E-IV: 170 of 788 banks across 12 discs are named, declare a length, and read nothing

[bmxcode]

Context

docs/README.md's "What is not done" carries this, from when three E-IV discs were in hand:

E4P1 presets are not read. The three E-IV discs carry 916, 901 and 284 of them. On eiv-studio 100 of 230 banks have no sample directory and are listed with a note; preset-only banks are the likely explanation, and it is not established.

Twelve E-IV files are in hand now, and the whole series was swept on 2026-08-24. The figure is no longer one disc's 100 of 230 — it is 170 banks of 788, across 12 discs. Every one of them is named by the folder directory, declares a non-zero length, and comes back with no files and the note no sample directory found for this bank; listed only. The ADR-0012 invariant holds throughout; nothing is silent. But 170 named banks yield nothing, and 166 of them are not Credits text banks.

The measurement

banks and bound are from EmuBackend._banks and _eiv; unread is banks - bound and reproduces exactly the count obtained independently by parsing samplerdisc list -v output.

Disc	Banks	Read	Unread	E3S1 chains
Vol. 1 – Studio Essentials	230	130	100	250
Vol. 5 – 3-D Audio Collection	96	61	35	147
Vol. 2 – More Studio Essentials	168	142	26	229
Vol. 4 – Denny Jaeger Private	48	44	4	73
Miroslav Vitous – String Ensembles	44	42	2	80
Vol. 3 – Hollywood Sound Effects	92	91	1	161
Vol. 8 – Platinum Phatt (CD 1)	17	16	1	127
Vol. 8 – Platinum Phatt (CD 2)	20	19	1	61
Vol. 6 – Analogia Project (CD 1)	16	16	0	54
Vol. 6 – Analogia Project (CD 2)	12	12	0	35
Vol. 7 – Old World Instruments (CD 1)	17	17	0	46
Vol. 7 – Old World Instruments (CD 2)	28	28	0	75
Total	788	618	170	1338

The loss concentrates hard: Vol. 1, Vol. 5 and Vol. 2 account for 161 of the 170, and the other nine discs lose 9 between them. Vol. 5 and Vol. 2 are filed in active/e-mu-eiv-series-factory-library/ as the specimens for this; Vol. 1 is eiv-studio in active/archive-oldschoolscds/.

Two observations, and one that looked like evidence and is not

The unread banks are not obviously different from the read ones. Every unread bank declares a non-zero length in the folder directory (Vol. 5 min 1 max 17, Vol. 2 min 1 max 16, Vol. 1 min 1 max 52 — zero on none of them), and the 4-byte magic at their declared start block has the same distribution as the read banks' (BBBB dominant in both groups on all three discs). Whatever separates them is not visible at the bank's own start.

Named Credits banks are a real sub-case but a small one. 4 of the 170 are Credits — a text bank with no audio, on Vol. 3, Vol. 4, Vol. 5 and Vol. 8 CD 2. Those are correctly empty and should probably say so with a different note. The remaining 166 are named like ordinary instrument banks: Studio Snare 01–04, Studio Toms, Studio Cymbals, 3DPAKick, 3DPAHat on Vol. 5; Oxygen Mask, Freedom Strings, CP-70 El.Grand, Chroma 1/2, Magellan, StereoGrand 8mb, Harpsichord on Vol. 2.

A chain surplus is not evidence. E3S1 chains outnumber bound banks on every disc — 1338 chains against 618 bound banks overall — and it is tempting to read the difference as sample directories that exist and fail to bind. It is not: the four discs that read completely show the same surplus (Vol. 6 CD 1 binds 16 of 16 banks and has 54 chains; Vol. 7 CD 2 binds 28 of 28 and has 75). E3S1 has two uses (fs/emu3.py:186), so unbound chains are normal and the count says nothing about whether a given unread bank's directory is present. Recording this because it is the first thing an investigation will reach for.

Where to start

• Test the preset hypothesis directly rather than by elimination: if these are preset-only banks, their declared extent should hold E4P1 records and no E3S1 sample directory. Nothing in fs/emu3.py looks for E4P1 today, so this is a measurement, not a change.
• Vol. 5 is the sharpest specimen — 35 of 96, the highest proportion in the collection, on a disc small enough to read end to end, and its unread banks (Studio Snare 01 through 04, consecutive and near-identically named) are exactly the shape where a bank present in one form and absent in another should be visible.
• _eiv_bases / _eiv_unit fit base == BLOCK * (unit * start + bias) + EIV_RECORD_OFFSET over EIV_UNITS = (256, 512, 1024, 2048, 4096, 8192). If the binding rather than the data is at fault, a disc needing a unit outside that tuple, or a bias the fit does not reach, would present exactly as this does.

Not being worked on

Filing so the number does not have to be rediscovered. docs/README.md's E4P1 bullet should be updated when this is picked up — it still says "the three E-IV discs" and quotes one disc's slice of what is now a 12-disc, 170-bank measurement.