CraftController

PACKET EVIDENCE / 22 NOVEMBER 2026

Connected, but no input.
The slot was wrong.

Two Wireshark captures on the same phone and PC isolate the difference between the original DSU Controller and CraftController 0.7.0.

Capture A used the original DSU Controller 3.2.5. Capture B used CraftController 0.7.0 with the independent DSU probe also listening. Both captures were taken on the same LAN: phone 10.225.163.178, PC 10.225.163.175, phone listener UDP 26760. The PC capture saw both directions.

What the Manager asked for

Original app

The Manager subscribed with flags=1 slot=1 and an all-zero MAC. The phone accepted it and sent 4,931 data packets to the Manager's port 53846 over 80.4 s, about 61.4 Hz.

CraftController 0.7.0

The Manager sent the same flags=1 slot=1 subscription. The app only accepted slot 0, so every Manager subscription was ignored. The only data went to the probe on port 60998, which had subscribed with flags=0 slot=0.

The probe working was the misleading part. flags=0 is a wildcard subscription, so it bypassed the slot check. The Manager was not wildcarding; it asked for slot 1 specifically.

Side-by-side evidence

FieldOriginal 3.2.5CraftController 0.7.0
Subscription receivedflags=1 slot=1, MAC 00:00:00:00:00:00flags=1 slot=1, MAC 00:00:00:00:00:00
Subscription acceptedYesNo — slot 0 only
Data destinationManager 10.225.163.175:53846Probe 10.225.163.175:60998
Data packets4,931 in 80.4 s2,178 to the probe; 0 to the Manager
Average data rate~61.4 Hz~52.4 Hz
Data slot10
Identity in INFO/DATAstate 2, model 2, conn 1, MAC “DSU001”, battery 5state 2, model 1, conn 0, random MAC, battery 0
SequenceStarts at 1Starts at 0
TimestampLive epoch microsecondsZero
Accelerometer / gyroscopeLive valuesZeros

Counts and rates are from the captures, not from a controlled benchmark. Frame rate depends on scheduling and is not a product specification.

What changed in 0.7.1

Protocol

  • Controller slot is now 1 in INFO and DATA.
  • Subscriptions for slot 1 are accepted; slot 0 is no longer treated as the only valid slot.
  • The identity matches the reference app: fixed DSU001, model 2, connection 1, battery 5.
  • DATA carries a live epoch-microsecond timestamp and a sequence starting at 1.

Timing

0.7.0 restarted the 16.67 ms frame timer after every receive timeout, so the 4 ms socket timeout was added to each frame. 0.7.1 advances the schedule by one frame instead, with a resync guard after a real stall.

The capture measured 52.4 Hz for 0.7.0. 0.7.1 targets 60 Hz; that target has not yet been measured on a phone.

What this does not prove

A subscriber count, or even a correct data stream, does not prove that Windows created a virtual gamepad or that a game received input. Those remain separate checkpoints. The next useful evidence is:

  1. The Manager's subscriber count goes above zero while the controller is open.
  2. Windows shows a gamepad (for example in joy.cpl).
  3. A game responds to a button or stick movement.

If step 1 works and step 2 does not, the remaining problem is on the PC side of the DSU connection, not the phone's UDP transport. The app sends model 2 to match the reference identity, but accelerometer and gyroscope values remain zero, so it does not provide motion input.

Reproduce it

The comparison scripts are small Python parsers because tshark was not installed. Capture on the PC with Wireshark, filter on UDP port 26760, and decode DSUS/DSUC packets. The project notes in docs/manager-qr-findings.md and specs/revision-0.7.md describe the announcement flow.

Windows setup