pinned DS5Dongle implementation documents feature 0x80, subcommand 0x70/0x01, preparing Edge profile snapshots. That previously reached the generic feature-command success path despite no profile store. It now gets a transactional STALL, including incomplete variants of that opcode, without changing command response, feedback or media state. Ordinary common queries and base-controller behavior are unchanged.
• Removed a remaining base/Edge model restriction from cold Bluetooth haptics converter negotiation. Although native feedback accepted the pairing, a virtual Edge on a physical standard DualSense still silently requested the legacy converter. Both physical Sony models now use the same verified native compatibility gate; USB stays native PCM, and an unverified or changed recipient cannot acquire the Sony-filtered media stream.
• Corrected virtual gyro calibration to describe the mapped values actually sent by DS4Windows: SixAxis defines 16 units per degree/second and the V5 mapper carries those calibrated values unchanged. The virtual feature 0x05 endpoints remain +/-8192, but their reference speeds are now +/-512 instead of +/-500. SDL's independent calibration equation shows the former values multiplied angular speed by 125/128 (2.34375% low) when a game recalibrated the already calibrated samples. Accelerometer scale remains 8192 units/g. Paired C# and Go tests use the same exact wire vector to cover the actual mapper, V5 decoder, USB input encoder and advertised calibration for both personas. No physical factory calibration or raw HID sensor format changed.
• Final focused .NET suite: 132 passed, 0 skipped, 0 failed (Release x64), artifacts/rc466-edge-audit/rc466-final-edge-audit.trx. It covers capability, USB/BT feature validation, native command preservation/rejection, actual cold converter negotiation and the canonical motion contract.
• Broker checks: go test ./device/dualsense ./internal/server/usb -run 'Edge|DualSense|VirtualSonyCalibration' -count=1 and the same command with -race both passed after these final corrections. Full-release suites are tracked separately by the release owner; no hardware was used by this review.
• Additional primary reference (MIT; protocol facts, not copied source): https://github.com/awalol/DS5Dongle/blob/c67c7f685fe8d8cc44f519d27710c5a639a1be7d/src/dse.cpp#L50-L75. Its complete physical profile bridge does not establish standalone virtual profile/firmware support for VIIPER; the remaining gates above still apply.