From the workshop

MPC.VAULT
The floppy archive.

Beats made on an Akai MPC2000 in the late ’90s live on 3.5″ floppy disks — 1.44 MB each, and most of them dying. This is the recovery log: every disk imaged raw, cataloged, and decoded down to the pad, one rescue at a time.

6Disks archived
56Files recovered
29Unique sounds
4Rescues

The disks

What’s been pulled back.

Each entry is one physical disk: what survived, what it took to read it, and the kit that was on it. Field notes are verbatim from the bench.

01

MPC2000 disk — rescued files (floppy02)

2 files · 23 KB · 3 sequences

MPC2000VOL MPC2000 DISSET · MPC2000 program (floppy02)

Recovered from a failing MPC2000 floppy on 2026-08-01. Only these two files were readable; all 8 SOUND*.SND returned I/O errors and are NOT here.

MPC2000 PROGRAMALL_PROGRAM

BANK A

A13
A14
A15
sound001
A16
sound000
A09
sound001c
A10
A11
A12
sound007
A05
sound005
A06
sound007
A07
A08
sound001c-23.8 st
A01
sound005
A02
sound006
A03
sound004
A04
sound003

BANK B

B13
B14
B15
B16
B09
B10
B11
B12
B05
B06
B07
B08
sound001
B01
B02
B03
B04

SEQUENCESOn this disk

SequenceBPMBarsSig
Sequence0187.324/4
Sequence029244/4
Sequence038744/4
02

DDJ WAHT / ALEJANDRO

11 files · 1.3 MB · 8 sounds · 2 sequences

MPC2000SET · MPC2000 program (floppy02)

mixtape tag Recovered by GNU ddrescue on the rescue station from the drive. 99.68% rescued, 0 read error(s), 9 bad area(s). A re-read is planned for the remaining sectors.

MPC2000 PROGRAMHYPE

BANK A

A13
sound001c-2.0 st
A14
tound001c-16.0 st
A15
sound001
A16
A09
A10
A11
A12
A05
sound018
A06
A07
A08
A01
LOOP-4.1 st
A02
LOOPc
A03
sound002
A04
HI_HAT_83+1.7 st

SEQUENCESOn this disk

SequenceBPMBarsSig
Sequence027944/4
Sequence017924/4
8 sounds on this disk — names & lengths only
SoundLengthRateTune
HI_HAT_830.15s44100—
LOOP0.10s44100-9.0 st
LOOPc0.10s44100-9.0 st
NOOPc trimmed0.10s44100-9.0 st
sound001c trimmed1.18s44100—
sound0020.12s44100—
sound0180.45s44100—
tound001c0.66s44100—
03

MPC2000 Demo Disk (built in)

15 files · 370 KB · 12 sounds · 11 sequences

MPC2000XLVOL MPC2000BUILT-IN DEMO

Bundled with the app so playback works with no floppy drive attached. The pad map and the sequences are real MPC2000XL content (a C.R.E.A.M. flip: 11 sequences on an 11-pad kit), from the GPL3 pyMPC2000xl / mpc2000xl_kaitai test corpus. The .SND samples are NOT the originals — those were never published. They are synthesised percussion stand-ins, one per pad, so the demo actually makes a sound. TIP.SND is a genuine sample from the corpus.

MPC2000XL PROGRAMCREAM

BANK A

A13
A14
A15
A16
A09
CASHRULESL+0.1 st
A10
CASHRULESR+0.1 st
A11
CASHRULES+0.1 st
A12
A05
JACK4+1.1 st
A06
JACK5+0.9 st
A07
JACK6+0.9 st
A08
JACKfix+1.0 st
A01
JACKint+1.0 st
A02
JACK1+1.2 st
A03
JACK2+1.2 st
A04
JACK3+1.2 st

SEQUENCESOn this disk

SequenceBPMBarsSig
Sequence0112024/4
Sequence0112024/4
Sequence0112024/4
Sequence0112024/4
Sequence0112024/4
Sequence0112024/4
INTRO12014/4
CHORUS112084/4
CHORUS212044/4
FINISH12024/4
INTROlead12014/4
12 sounds on this disk — names & lengths only
SoundLengthRateTune
CASHRULES0.33s44100—
CASHRULESL0.28s44100—
CASHRULESR0.30s44100—
JACK10.42s44100—
JACK20.07s44100—
JACK30.11s44100—
JACK40.22s44100—
JACK50.26s44100—
JACK60.34s44100—
JACKfix0.05s44100—
JACKint0.06s44100—
TIP0.07s44100—
04

FAILED READ — the drive (2026-08-01, not blank)

no data recovered

NOT a blank disk. Raw dd on the drive returned "Invalid argument" then 2880 x "Device not configured" — every sector errored and conv=sync zero-filled the image. The whole 1.44MB "read" finished in 0.92s; a real floppy read takes 1-3 minutes. This image contains no data from the disk. The disk itself is unassessed.

05

RESCUED via ddrescue on the rescue station (2026-08-01)

17 files · 904 KB · 7 sounds

MPC2000

100% recovered by GNU ddrescue 1.30 on the rescue station after macOS raw dd returned only zeros. 486 read errors, 42min, all sectors ultimately recovered. Named MPC kit: 112_KICK, HI_HAT_8, SNARE_70, BASS, HORN_168, ENGAGER, BLAHZEY, DANGER, MICROPHO, RIGHT, GO_GO.

7 sounds on this disk — names & lengths only
SoundLengthRateTune
112_KICK trimmed0.15s44100+3.1 st
BASS0.80s44100-3.8 st
BLAHZEY1.36s44100-12.0 st
ENGAGER1.08s44100-12.0 st
HI_HAT_83 trimmed0.17s44100—
HORN_1680.38s44100—
SNARE_700.23s44100-1.3 st
06

RESCUED via app (station)

11 files · 1.3 MB · 8 sounds · 2 sequences

MPC2000

Recovered by GNU ddrescue on the rescue station from the drive. 99.65% rescued, 105 read error(s), 10 bad area(s). A re-read is planned for the remaining sectors.

MPC2000 PROGRAMHYPE

BANK A

A13
sound001c-2.0 st
A14
tound001c-16.0 st
A15
sound001
A16
A09
A10
A11
A12
A05
sound018
A06
A07
A08
A01
LOOP-4.1 st
A02
LOOPc
A03
sound002
A04
HI_HAT_83+1.7 st

SEQUENCESOn this disk

SequenceBPMBarsSig
Sequence027944/4
Sequence017924/4
8 sounds on this disk — names & lengths only
SoundLengthRateTune
HI_HAT_830.15s44100—
LOOP0.10s44100-9.0 st
LOOPc0.10s44100-9.0 st
NOOPc trimmed0.10s44100-9.0 st
sound001c trimmed1.18s44100—
sound0020.12s44100—
sound0180.45s44100—
tound001c0.66s44100—

The method

How a dead disk comes back.

Purpose-built tooling, because nothing off the shelf reads these.

Raw FAT12, no mount

Images are read sector-by-sector with a custom FAT12 decoder that survives a wrecked boot record, recovers deleted files, and tells a genuinely blank disk from an unreadable one.

Hardware rescue

Disks the Mac gives up on go to a dedicated rescue rig running GNU ddrescue — hundreds of read errors, dozens of passes, whole kits recovered from disks that grind for minutes per sector.

Sounds, not filenames

Every sample is identified by its audio content, not its name — so two unrelated KICK.SNDs stay two sounds, and the same break on two disks is filed once.

This page publishes metadata only — labels, pad maps, sequence lists, recovery notes. No audio is hosted here. The demo disk’s pad map and sequences are real MPC2000XL data from the open-source pyMPC2000xl / mpc2000xl_kaitai corpus (GPL-3.0); its samples were never published.