KM003C firmware analysis
1. Findings for the CLI
- Decryption: the method is from okhsunrog/km003c-rs FIRMWARE.md @ c063388, and I verified it independently.
- The container is a Qt
QDataStream: tenQStrings, then oneQByteArrayof ciphertext. - The cipher is AES-128-ECB with key 0,
Lh2yfB7n6X7d9a5Z. - All 21 Drive packages decrypt to a valid Cortex-M image (base
0x4000, reset vector0x4295) whose CRC-32 matches the header. - Seven packages carry stale bytes after the ciphertext; ignore them.
- The container is a Qt
- The meter's firmware is not byte-identical to the Drive 1.9.5. The meter's app info block (captured MemoryRead of
0x4420) says 1.9.5, 2025.1.7, 451,748 bytes, stored CRCDD9AD21D. The Drive 1.9.5 is 2025.1.6, also 451,748 bytes, CRC3F85AC89. Patching only the date string does not reproduce the meter's CRC. Every finding below comes from the Drive 1.9.5 and agrees with the captured 1.9.5 frames. - ADC tail by version: the old layout is used by every image from 1.7.9 through 1.9.9 and by the Windows-app 2.0.0 (2025.2.20). The new layout starts at 2.0.1 (2026.4.9) and continues through 2.0.7. Select it with
version <= 2.0.0→ old (§8 table). On the old layout (1.9.5):0x24is ani16second-temperature slot. Its only writer stores0x8000, meaning "invalid".0x26is the CC2 50-sample average in mV, with the low 2 bits overwritten by the sample-rate index. Your0x1A= rate 2 = 50 SPS, CC2 24–27 mV.0x28and0x2Aare the D+ and D− 50-sample averages in mV.
0x24is unchanged (0x8000).0x26is the rate only.0x27is the accelerometer status and orientation.0x28is the PD-reported battery charge in per-mille, or 0.0x2Ais0xFF.
- IBUS sign (1.9.5 and 2.0.7): the firmware passes the INA229
CURRENTregister sign through unchanged. It calibrates positive and negative currents with separate gain tables. It does not decide which physical direction is positive; that takes a bench test. - Protocol shown on screen: no GetData attribute carries it.
- The 1.9.5 dashboard label is a function of (VBUS mV, IBUS mA, D+ mV, D− mV). D+ and D− are exactly payload
0x28/0x2Aon 1.9.x. The function is reproduced in §10 so the CLI can compute it. - PD details (PD3.0/3.1, PPS/QC4, AVS, Apple/OPPO/Xiaomi VIDs) come from PD messages, which the CLI receives through attribute
0x10. - The passively detected proprietary protocol state lives only in RAM.
- The 1.9.5 dashboard label is a function of (VBUS mV, IBUS mA, D+ mV, D− mV). D+ and D− are exactly payload
- Unauthenticated GetData: on 1.9.5 it is masked with
0x19, which leaves ADC0x01, settings0x08and PD0x10. AdcQueue, the PD and QC traces and the log list need auth level 1. On 1.9.x that only means echoing the MCU UID under key 3, and it has no other side effect. From 2.0.1 the mask is0x7F. - MemoryRead (0x44) needs Connect but no auth. I checked the dispatcher on 1.9.5, 1.9.9 and 2.0.7; the gate itself is identical in every image from 1.9.5 to 2.0.7. It requires magic
0xFFFFFFFF, size <0x3D0901and address <0x983D0901, with no whitelist. The reply is AES-ECB(key 0) in 2,544-byte chunks. On 2.0.5 a reply that ends in a short chunk sends that last chunk unencrypted while the header flag says plaintext (FIRMWARE.md). - Dangerous commands:
- vendor bulk interface:
0x04reset,0x09DFU,0x48settings batch (no auth: persistent settings, reboot, factory reset),0x4ANOR write and erase (auth ≥ 1),0x4Dcalibration (auth 2); - CDC virtual serial port:
0x0Asettings batch with no auth; - CDC text "protocol trigger" mode (
pdm open,entry pd,qc 20V,pd req=, ...): makes the attached charger raise VBUS.
- vendor bulk interface:
- Session on 1.9.5: the first Connect sets a 5 s receive timeout. A second Connect while connected raises it to 120 s. On timeout the firmware resets auth and sends an unsolicited
0x03. GetData works without Connect.
2. .mencrypt container
The container is Qt QDataStream, all integers big-endian. A QString is a u32 byte length (0xFFFFFFFF = null; never used) followed by UTF-16BE. A QByteArray is a u32 length followed by bytes. Layout of KM003C_V2.0.7.mencrypt; other versions differ only in string lengths:
| Offset | Size | Type | Field | 2.0.7 value |
|---|---|---|---|---|
0x000 | 4 | u32 BE | string count | 10 |
0x004 | 4+26 | QString | [0] magic | "Firmware file" |
0x022 | 4+20 | QString | [1] vendor | "ChargerLab" |
0x03A | 4+8 | QString | [2] format version | "V1.0" |
0x046 | 4+12 | QString | [3] model | "KM003C" |
0x056 | 4+10 | QString | [4] firmware version | "2.0.7" |
0x064 | 4+18 | QString | [5] build date | "2026.9.29" |
0x07A | 4+16 | QString | [6] CRC-32 of plaintext, hex | "A175E4A0" |
0x08E | 4+12 | QString | [7] plaintext size, decimal | "456896" |
0x09E | 4+216 | QString | [8] release notes, Chinese | |
0x17A | 4+586 | QString | [9] release notes, English | |
0x3C8 | 4 | u32 BE | QByteArray length | 0x0006F8D0 = 456,912 |
0x3CC | 456,912 | bytes | AES-128-ECB ciphertext | ends at EOF |
| after | 0–1,746 | bytes | stale trailer (some packages) | none |
Trailers. Drive packages 1.8.5 (226 B), 1.8.8 (258), 1.9.3 (2), 1.9.9 (4), "2.0.0" (1,746), 2.0.1 (452) and 2.0.3 (168) have bytes after the QByteArray. In 1.8.5, 1.8.8 and 2.0.1 these bytes contain ccf2649e…f1f1, the encrypted all-0xFF block, and other ciphertext. They look like the tail of a previously written, longer file that the packaging tool overwrote without truncating. Parse exactly the QByteArray length and ignore the rest. FIRMWARE.md reaches the same rule ("exclude outer file trailers").
3. Decryption
key = "Lh2yfB7n6X7d9a5Z" (protocol key 0, also the MemoryRead key)
plain = AES-128-ECB-Decrypt(key, QByteArray)[:int(QString[7])]
check = crc32(plain with bytes 0x428..0x42F set to 0xFF) == int(QString[6], 16) # zlib CRC-32
image = flash 0x00004000; padding after 'size' is 0xFF (a full extra block when size % 16 == 0)
- Source: km003c-rs FIRMWARE.md (§ Package validation). It also documents the bytes the bootloader rewrites at
0x428..0x42Fand the0x4000base. The bootloader size limit0x6FFFEwas not re-checked here. - Independent evidence of ECB before trying a key: identical 16-byte ciphertext blocks repeat within each file, for example
2752eb11…aece28 times in 2.0.7. Blocks are shared across versions and models; KM002C 2.0.0 and MI2 2.0.0 are identical in 25,299 of 25,302 blocks. The same key also decrypts KM002C, KM002C Lite and MI2. - Verification: all 21 Drive packages and the four Windows-app 3.0.0 packages decrypt to
SPin SRAM,Reset_Handler = 0x00004295, the app info block at0x4420(0x00004000, "KM003C", version, date) matching the header, and a matching CRC-32. The MemoryRead handler addresses that FIRMWARE.md lists for 1.9.9, 2.0.5, 2.0.6 and 2.0.7 (0x46CAC,0x45514,0x45524,0x455BC) are exactly where Ghidra finds them in my images, which confirms the base. - Host side: per FIRMWARE.md, the updater sends the original ciphertext and the bootloader decrypts it. CRC-32 is not a signature, so the CLI must never contain an update path.
4. All official images
"Tail" is the ADC payload layout from §8. "L0 mask" is the GetData attribute mask applied at auth level 0. "MR flag" is the MemoryRead confirmation byte 2. All 22 images were analysed.
| Package | Embedded ver / date | Size | CRC (✓) | Trailer | SP | Tail | L0 mask | GetData / MemoryRead | MR flag |
|---|---|---|---|---|---|---|---|---|---|
| 1.7.9 | 1.7.9 / 2023.7.14 | 426,928 | 65C433B5 ✓ | 0 | 0x20008500 | old | 0x19 | 0x42F14 / 0x43388 | 1 |
| 1.8.0 | 1.8.0 / 2023.8.4 | 440,832 | B513EA7F ✓ | 0 | 0x200087F8 | old | 0x19 | 0x44560 / 0x449D4 | 1 |
| 1.8.4 | 1.8.4 / 2023.9.25 | 441,480 | C7BF2374 ✓ | 0 | 0x20008648 | old | 0x19 | 0x445F4 / 0x44A68 | 1 |
| 1.8.5 | 1.8.5 / 2023.12.6 | 441,552 | 1FAF5B0B ✓ | 226 | 0x20008648 | old | 0x19 | 0x4468C / 0x44B00 | 1 |
| 1.8.6 | 1.8.6 / 2024.1.15 | 444,456 | 45E63082 ✓ | 0 | 0x20008690 | old | 0x19 | 0x44B20 / 0x44F94 | 1 |
| 1.8.7 | 1.8.7 / 2024.2.18 | 444,448 | 685B68A5 ✓ | 0 | 0x20008690 | old | 0x19 | 0x44B10 / 0x44F84 | 1 |
| 1.8.8 | 1.8.8 / 2024.7.2 | 447,816 | 87B21A94 ✓ | 258 | 0x20008698 | old | 0x19 | 0x456C0 / 0x45B34 | 1 |
| 1.8.9 | 1.8.9 / 2024.7.18 | 448,000 | A0B15145 ✓ | 0 | 0x200086A0 | old | 0x19 | 0x456F8 / 0x45B6C | 1 |
| 1.9.0 | 1.9.0 / 2024.9.13 | 450,396 | B1769B86 ✓ | 0 | 0x200067C0 | old | 0x19 | 0x46074 / 0x464B4 | 1 |
| 1.9.2 | 1.9.2 / 2024.10.25 | 450,916 | DEC1AF5E ✓ | 0 | 0x200067C0 | old | 0x19 | 0x462A4 / 0x466E4 | 1 |
| 1.9.3 | 1.9.3 / 2024.11.12 | 451,524 | 0CFDB021 ✓ | 2 | 0x200067A8 | old | 0x19 | 0x463C4 / 0x46804 | 1 |
| 1.9.5 | 1.9.5 / 2025.1.6 | 451,748 | 3F85AC89 ✓ | 0 | 0x200067A8 | old | 0x19 | 0x46494 / 0x468D4 | 1 |
| 1.9.7 | 1.9.7 / 2025.8.7 | 452,460 | 4A453F41 ✓ | 0 | 0x200067A8 | old | 0x19 | 0x46654 / 0x46A94 | 1 |
| 1.9.9 | 1.9.9 / 2025.9.22 | 453,612 | E7A75D5B ✓ | 4 | 0x200067B0 | old | 0x19 | 0x4686C / 0x46CAC | 1 |
| Win app 2.0.0 | 2.0.0 / 2025.2.20 | 439,396 | 64118EC3 ✓ | 0 | 0x200068C8 | old | 0x19 | 0x435D8 / 0x439E4 | 1 |
| Drive "2.0.0" | 2.0.1 / 2026.4.9 | 454,396 | 34049BEC ✓ | 1,746 | 0x20008658 | new | 0x7F | 0x44760 / 0x44BB0 | 1 |
| 2.0.1 | 2.0.1 / 2026.4.21 | 454,896 | AD739165 ✓ | 452 | 0x1FFFCA80 | new | 0x7F | 0x44970 / 0x44DC0 | 1 |
| 2.0.2 | 2.0.2 / 2026.4.29 | 455,020 | C28A54B1 ✓ | 0 | 0x1FFFCA80 | new | 0x7F | 0x449D8 / 0x44E28 | 1 |
| 2.0.3 | 2.0.3 / 2026.5.20 | 454,864 | 62891FA8 ✓ | 168 | 0x1FFFC9F0 | new | 0x7F | 0x44B74 / 0x44FC4 | 1 |
| 2.0.5 | 2.0.5 / 2026.7.15 | 456,424 | 01ED2B08 ✓ | 0 | 0x1FFFD260 | new | 0x7F | 0x450C4 / 0x45514 | 0 |
| 2.0.6 | 2.0.6 / 2026.8.9 | 456,480 | 18F401E1 ✓ | 0 | 0x1FFFD260 | new | 0x7F | 0x450D4 / 0x45524 | 1 |
| 2.0.7 | 2.0.7 / 2026.9.29 | 456,896 | A175E4A0 ✓ | 0 | 0x1FFFD260 | new | 0x7F | 0x4516C / 0x455BC | 1 |
The Drive file named KM003C_V2.0.0.mencrypt declares version 2.0.1 (2026.4.9) in both its header and the embedded app info. It is not the 2.0.0 (2025.2.20) that ships with the Windows app. The Drive 1.9.9 and 2.0.7 packages are byte-identical to the copies in km003c-protocol-research and the 2.0.7 package from the official update catalog. Package SHA-256 prefixes, in order: f8974b15, a9ddb5f2, 48b9a98c, b36be632, 6043bbab, d1cf2ecb, 20de23e9, 9a0da5a9, ec87e015, b0fd6412, ea6d1781, a8cb82cd (1.9.5), 71c81457, cdbdd032, e1e67c5c, b0953bf1, 7f229082, 9f73eba2, b3f6705b, 33e7034c, 7a081af8.
5. The meter's 1.9.5 vs the Drive 1.9.5
The repository fixture hardware/km003c/testdata/fw1.9.5/bulk_W6_2_plain.hex is a MemoryRead of 0x4420 captured from the user's meter. It decodes to:
00 40 00 00 ff ff ff ff a4 e4 06 00 1d d2 9a dd "KM003C" "1.9.5" "2025.1.7" "1"
base 0x4000 (unused) size 451748 CRC DD9AD21D (written by the bootloader after flashing)
The Drive 1.9.5 has the same size, but its date is 2025.1.6 and its CRC is 3F85AC89. Writing "2025.1.7" into the Drive image and recomputing gives 5097DFBD, not DD9AD21D, so the meter runs a different build of the same size. The bootloader block at 0x0420 on the meter reads "KM003C", "2.2", "2024.9.23" (bulk_W6_1_plain.hex). Nothing visible in the captures contradicts the Drive 1.9.5 analysis:
- ADC tail
00 80 1a 00 56 00 5a 00; - PD preamble
u32 tick, u16 vbus_mV, i16 ibus_mA, u16 cc1, u16 cc2; - a 180-byte settings block;
- MemoryRead confirmation byte 2 =
0x01; - GetData answered on bulk, HID and CDC alike.
A read-only MemoryRead of the meter's application flash (0x4000..0x72CA4, about 450 KB) compared against the decrypted Drive 1.9.5 (plaintext sha256 22281fde…443d) would settle it. That is a decision for whoever owns the meter.
6. USB command dispatch per interface (1.9.5)
The 1.9.5 USB task is at 0x0001EDD4. It waits for an event, with the session timeout as the wait time, and serves three inputs. The table shows what each interface accepts:
| Interface | RX buffer | Handler | Accepted |
|---|---|---|---|
IF0 vendor bulk, EP 0x01/0x81 (gopower-z) | 0x20010000 | FUN_00052514 | full command set below |
| CDC virtual serial port, OUT EP 2 | 0x20010A00 (256 B) | inline in the task | 0x0C GetData (with bit 7: encrypted sub-read limited to XIP, OTP 0x3000C00+<0x3BF and flash < 0x7D000, size < 0xFF0); 0x0A settings batch, no auth; anything else goes to the text parser FUN_00042df2 (§12) |
| HID (IF3), OUT EP 5, 64-byte reports | 0x20011600 | inline | 0x0C GetData only |
The OUT endpoint-to-buffer mapping comes from the receive-complete callback FUN_00051ee8: EP 1 → bulk, EP 2 → CDC, EP 5 → HID. In 1.9.5 the CDC branch has no binary 0x02 handler, so such a frame reaches the text parser and is ignored there. 2.0.x added CDC 0x02/0x03 start/stop streaming. The repository's guard.go says CDC 0x02 starts a push stream; that matches 2.0.x, not this 1.9.5 code path.
The bulk dispatcher, FUN_00052514 in 1.9.5, is identical to FUN_00052af0 in 1.9.9. If the connected bit (0x20003CD6 bit 2) is clear, it sets auth to 0 and accepts only 0x0C and 0x02; everything else gets Reject 0x06.
| Type | Name | 1.9.5 behaviour |
|---|---|---|
0x01, 0x0A, 0x12, 0x13 | ignored, no reply | |
0x02 | Connect | from unconnected: set connected, receive timeout 5,000 ms; while connected: clear flags 0xFFFC, timeout 120,000 ms; reply Accept 0x05 |
0x03 | Disconnect | Only accepted while connected; otherwise Reject 0x06. It calls the same routine as the idle timeout, FUN_0001c374, which:
0x81. 2.0.7 (FUN_00053f84) does the same but clears only flags 0x64. |
0x04 | Reset | NVIC_SystemReset (FUN_00011de4) |
0x09 | JumpDfu | 0x12345678 → 0x200F0FF0 (retention SRAM), reset into the bootloader |
0x0C | GetData | FUN_00046494(src, attr), §7 |
0x0D | GetFile | FUN_00046872: attribute bit 0 gives the log catalog, otherwise an empty 0x41 |
0x0E / 0x0F | Start/StopGraph | sets the live rate (the rate setter FUN_00018138 resets the averages) |
0x10 / 0x11 | PdEnable/Disable | No auth check. PdEnable is rejected (0x06) if the PD flag (0x20003CD6 bit 4) is already set, or if attr is 0. Otherwise it sets that flag (FUN_0005243a), which mainly drives a status-bar icon, and then:
Recording does not depend on PdEnable. The PD task |
0x40 | Head | reply 0x01000040, 0x1000 (APP mode) |
0x44 | MemoryRead | §11; connected, no auth |
0x48 | SettingsBatch | FUN_00046a1c, no auth at dispatch (per-op gates in §12) |
0x4A | FlashWrite | auth ≥ 1; external NOR 0x300000–0x37FFFF, ≤ 256 B, erases on a 64 KiB-aligned address (FUN_0000d18c, FUN_0000d364) |
0x4B | NOR read | MemoryRead of 0x98000000 + addr |
0x4C | Auth | decrypts 32 B with Fa0b4tA25f4R038a. Bytes 8–19 equal to the UID (0x40010450) give level 1; equal to the first 12 bytes of the OTP calibration record they give level 2. Replies 0x4C, attribute 0x201|level<<1, encrypted with FX0b4tA25f4R038a. On 1.9.x nothing else changes; 2.0.7 also stores a wall-clock offset from bytes 0–7. |
0x4D | SettingsBatch | same handler, only at auth 2 |
For types ≥ 0x40, header bit 16 means the payload is AES-ECB(key 0). Requests without bit 16 are accepted as plaintext.
2.0.7 differences. One task, 0x0001F894, handles all three inputs. The CDC path gained Start/Stop streaming (0x02/0x03), reset/DFU, an authenticated memory or NOR read (0x0D), device info (0x14), an alternative auth (0x15) and NOR write (0x16), with a CRC-8 (poly 0x29) in header byte 3 for some commands. The bulk session deadline is 5 s after the last command, checked when the 10 s receive wait expires.
7. GetData attributes
| Attr | Content (1.9.5; same layout in 2.0.7) | Level 0 on 1.9.5 | Level 0 on ≥ 2.0.1 |
|---|---|---|---|
0x001 | ADC, 44 B (§8) | yes | yes |
0x002 | AdcQueue, 20 B each: u32 uptime_ms, i32 vbus_uV, i32 ibus_uA, u16 cc1, cc2, d+, d− | no | yes |
0x008 | settings-A (96 B) + settings-B (84 B) = 180 B | yes | yes |
0x010 | u32 uptime_ms, u16 vbus_mV, i16 ibus_mA, u16 cc1_mV, u16 cc2_mV + PD event records | yes | yes |
0x020 | two PD trace queues (u8 len + ≤ 200 B each) | no | yes |
0x040 | u32 uptime_ms, u16 vbus_mV, i16 ibus_mA, u16 d+_mV, u16 d−_mV + D+/D− protocol-engine events | no | yes |
0x080 | u32 uptime_ms, u32 byte at 0x40054062 (clock trim) | no | no |
0x200 | offline-log catalog, N×48 B | no | no |
0x800 | bare header 0x00000800 | no | no |
Masked attributes are dropped silently. Your capture of attribute 0x11 returned the PD preamble tick 0x1EA9A8, vbus 6 mV, ibus 0 mA, cc1 70 mV, cc2 25 mV, which matches this layout.
8. ADC payload and the 0x24–0x2B tail
The payload is a memcpy of the first 44 bytes of the ADC state struct, followed by overrides. The struct is at 0x20002560 on 1.9.5, 1.9.7 and 1.9.9 (accessor FUN_000180ac on 1.9.5) and at 0x1FFFACF0 on 2.0.7. Struct fields come from the 1.9.5 ADC task 0x0001D934 and the measurement function FUN_00017e20:
| Offset | Type | Meaning |
|---|---|---|
0x00 / 0x04 | i32 | VBUS µV (calibrated, clamped ≥ 0) / IBUS µA (calibrated, signed) |
0x08 / 0x0C | i32 | 1-second averages (window = SPS samples), calibrated |
0x10 / 0x14 | i32 | 1-second averages, uncalibrated ("ori") |
0x18 | i16 | INA229 die temperature, 1/128 °C; 0x8000 right after a rate change |
0x1A–0x20 | u16×4 | CC1, CC2, D+, D− in 0.1 mV (refreshed every 250 ADC iterations) |
0x22 | u16 | internal VDD |
0x24–0x2A | u16×4 | CC1, CC2, D+, D− in mV, 50-sample averages |
0x2E–0x35, 0x38–0x3F | u16×4 | CC1, CC2, D+, D− instantaneous mV (two copies) |
0x4C | i16 | second-temperature slot; the only writer (rate setter) stores 0x8000; Ghidra xrefs and a MOVW/MOVT scan found no other |
0x4E | u8 (1.9.x) / u16 (2.x) | bits 0–1 rate index; 2 new sample; 3 ADC done; 4–6 average ready; 2.0.1+ bits 7–10 accelerometer I²C status, 11–14 orientation |
0x66 / 0x68 (1.9.5); 0x68 / 0x6A (2.0.7) | u16 / i16 | VBUS mV (INA229 raw×100/512) / IBUS mA; used by PD preambles and the dashboard |
Old tail: 1.9.5, 1.9.7, 1.9.9, Windows-app 2.0.0
/* 1.9.5 FUN_00046494 */
memcpy(p, adc, 0x2c);
*(u16*)(p+0x18) = adc[0x18];
*(u16*)(p+0x24) = adc[0x4c]; /* 0x8000 */
p[0x26] = (adc[0x4e] & 3) | (p[0x26] & 0xFC); /* rate index into the CC2 average's low bits */
| Payload | Meaning | Your capture 00 80 1a 00 56 00 5a 00 |
|---|---|---|
0x24 i16 | second temperature; 0x8000 = invalid or absent | 0x8000 |
0x26 bits 0–1 | sample-rate index | 2 = 50 SPS |
0x26 u16 & 0xFFFC | CC2 50-sample average, mV, low 2 bits lost | 24 (true 24–27) |
0x28 u16 | D+ 50-sample average, mV | 86 |
0x2A u16 | D− 50-sample average, mV | 90 |
New tail: every build from 2.0.1 (2026.4.9) through 2.0.7
/* 2.0.7 FUN_0004516c; identical in Drive "2.0.0" (=2.0.1), 2.0.1, 2.0.2, 2.0.3, 2.0.5, 2.0.6 */
memcpy(p, adc, 0x2c); p.u16[0x18] = adc[0x18]; p.u16[0x24] = adc[0x4c]; /* 0x8000 */
p.u16[0x28] = (pd[0x29]==0x79 && full!=0 && full!=0xFFFF && present<=full) ? present*1000/full : 0;
/* pd = PD object; full = +0x460 (Battery_Capabilities last-full capacity),
present = +0x462 (Battery_Status present capacity) */
p[0x26] = adc[0x4e] & 3; /* rate only */
p[0x27] = (adc[0x4e] >> 7) & 0xFF; /* lo: SC7A20 I2C status (0 = ok), hi: orientation 0-3 or 0xF = flat */
p[0x2a] = 0xFF; /* p[0x2b] = stale high byte of the D- average */
0xF001: rate 1 (the boot default), accelerometer OK, orientation 0xF (flat). Pick the layout by firmware version: ≤ 2.0.0 old, ≥ 2.0.1 new (see the version table below). The rate, p[0x26] & 3, is valid in both. On 2.0.1+ read D+/D− from 0x1E/0x20 (0.1 mV).Rate index: the per-rate SPS table ({2, 10, 50, 1000}; 1.9.9 0x55DF8, 2.0.7 0x558E6) matches the UI strings "2SPS", "10SPS", "50SPS" and "1KSPS". AdcQueue line voltages come from struct offset 0x1A + 10×rate:
- rate 0: 0.1 mV, averaged;
- rate 1: mV, 50-sample average;
- rates 2 and 3: instantaneous mV.
That is why the queue units change with rate.
Layout by version
Each of the 22 images was checked the same way. I located the GetData builder (extended header 0x8001/0xB00, 44-byte copy) and read its tail overrides, found every writer of ADC struct +0x4C, and read the SPS table the rate setter indexes. 1.7.9–1.8.9 copy the 44 bytes as eleven words rather than calling memcpy; the overrides are the same.
| Firmware (date) | Tail layout | GetData | 0x24 | 0x26 | Level-0 mask | Dashboard label |
|---|---|---|---|---|---|---|
| 1.7.9 (2023.7.14) | old | 0x42F14 | adc+0x4C = 0x8000 | CC2 avg, rate in bits 0–1 | 0x19 | 1.9.5 logic |
| 1.8.0 (2023.8.4) | old | 0x44560 | same | same | 0x19 | 1.9.5 logic |
| 1.8.4 (2023.9.25) | old | 0x445F4 | same | same | 0x19 | 1.9.5 logic |
| 1.8.5 (2023.12.6) | old | 0x4468C | same | same | 0x19 | 1.9.5 logic |
| 1.8.6 (2024.1.15) | old | 0x44B20 | same | same | 0x19 | 1.9.5 logic |
| 1.8.7 (2024.2.18) | old | 0x44B10 | same | same | 0x19 | 1.9.5 logic |
| 1.8.8 (2024.7.2) | old | 0x456C0 | same | same | 0x19 | 1.9.5 logic |
| 1.8.9 (2024.7.18) | old | 0x456F8 | same | same | 0x19 | 1.9.5 logic |
| 1.9.0 (2024.9.13) | old | 0x46074 | same | same | 0x19 | 1.9.5 logic |
| 1.9.2 (2024.10.25) | old | 0x462A4 | same | same | 0x19 | 1.9.5 logic |
| 1.9.3 (2024.11.12) | old | 0x463C4 | same | same | 0x19 | 1.9.5 logic |
| 1.9.5 (2025.1.6) | old | 0x46494 | same | same | 0x19 | reference |
| 1.9.7 (2025.8.7) | old | 0x46654 | same | same | 0x19 | 1.9.5 logic |
| 1.9.9 (2025.9.22) | old | 0x4686C | same | same | 0x19 | 1.9.5 logic |
| 2.0.0 (2025.2.20, Windows app) | old | 0x435D8 | same | same | 0x19 | 1.9.5 strings and thresholds, inputs computed differently (see §10) |
| 2.0.1 (2026.4.9, Drive file "V2.0.0") | new | 0x44760 | adc+0x4C = 0x8000 | rate only; 0x27 accelerometer | 0x7F | 2.0.7 logic |
| 2.0.1 (2026.4.21) | new | 0x44970 | same | same | 0x7F | 2.0.7 logic |
| 2.0.2 (2026.4.29) | new | 0x449D8 | same | same | 0x7F | 2.0.7 logic |
| 2.0.3 (2026.5.20) | new | 0x44B74 | same | same | 0x7F | 2.0.7 logic |
| 2.0.5 (2026.7.15) | new | 0x450C4 | same | same | 0x7F | 2.0.7 logic |
| 2.0.6 (2026.8.9) | new | 0x450D4 | same | same | 0x7F | 2.0.7 logic |
| 2.0.7 (2026.9.29) | new | 0x4516C | same | same | 0x7F | reference |
The switch happened between 2.0.0 (2025.2.20) and 2.0.1 (2026.4.9). No official build exists between those two dates. Version strings are the reliable selector: <= 2.0.0 uses the old layout and >= 2.0.1 the new one. Both 2.0.1 builds use the new layout.
0x24in both layouts: the payload copies ADC struct+0x4C. In all 22 images the only writer of that field is the rate setter, which stores0x8000. The field is never filled with a measurement, so on a KM003C it always reads0x8000. The official app treats the i16 at0x24as a second temperature and0x8000as invalid, so "i16 second temperature,0x8000= invalid" is the right decoding for both layouts.- Rate in the old layout: the firmware writes
(adc[0x4E] & 3) | (p[0x26] & 0xFC)into byte0x26, the low byte of the little-endian u16 at0x26. The rate is therefore bits 0–1 of that u16, and the CC2 average isu16 & 0xFFFC. In the new layout byte0x26holds the rate alone. - SPS mapping: every image carries the same per-rate table
{2, 10, 50, 1000}, indexed by that rate value, and the strings "2SPS", "10SPS", "50SPS" and "1KSPS". The mapping 0 = 2, 1 = 10, 2 = 50, 3 = 1000 SPS holds in all versions.
9. IBUS sign
- INA229 read (1.9.5
FUN_00006e6e, 2.0.7FUN_00006d78): reads register 7 (CURRENT, 20-bit two's complement) and sign-extends it. It switchesADCRANGEnear full scale and rescales by 4. It returns the value without negation. - Calibration (1.9.5
FUN_00017e20, 2.0.7FUN_00017b1c):ibus_uA = gain[k]×(raw − offset)/1e6, sign kept.kpicks one of 10 gains: five positive ranges with breakpoints at 2.1, 4.1, 6.1 and 8.2 A, and the mirrored negative ranges. Settings op0x0Dwrites the negative gains. - UI: prints the signed value (
"%5.3fV %6.3fA %0.1fW"). There is no direction text or arrow logic.
So the meter measures both directions symmetrically and the firmware defines no physical convention. Which way is positive depends on how the shunt is wired. The official "new ADC" PDF (negative = receptacle → plug) and the km003c-rs comments (positive = female → male, which is the same thing stated the other way) conflict on paper. A bench test with a known load direction settles it. Until then the CLI should report the raw sign.
10. Fast-charge protocol labels
No GetData attribute carries the dashboard's protocol text or the engine's detected-protocol state. The engine object is at 0x200056B0 on 1.9.5 (FUN_0004f8c6) and 0x1FFFC140 on 2.0.7. Reading it needs a version-specific RAM MemoryRead, which is not recommended. What the CLI can do instead:
Dashboard voltage label, 1.9.5 (FUN_0001fb44, called at 0x0004b476 with the ADC struct)
Inputs:
v= VBUS mV (struct0x66; use payloadvbus_uV/1000);i= IBUS mA as u16 (struct0x68; negative currents compare as large);dp,dm= D+/D− 50-sample averages in mV (payload0x28/0x2A).
This is a branch-by-branch translation:
line1(v, i, dp, dm):
if v < 3300 -> "[ N/A ]"
if dp > 2800:
if dm > 2800:
if v >= 19001 -> "QC2.0 20V"
if v > 7000 && i > 1000 -> "VFCP | UFCS"
return v < 5500 ? "DCP 1.5A" : "[ N/A ]"
if 451 <= dm <= 1099: return v > 8500 ? "QC2.0 9V" : "VOOC"
return v > 8500 ? "SVOOC" : "[ N/A ]"
if 2501 <= dp <= 2800: return (v < 5500 && 2501 <= dm <= 2899) ? "Apple 2.4A" : "[ N/A ]"
if 1101 <= dp <= 1300: return (v < 5500 && 1101 <= dm <= 1299) ? "Samsung 2.0A" : "[ N/A ]"
if 501 <= dp <= 1000:
if dm > 2800 -> "QC3.0 | QC3+"
if 501 <= dm <= 699:
if 10501 <= v <= 12499 -> "QC2.0 12V"
return 4501 <= v <= 5499 ? "DCP 1.5A" : "[ N/A ]"
if dm < 200:
if i >= 2001 -> "QC2.0 | SCP"
if 8501 <= v <= 12499 -> "FCP | AFC"
return 4501 <= v <= 5499 ? "QC2.0 5V" : "[ N/A ]"
return "[ N/A ]"
line2(v, i, dp, dm):
if v < 3300 -> "[ N/A ]"
if 8501 <= v <= 9999 && 1100<=dp<=1699 && 1100<=dm<=1699 -> "SFCP 9V"
if 10000 <= v <= 12499 && 2100<=dp<=2699 && 2100<=dm<=2699 -> "SFCP 12V"
if v >= 12500 && dp < 700 -> "FCP | SCP"
if i <= 2000:
alt = "[ N/A ]"
if |dp - dm| < 100:
if 4501 <= v <= 5499: alt = "DCP 1.5A"
elif v < 19500 && dp > 3000: alt = "QC2.0 20V Err"
return (501 <= dp <= 699 && dm < 300) ? "FCP | SCP | AFC" : alt
if dp > 2900: return v >= 7001 ? "SVOOC" : (v < 5500 ? "VOOC" : "[ N/A ]")
return "[ N/A ]"
The function first checks whether the engine reports an identified proprietary session (engine +0xD == 0x13). If so, it shows VOOC/SVOOC, VFCP %d.%dW or UFCS %d.%dW from engine state instead of the labels above.
Same logic in other versions. I normalised the decompiled function in every image by stripping addresses and variable numbers, then diffed it against 1.9.5.
- 1.8.8, 1.8.9, 1.9.0 and 1.9.2 are identical to 1.9.5.
- 1.7.9–1.8.7, 1.9.3, 1.9.7 and 1.9.9 differ only in how Ghidra renders constants, or in the format string of the proprietary (UFCS/PPS/SVID) branch. The voltage classifier, its thresholds and its input fields are the same.
- The Windows-app 2.0.0 has the same strings and threshold constants. It computes VBUS and D+/D− from another structure (µV/1000), so it was not compared line by line.
Dashboard voltage label, 2.0.1–2.0.7 (FUN_00022098 line 1, FUN_00021bd4 line 2, called from FUN_000213c8)
Inputs:
v= VBUS mV (struct0x68; use payloadvbus_uV/1000);dp,dm= D+/D− 50-sample averages in mV (struct0x28/0x2A).
The 2.0.x payload no longer exports dp/dm. The closest exported values are payload 0x1E/0x20 divided by 10: 0.1 mV values refreshed every 250 ADC iterations. There is no current input. Both functions normalise to identical text in all seven 2.0.1–2.0.7 images.
line1(v, dp, dm): // FUN_00022098
if v < 3300 -> "[ N/A ]"
if dm > 2800 && 501 <= dp <= 699 -> "QC3.0 | QC3+"
if v < 5500:
if dp < 110 && dm < 100 -> "DCP"
if dp > 2500 && 2501 <= dm <= 2899 -> "Apple 2.4A"
if dm < 100 && 501 <= dp <= 699 -> "QC2.0 5V"
return |dp - dm| < 100 ? "DCP" : "[ N/A ]"
if 8501 <= v <= 9499:
if dp > 2800 && 451 <= dm <= 1099 -> "QC2.0 9V"
if 1101 <= dp <= 1299 && 1101 <= dm <= 1299 -> "Samsung 2.0A"
if dm < 100 && 501 <= dp <= 699 -> "FCP | AFC"
if 501 <= dp <= 699 && 501 <= dm <= 699 -> "QC2.0 12V Error"
return (dp > 2800 && dm > 2800) ? "QC2.0 20V Error" : "[ N/A ]"
if 11001 <= v <= 12999:
if 501 <= dp <= 699 && 501 <= dm <= 699 -> "QC2.0 12V"
return (501 <= dp <= 699 && dm < 100) ? "FCP | AFC" : "[ N/A ]"
if 18501 <= v <= 20999 && dp > 2800 && dm > 2800 -> "QC2.0 20V"
return "[ N/A ]"
line2(v, dp, dm): // FUN_00021bd4
if v < 3300 -> "[ N/A ]"
if 8501 <= v <= 9999 && 1100<=dp<=1699 && 1100<=dm<=1699 -> "SFCP 9V"
if 10000 <= v <= 12499 && 2100<=dp<=2699 && 2100<=dm<=2699 -> "SFCP 12V"
if v >= 12500 && dp < 700 -> "FCP | SCP"
return (dm < 300 && 501 <= dp <= 699) ? "FCP | SCP | AFC" : "[ N/A ]"
In 2.0.x the dashboard uses these labels only while the engine's protocol code (+0x234) is not one of the values below. Otherwise it shows AFC %dV (2), FCP %dV (3), Huawei SCP (4), UFCS %d.%dW (5), VFCP %d.%dW (6), TFCP %d.%dW (8) or VOOC/SVOOC (10), all taken from RAM state that is not on USB.
Differences from 1.9.5:
- no current input, so no "QC2.0 | SCP" and no "VFCP | UFCS";
- the new labels "DCP", "QC2.0 12V Error" and "QC2.0 20V Error";
- no "DCP 1.5A", "QC2.0 20V Err" or voltage-based VOOC/SVOOC;
- different VBUS windows.
Use the classifier that matches the meter's firmware generation.
Label variant by version
Method: normalise each image's decompiled label functions (strip addresses and variable numbering) and diff them against the 1.9.5 and 2.0.7 references, including the 2.0.x wrapper that picks between the protocol-code labels and the voltage classifier. The arguments come from each wrapper's call site.
| Firmware (date) | Variant | Functions (line 1 / line 2 / wrapper) | Inputs | Difference from the reference |
|---|---|---|---|---|
| 1.7.9–1.9.9 | 1.9.5 | one function (1.9.5: 0x1FB44) | VBUS mV, IBUS mA (u16), D+/D− 50-sample avg | none in the classifier; only the format strings of the proprietary branch differ |
| 2.0.0 (2025.2.20, Windows app) | 1.9.5 constants | 0x20728 | same quantities, computed from another structure | not proven identical line by line |
| 2.0.1 (2026.4.9, Drive "V2.0.0") | 2.0.7 | 0x21C50 / 0x2182C / 0x21020 | struct +0x68, +0x28, +0x2A | none (0 diff lines in all three) |
| 2.0.1 (2026.4.21) | 2.0.7 | 0x21DE4 / 0x21958 / 0x2114C | same | none |
| 2.0.2 (2026.4.29) | 2.0.7 | 0x21E48 / 0x219BC / 0x211B0 | same | none |
| 2.0.3 (2026.5.20) | 2.0.7 | 0x21D54 / 0x218C8 / 0x210BC | same | none |
| 2.0.5 (2026.7.15) | 2.0.7 | 0x21FF0 / 0x21B64 / 0x21358 | same | none |
| 2.0.6 (2026.8.9) | 2.0.7 | 0x21FF8 / 0x21B6C / 0x21360 | same | none |
| 2.0.7 (2026.9.29) | reference | 0x22098 / 0x21BD4 / 0x213C8 | struct +0x68, +0x28, +0x2A | — |
The label variant changes at exactly the same version as the ADC payload tail, so choosing it by layout (<= 2.0.0 → 1.9.5 logic, >= 2.0.1 → 2.0.7 logic) is correct for every official build.
0x4A780, pointer from FUN_00017dae). It passes +0x28/+0x2A, the 50-sample D+/D− averages that the ADC task fills, which are the same fields 1.9.x exports at payload 0x28/0x2A. 2.0.x no longer exports them in the ADC payload, so a host has two approximations:
- payload
0x1E/0x20divided by 10, a 0.1 mV value refreshed every 250 ADC iterations; - AdcQueue (
0x02, readable without auth on 2.0.x) records taken while the rate index is 1. At rate 1 the four line voltages in each record are exactly struct+0x24..+0x2A, the 50-sample averages. Changing the rate to get there is a state change.
PD label (1.9.5 FUN_0002768c; 2.0.7 FUN_00025590)
The label is the PD revision ("PD3.0", or "PD3.1" with "EPR"; 2.0.x adds "PD3.2") plus suffixes:
PPS,QC4orPPS,QC4+(2.0.x addsPPS,QC5) when an APDO is present, orAVSwhen the APDO type bits say so;Applewhen a VID stored in the PD object is0x05AC;SVOOC,%dWfor0x22D9(OPPO) andXIAOMI,%dWfor0x2717.
All of these come from PD traffic, which the CLI sees with PdEnable(1) and attribute 0x10. I did not trace which message fills the VID field; Discover Identity or Source_Capabilities_Extended are the candidates.
11. MemoryRead (0x44) validation
/* 1.9.5: dispatcher case 0x44 + FUN_000468d4. Identical logic in every analysed version. */
req = (hdr bit16) ? AES-ECB-Decrypt(key0, payload) : payload // {u32 addr, u32 size, u32 magic, u32 tag}
if (addr - 0x03000C00) < 0x180: // OTP calibration window
p = first 64-byte record from 0x03000C00 whose first word is 0xFFFFFFFF (max 0x03000D80)
if p > 0x03000D40: addr = 0x03000D80
if magic == 0xFFFFFFFF && size < 0x3D0901 && addr < 0x983D0901:
send {0xC4, id, 0x01, 0x01} + {addr, size, 0xFFFFFFFF, tag} // 20 bytes; byte 2 bit 0 = encrypted
for each 0x9F0-byte chunk (last rounded up to 16): send AES-ECB-Encrypt(key0, mem) // raw, no header
else: send Reject {0x06, id}
- The only requirement is the connected state; there is no auth check. There is no lower bound and no whitelist: flash, bootloader, OTP, SRAM, peripheral registers and the XIP window are all readable. Reading FIFO or clear-on-read registers changes peripheral state.
- The copy loop (
FUN_00005090on 1.9.5) dereferences the address directly, so an unmapped address raises a bus fault. km003c-protocol-research saw a NOT_READABLE (0x27) reply for one such address on 1.9.9, and the repository's guard notes report crashes. Either way, whitelist. - Large sizes, up to about 3.8 MB, keep the USB task busy.
- The tail is padded to 16 bytes.
0x4Bapplies the same check to0x98000000 + addr.- 2.0.5 sets the confirmation flag to 0 and sends its final short chunk unencrypted. FIRMWARE.md documents this; its library limits requests to 2,528 bytes to avoid it. The version table in §4 confirms the flag per image.
0x0000–0x7FFFF, OTP 0x03000C00–0x03000FFF, SRAM and the 12-byte UID at 0x40010450. Keep each request ≤ 2,528 bytes and never touch other peripheral space. Useful fixed reads: 0x0420 (bootloader info) and 0x4420 (app model, version, date, installed size and CRC).12. State-changing commands
0x04and0x09;0x48and0x4D(every op changes state);0x4A;- level-2 auth;
- CDC
0x0A; - any CDC text line.
0x0E/0x0F) changes the live sample rate, which also shows on screen, but is not persisted.Settings-batch operations (1.9.5 FUN_00046a1c; identical in 2.0.7 FUN_00045704)
Format: [op:15 | next:1 | rsv:6 | size:10] + payload, chained while next is set. The handler is reachable from bulk 0x48 (no auth), bulk 0x4D (auth 2) and CDC 0x0A (no auth).
| Op | Gate | Effect |
|---|---|---|
0x01 | none | settings-A bit 0 (language) |
0x02 | none | clear settings-A bit 2 (calibration-required flag) |
0x03 | none | settings-A bits 3–9 (brightness) |
0x04, 0x05 | none | settings-A bits 10–11, 12–13 |
0x06, 0x1C | none | settings-A bits 14–15, applied |
0x07 | none | settings-A word 1 bits 0–2 |
0x08 | none | save, reinitialise, reset MCU |
0x09 | none | factory reset of both settings blocks, then op 8 |
0x0A | auth 2 | write the 5 VBUS calibration offsets; may re-zero the current offset |
0x0B, 0x0C | auth 2 | write positive-current gains |
0x0D | auth 2 | write negative-current gains; brightness forced to 50 |
0x0E | auth ≥ 1 | unlock PWC.FPRC and write a signed byte to CMU register 0x40054062 (clock trim; exact name unconfirmed) |
0x0F | auth 2 | write the current zero offset (settings-A 0x58) |
0x14–0x17 | none | settings-B: screen orientation, default sample rate (applied at once), bits 4–5, main page; persisted to NOR |
0x28 | none | append 48-byte offline-log catalog entries |
0x7FFF | none | persist settings-A |
The OTP calibration record that level-2 auth compares against is itself readable through 0x44. Level 2, and with it calibration writes, is therefore reachable by any host.
CDC text mode: "Protocol Trigger by Virtual Serial Port"
On the CDC port, any frame that is not a 0x0C or 0x0A header goes to the text parser (1.9.5 FUN_00042df2, 2.0.7 FUN_00040718). ChargerLAB's own document, "KM003C/002C Protocol Trigger by Virtual Serial Port (Instructions)", published on their Google Drive next to the firmware, lists the commands:
- session:
pdm open/pdm close,pdm set type=,em=(eMarker simulation up to 50 V 5 A EPR),reset; - protocol entry:
entry pd|ufcs|qc|fcp|scp|afc|vfcp|sfcp; - fixed voltages:
qc 5V|9V|12V|20V,fcp|afc|sfcp ?V; - stepped QC3:
qc3 volt=3600..20000,qc3 inc= / dec=; - variable voltage and current:
scp volt=,cur=,vfcp volt=,cur=,ufcs req=,volt=,cur=,pd req=,volt=,cur=; - raw PD:
pd pdo,pd cmd=,pd data=(raw PD messages).
The firmware parser also knows debug, list, ver, info, hrst (hard reset), drp, tfcp, apple and bc.
These commands make the meter act as a decoy sink and negotiate higher voltages, up to 20 V, or 48–50 V with EPR, onto whatever is connected on the output side.
13. Address index
| What | 1.9.5 (Drive, 2025.1.6) | 2.0.7 |
|---|---|---|
| USB task (bulk, CDC and HID inputs) | 0x0001EDD4 | 0x0001F894 |
| Bulk dispatcher | 0x00052514 | inline in task |
| GetData / MemoryRead / SettingsBatch | 0x00046494 / 0x000468D4 / 0x00046A1C | 0x0004516C / 0x000455BC / 0x00045704 |
| ADC struct / accessor | 0x20002560 / 0x000180AC | 0x1FFFACF0 / 0x00017DAE |
| ADC task / measurement / rate setter | 0x0001D934 / 0x00017E20 / 0x00018138 | 0x0001DE5C / 0x00017B1C / 0x00017E34 |
| INA229 CURRENT read | 0x00006E6E | 0x00006D78 |
AES-ECB decrypt / encrypt (HW AES 0x40008000) | 0x00004FB0 / 0x00005090 | 0x00004F9C / 0x00005074 |
| Auth level byte / session flags | 0x20004041 / 0x20003CD6 | 0x1FFFB485 / 0x1FFFB1A2 |
| Disconnect + timeout | 0x0001C374 | 0x00053F84 |
| NVIC_SystemReset | 0x00011DE4 | 0x000119D8 |
| Dashboard voltage label / PD label | 0x0001FB44 / 0x0002768C | 0x000213C8 (0x22098, 0x21BD4) / 0x00025590 |
| D+/D− protocol engine object | 0x200056B0 | 0x1FFFC140 |
| CDC text parser | 0x00042DF2 | 0x00040718 |
NOR program / erase (QSPI 0x9C000000) | 0x0000D364 / 0x0000D18C | 0x0000CE84 / 0x0000CCAA |
14. Not established
- Whether the meter's 1.9.5 (2025.1.7, CRC
DD9AD21D) differs in code from the Drive 1.9.5 (2025.1.6). §5 describes a read-only way to check. - Which physical direction a positive IBUS means (§9).
- The name of PD state
0x79that gates the 2.0.1+ battery field, and the CMU register name at0x40054062. - The Windows-app 2.0.0 dashboard classifier has the same constants as 1.9.5, but its inputs are computed differently, so it was not proven equivalent line by line.
- The CDC/HID interface numbering beyond what the captures show; the USB descriptors are in compressed init data.
15. Reproduction
No firmware, decrypted or encrypted, is stored in this repository. The inputs are the official .mencrypt packages from ChargerLAB's Google Drive and the four packages bundled with the Windows app 3.0.0, identified below by SHA-256.
- Decrypt. A short Python decryption script (pycryptodome) parses the QDataStream header, AES-128-ECB-decrypts the QByteArray with key 0 and truncates to the declared size. It checks the CRC-32 with
0x428..0x42Fset to0xFF, the vector table and the app info block at0x4420. It then prints the table in §4. - Analyse. Ghidra 12.2 headless with
-processor ARM:LE:32:Cortex -loader BinaryLoader -loader-baseAddr 0x4000. A pre-script adds the HC32F460 memory map (SRAMH, SRAM, retention SRAM, OTP, peripherals, QSPI XIP and registers) and seeds functions from the 160-entry vector table. A post-script decompiles every function. The never-returning RTOS tasks have no prologue, so I created them by hand at the addresses in §13. ECB was fingerprinted by counting repeated and shared 16-byte ciphertext blocks. - Per-version ADC check. In each decompiled image, find the GetData builder: the function that copies
0x2Cbytes with extended header0x8001/0xB00. Record the level-0 mask and the tail overrides, find every writer of ADC struct+0x4C, and read the per-rate SPS table that the rate setter indexes.
| Package | Package SHA-256 | Plaintext SHA-256 |
|---|---|---|
| 1.7.9 | f8974b15601e2a6c58e9daddf9136ee9f903db723f2bca190d4249f659a65fc2 | fab69c2f14a137d578996a6ddb36243dd9dd855ffb74a575d452b5ff58d22d3c |
| 1.8.0 | a9ddb5f20307562182acde6797e819cf5fbdc578e0851a19712b14e445b6661d | 8ab421d5f3cd4657633ebb088fddf1c14060dd5ce6086f5f1a9ddc8d25434687 |
| 1.8.4 | 48b9a98c9cec03331e964ad707737ad8f5d69adbcc8c4103fd93782265e21ef3 | 78cf2e21ad4f15bb8957d9f43094d45f9cae8c95612d3aa8533c3ccfa1b5f142 |
| 1.8.5 | b36be632b21baddb63379ec24d76be7d7cbfeabd20ca046f1a0686de7a0740d7 | a897ece9eb00a8090944be2b905fe8f1460d83dc60d440630b6d61e87ae5c9f8 |
| 1.8.6 | 6043bbab97dccb93738a77b80cace13d4a7253cde320aceb241c6b2979a6a275 | 9f2fa172b18ba5bf55efe9dce02d9d24ca969288eb39912416e50e5fbe1535ae |
| 1.8.7 | d1cf2ecb63469338d5b577ac9a3fc9d38b00dcdf883f44b3454f8637cb21f093 | 224af714526d0a3a4bd44d8e52fb45af79e7b494391362ee1fc6f21876e8c40a |
| 1.8.8 | 20de23e9032bf2b33a611b0bc23122f39f7236662a487ab15c9cc4b962a0bbc3 | 89b07ebaf1b0fefed9e7b8057edf4bdfc8b484026c5207bb499bac20ed7baf80 |
| 1.8.9 | 9a0da5a9565ac7de1b24948bd52cae39cba50f113af21a348e4a70962c961665 | 64ad5c4728b7ed56d1e8b16fe10528ad4ee9ef684f77e475f0a61c540fb0bcc1 |
| 1.9.0 | ec87e015e1198ebc531570a224a6ad902c61f0bd2b66bad7cd9dfc12dc08c361 | e9f3b166fc029ecbd80afa064f740970264107b5078d14c2927ad55e62e39315 |
| 1.9.2 | b0fd64123147638742b5cbdfb0304c41521ff74371b4eb214223fc1191961f25 | fdf6f0470bb97b2b1ed886869896e250fbff43699c28946f9f9a450073ead6f5 |
| 1.9.3 | ea6d1781beb748740e3904c0b3b5735803b29e5c0c39e26a10c1922f20559c39 | 4a4810d3f5e6303396fe6a14a644139d41f019215f30d263cc07b6f474b147d7 |
| 1.9.5 | a8cb82cd1980fbdab8df54b2d1dfc8a6bbc4a9092815a5d2d2210b3407262946 | 22281fde35c4622e1f5e5bd7a1fd496de8f72517c43befb7cf701f02d343443d |
| 1.9.7 | 71c814571f0d3384ea294bb3f6e83d0b863b6b152e5ec553e8f581728069c9db | 933e7ca6f50db208dc2a8a9951a125655607c7dc2bd24334082601c47b5a65b7 |
| 1.9.9 | cdbdd032c49f074510dcb8f746daf227e77ca4c40e63df7161f18a879f4aecdc | 0787c2284d95db50adccdd887dbe137fe0d8e68b056d099d5f848729a6c9358b |
| Windows-app 2.0.0 | 57cd634713090fb2993c44e0edb09d53405e18d24d1234ff5e3fd635cfc326fc | 2b00081aa5f5c2ca3107de26420a4a62375c7bb5fefe19cafc3628cf03f6f2d1 |
| Drive "2.0.0" (2.0.1, 2026.4.9) | e1e67c5c60ec3e0449561ac36005c03215e22f30d1ddeeb3351f814217b65efe | 91a6f2ff3ac88a9d5dd6ec388452a5165aa954b9669f36705bad26e9be1b1b5a |
| 2.0.1 | b0953bf12accfe1b3aff2d1034b21180218fc17a69ac1f84c63aee001c192907 | af41f168720bc8675c7193d7ad9e4a2acb99b05cf0de6decacd611c4b892910f |
| 2.0.2 | 7f229082db7ec763ef083af06684ff73be986862e21063593dddc939bb73e634 | fc7758010a8bf57a73b9f21f58fac62eb17314cf84203ecf026d07d2fd792e7e |
| 2.0.3 | 9f73eba25b74aa6c6d5ac1baf71117900a85c5a2faa543b8b67a942f5b162cd0 | 866683d04a5761c6d2c184ba9abfa34ef044cd78023ed41f58ccd7658930abdb |
| 2.0.5 | b3f6705b14bb959576905b2f4c72a79556c8e1944b2232b60eb6c5900f38bd7b | 627b60ea653771e431c532c161ab3bb95a3f0f549daeeba98fe60635ad3cb141 |
| 2.0.6 | 33e7034cb15c221e6d772389b42b01f37eab55334d04282e4763bb34868a2444 | 2e2a959aad8cfbda85b88d9f4f1b036bf28bc9d4e0daaa4d3864e51656590aec |
| 2.0.7 | 7a081af887fa8873d9e0cfc9cee9654187ab392f882efed16c1904a184ce8454 | 1440925005e848ff4d7b17fabb4a015a972b2b4dd519ed7c7a7352f8b236d1bf |