Skip to content

Unit

RolandSC-8850

roland-sc8850-01

Identity reply
F0 7E 10 06 02 41 42 00 00 06 00 00 00 00 F7
Specifications
ClaimedGM · GSMeasuredGS
  • 854Regions
  • 45,611Addresses
  • 149Heard
  • 19Banks
  • 461Blocks

Roland SC-8850

One physical unit, measured on one path, on the days these runs were taken. What follows is how much of it has been reached, what was asked of it, what stood between it and the archive, and how the measuring itself checked out.

Coverage

How much of the address space each question was actually put to. The whole is every address that answered a read at all; a bar that stops short is a run that stopped, not a unit that ends there.

  • Answered a read45,611
  • Held a value at power-on41,200
  • Asked what it accepts37,487
  • Reached by a named MIDI message78
  • Listened to671

149 of the 671 listened to changed the sound. 522 did not — which is a finding about the note it was asked with as much as about the address.

roland-sc8850-01/

Stages measured

Each is one command that produced records, and each puts a different question to the same unit. An underlined name has an entry in the glossary.

roland-sc8850-01/

Measurement chain

What stood between the instrument and the archive. A measurement is of a unit through a path, and the path is part of the finding.

midi_interface
Focusrite Scarlett 4i4 4th Gen
midi_cables
Roland RMIDI-B3 x2
midi_wiring
interface MIDI OUT -> unit MIDI IN 1; unit MIDI OUT 1 -> interface MIDI IN
audio_cable
Roland RCC-3-2R28
audio_wiring
unit OUTPUT 1 L -> interface INPUT 3; R -> INPUT 4 (rear, fixed gain)
unit_volume
maximum (hard stop, exactly reproducible without markings)
capture
CoreAudio HAL AudioDeviceCreateIOProcID, 48000 Hz, 6 ch, float32
capture_note
ffmpeg -f avfoundation silently loses about 11 percent of samples on this path and must not be used.
meta.json

Measurement health

What the chain was checked against while it ran. These are readings on the measuring, not on the unit.

sysex_reply_checksums
6/6 verified
repeat_read_identical
100/100
sysex_round_trip_median_ms
23.4
sysex_round_trip_range_ms
20.2 – 25.1
interface_loopback_sysex_bytes
203
interface_loopback_result
byte-exact
capture_timeline_ratio
1.0004
capture_timeline_stddev_ms
2.51
capture_dropouts
0
noise_floor_dbfs
-90.3
peak_dbfs_at_maximum
-25
peak_conditions
CC7=127, CC11=127, velocity=127
capture_open_latency_ms
650
capture_open_latency_note
Stream open to first block. A stimulus timed from the call to the recorder rather than from the first block sounds before anything is captured, and only its tail is recorded -- at the right length, with no dropouts, and healthy by every other check.
take_repeatability_db
-56.4
take_repeatability_note
Two takes of one struck note, aligned to a fraction of a sample and levelled, on program 0 with no effect on. 0.4 to 2.3 dB above the floor two takes' noise would leave.
take_level_agreement_db
0.001
trigger_scatter_ms
-3.6 – 0.3
chain_frequency_wander_ppm_over_20s
10
chain_frequency_wander_note
An upper bound, not a reading: every voice measured passed through the same converters, so the chain cannot wander more than the least wobbly of them does.
meta.json

First published2026-09-01

Address mapObservationsHow to read the archive

Measurements are CC0 1.0 · Not affiliated with, or endorsed by, any instrument manufacturer. Product names identify the equipment measured.