Observed on one physical unit on 2026-10-28, using the Ajazz Linux transport (feature report 8, 17 bytes) and static reading of the vendor's Windows driver for interpretation. Everything marked observed was read from, or written to and read back from, that unit. Inferred means interpretation without a direct test. Other units, firmware revisions and the wireless receiver (PID 0xFA7C) were not tested.
Register layout used below: a value is stored as a pair (value, 0x55 - value); a color is (r, g, b, 0x55 - r - g - b); all arithmetic is modulo 256.
The light on the scroll wheel is the DPI effect at EEPROM 0x4C..0x53, not the "main lighting" block at 0xA0.
| Offset | Content |
|---|---|
0x4C |
mode, as a pair |
0x4E |
brightness level, as a pair (see table) |
0x50 |
parameter, as a pair (breathing speed) |
0x52 |
padding pair (0x00, 0x55) |
Brightness levels 1 to 10 map to 0x10, 0x1E, 0x3C, 0x5A, 0x80, 0x96, 0xB4, 0xD2, 0xE6, 0xFF (table taken from the vendor driver; only levels 1 and 10 were exercised on hardware). The unit shipped with level 1, which is dim enough that stage color changes looked like no effect.
The light shows the color of the active DPI stage (0x2C + 4 * stage). Observed: changing a stage color is stored, and the scroll light shows it once brightness is raised.
| Mode | Result |
|---|---|
| 0 | off |
| 1 | steady |
| 2 | breathing |
| 3, 4, 6, 7, 8 | dark |
| 5 | steady |
The firmware accepted every value written and read it back identically. Modes above 8 were not tried.
| Parameter | Result |
|---|---|
| 1 | slow |
| 5 | about one second per cycle |
| 8 | fast (vendor default is 8) |
| 9 | slow, less than 10 |
| 10 | slow |
| 20 | very slow |
The useful range is 1 to 8. Values above 8 behave unpredictably and the app does not offer them.
The mode byte read 05 instead of 01 once, after a long session in which it had been restored to 01. It was not caused by pressing the DPI button, by command 0x02, or by writing 0xB5 or 0xB7 (each tested separately). A 7 minute watch with the mouse untouched (reads every 15 s, which may itself keep the mouse awake) showed no change in 0x04 or 0x4C..0x53. The cause is unknown.
| Offset | Observed | Notes |
|---|---|---|
0x00 |
polling rate | stored as 1, 2, 4, 8 for 1000, 500, 250, 125 Hz |
0x02 |
DPI stage count | the vendor driver accepts up to 8; its config for this model uses 6 |
0x04 |
active DPI stage | changed by the DPI button; see notifications |
0x06, 0x08 |
00 55 |
unknown, never written by this project |
0x0A |
lift-off distance | 1 or 2 mm |
0x0C..0x2B |
8 DPI slots | all 8 populated; unused slots hold the 26000 default |
0x2C..0x4B |
8 stage colors | |
0x54..0x5D |
secondary light | valid record: color, mode 0 (off), brightness level 5, parameter 3. The vendor config hides this light for this model. No physical effect was tested. |
0x60..0x9F |
key matrix, 16 entries | see below |
0xA0..0xA6 |
main light block | the vendor config sets ShowMainEffect=0 for this model; the Ajazz Linux Lighting page used to edit this block and had no visible effect |
0xA7 |
00 55 |
unknown, never written by this project |
0xA9 debounce, 0xAB light on, 0xAD idle (x10 s), 0xAF angle snapping, 0xB1 ripple, 0xB3 light off while moving |
known | |
0xB5, 0xB7 |
ff ff (uninitialized) |
the firmware accepts and stores writes to both (tested with 06 4f and 01 54, then restored to ff ff). Whether they have a physical effect is unknown; the vendor driver treats them as auto-sleep and a sensor option (Motion Sync candidate). |
Each was changed and then restored, with the scroll light watched by eye:
| Register | Change | Scroll light |
|---|---|---|
0xB3 light off while moving |
01 to 00 (it was 01, and the light stayed on while the mouse moved) |
no change |
0xAB light on |
01 to 00 |
no change |
0xAD idle timeout |
300 s to 10 s, mouse still for 20 s | no change |
These belong to the main lighting block, which the vendor config hides for this model. The app keeps them behind the experimental option.
Nothing other than 0x00..0xB4 is populated in 0x0000..0x0FFF on this unit (no key combinations, no macros). Memory above 0x1000 was not scanned.
The entries after the six physical buttons are the vendor's factory actions for keys this unit has no button for:
| Entry | Bytes | Action |
|---|---|---|
| 6 | 04 0a 03 |
three quick clicks (interval 0x0A) |
| 7 | 08 00 00 |
light on/off |
| 8 | 07 00 00 |
polling switch |
| 9 | 02 02 00 |
DPI up |
| 10 | 02 03 00 |
DPI down |
The mouse sends unsolicited reports (report id 9, command 0x0A). Observed: pressing the DPI button sends one with bit 0 set and the new stage index as value (stage 2 to 0 to 1 produced values 0 then 1), and 0x04 in EEPROM changes in step. Mouse movement, scroll and ordinary clicks produced no notification. From the vendor driver, six notification types exist; the others were not triggered.
The client now keeps these reports when a query drains them and pump() reads them without sending anything. In the app a stage change reaches the screen and a desktop toast in under a second (checked by hand on this unit).
| Command | Observation |
|---|---|
0x03 |
connection check, 01 on the cable device; 0xFA7C (receiver) did not respond |
0x04 |
battery and charging, 64 00 (100 percent, not charging) |
0x0E |
current profile, 00 |
0x0F |
state, always 01 in every state tried |
0x02 |
acknowledged with status 0 for P5 = 1 and P5 = 0; no change in any register or in 0x0F |
- Every write was preceded by a backup and followed by a read back; all values were restored afterwards except where stated.
- The app refuses to remove the last Left click and refuses to write outside
0x0000..0x00B8,0x0100..0x01BFand0x0300..0x0AFF. - Registers
0x06,0x08and0xA7were not written: an unknown register may trigger an action when written.