Draft. Draft — assembled from the working notes by `tools/build_book.py`. The source document is the authority; this page is a copy for reading, and it is regenerated from that source.

Status: Solved. The mechanism is proven from the 1.44 CBIOS source (occbio95.asm) and exercised end to end in MAME on an Osborne single-density disk in a double-density system — the recognition step itself; the four foreign formats are taken as written, not synthesised, because the archive holds no media for them and the O1 never has to recognise them (#95).

The DD upgrade is usually described as “twice the capacity”. Its own manual describes something else: a controller that recognises six disk formats, five of them other manufacturers’, with the computer doing the recognising. The six, at the manual’s own figures, are Osborne 1 single density (92 k), Osborne 1 double density (182 k), Xerox 820 single density (80 k), Cromemco single density (80 k), IBM Personal Computer, CP/M-86 format (156 k) and DEC VT-180 (171 k). Six names, five disk parameter blocks — Cromemco shares the Xerox geometry — and the mechanism is two cooperating parts: the monitor ROM senses the disk by trying both densities and reading an address mark, and the DD CBIOS matches the sensed geometry against a DPB + sector-skew table (§”Why the ROM is part of the answer”). That table’s five rows reproduce the manual’s capacities field for field, which is what makes this the answer rather than a guess.

The list below is the manual’s, and it is the list to follow: where a secondary source differs — oldcomputers.net gives the Xerox 820 entry as 82K — the manufacturer’s own page-4 statement of what the option does wins.

Thus, you’ll find that you didn’t just purchase double density, you purchased a multi-density disk option. Under the current release of double density, the following formats are all automatically recognized by the computer.

Format Capacity
Osborne 1 single density 92k per diskette
Osborne 1 double density 182k per diskette
Xerox 820 single density 80k per diskette
Cromemco single density 80k per diskette
IBM Personal Computer (CP/M-86 format) 156k per diskette
DEC VT-180 171k per diskette

— Osborne 1 Double Density User Manual, p. 4

That is the primary source for #95. Note the wording — “automatically recognized by the computer”: the recognition is in the machine, which is what makes the claim testable in MAME without the hardware.

The manual also records what the option deliberately does not handle, which matters as much as what it does:

  • Hard-sectored diskettes (North Star, Micropolis, Vector) — the controller cannot see the sector holes.
  • Non-block formats (Apple, NEC) — no block-of-bytes structure to read.

and names the formats that were anticipated rather than shipped: TRS-80 Model I and III CP/M, and Zenith/Heath double density, “most likely in our user magazine, The Portable Companion”.

Why the ROM is part of the answer

The upgrade is described as coming with the 1.4 ROM, and the recognition is a property of the computer rather than of the drive electronics. The 1.44 BIOS source and the overlay CBIOS modules are in the repo (research/source-code/BIOS-144-source/), and reading them settles #95 without hardware. The mechanism is two cooperating parts, and neither is the monitor ROM alone:

  1. The monitor ROM senses the disk. SENDEN (jump-table entry 0130, SENSE_DISK_DENSITY) tries both densities and issues a WD179x READ ADDRESS (RADR) on track 0. The address mark the controller finds is the disk’s geometry: it yields the density (FM/MFM), the sector length and the sectors per track. There is no format table in the monitor ROM — the geometry comes off the disk. SAVTYP is set to a code of density (bit 0) | sector length (bits 2–3), where the length field is 0=128, 1=256, 2=512, 3=1024 bytes.
  2. The DD CBIOS carries the DPB table. occbio95.asm defines a table of Disk Parameter Blocks, and occbio45.asm (GDEN) converts the sensed physical sector count into a count of 128-byte logical sectors (×1/×2/×4/×8 for 128/256/512/1024) and searches the table for the entry whose sectors-per-track and type code both match. On a match it installs that DPB and its sector-skew translation table.

The table is the smoking gun (occbio95.asm / occbio15.asm, the names are the manufacturer’s own):

Entry Type code Sectors/track Block Dir boottrk DSM formula = capacity Manual’s figure
Osborne DD (DSKD1) 0Ch = MFM, 1024 B 40 (=8×5) 1 K 64 3 ((40−3)*8*5)/8 = 185 185 K 182 k
Osborne SD (DSKS1) 5 = FM, 256 B 20 (=2×10) 2 K 64 3 ((40−3)*2*10)/16 = 46 46 blk × 2 K = 92 K 92 k
Xerox (XEROX) 1 = FM, 128 B 18 (=1×18) 1 K 32 3 ((40−3)*1*18)/8 = 83 83 K 80 k
IBM (IBM) 8 = MFM, 512 B 32 (=4×8) 1 K 64 1 ((40−1)*4*8)/8 = 156 156 K 156 k
DEC (DEC) 8 = MFM, 512 B 36 (=4×9) 1 K 64 2 ((40−2)*4*9)/8 = 171 171 K 171 k

NUMDPB = 5. The capacities match the manual’s list exactly — IBM 156 k and DEC 171 k to the kilobyte — which is what makes this the answer rather than a guess. The manual lists six formats and the table has five entries because Cromemco shares the Xerox geometry (both 5.25” single-density, 80 k), so one DPB serves both names.

Each entry also carries a sector-skew translation table (XTAB in occbio95.asm): Osborne DD is 1:1, Osborne SD is 2-to-1, Xerox is 5-to-1, IBM is 1:1, DEC is 2-to-1. “Mostly DPB + FDC parameter tables”, as #30’s text put it — confirmed, and the skew is the part that would be missed by a geometry-only description.

So the DD upgrade’s multi-format claim is a CBIOS DPB+skew table selected by a monitor-ROM density/sector probe — recognition in the computer, exactly as the manual says, with the filesystem parameters living on the diskette’s own CBIOS rather than in the ROM. That also explains the G2 answer (#30): a “G2 format set” is more entries in the same table.

What this means for the project

Decision (2026-10-11): take the manual as written, and do not synthesise the four foreign formats. The claim has two independent sources that agree — the manufacturer’s own page-4 list, and the CBIOS DPB + skew table whose five rows reproduce the manual’s capacities field for field — and the mechanism is exercised end to end in MAME on the one format we hold media for (Osborne SD, in the test below). What a four-format test would add is the four skew entries and nothing else: a capacity cannot catch a wrong skew, only a DIR can. Those are skews for formats we are choosing not to serve.

  • The floppy personalities (#20, #62) must not assume Osborne geometry — but they do not have to implement six formats either. A Gotek-class emulator is format-agnostic: an .HFE carries the track layout, so replaying it is not the same thing as knowing the format. Format knowledge is only needed to synthesise a blank foreign diskette, and that is out of scope.
  • Foreign media is converted on the modern side, before it is mounted. The O1 never has to recognise anything: read the files out of a foreign image with tools/cpmfs.py (it already reads these geometries) and write them into an Osborne-format image with mkfs.cpm + cpmcp -T raw. That is why the four skew tables do not need verifying.
  • The G2 (#30) is more rows in the same table — and no rows get written. #30 asked for a G2 format catalogue; the fold into #95 carried the mechanism, not the catalogue, and with the decision above the catalogue is not needed either.

A second vendor made the same claim, and a bigger one: Nuevo

OCC is not the only multi-format claim of the period. Nuevo’s four-way product line, advertised in The Portable Companion #15 (Jan/Feb 1985, PDF p. 39), escalates from density to format:

Nuevo product Price Claim
Double Density Upgrade $150 “a full 183K of data storage per disk” — and “upgradable to double-sided double density”
2 x 2 Upgrade (780K) $595 double-sided double density, “nearly 400K each”, “Format compatible with the new ACC Vixen”, 128 directory entries
4 x 4 Upgrade (1580K) $745 “both drives are replaced with double-sided 96 track-per-inch drives”

The 4 x 4 text is the important one:

This system will automatically recognize single-sided disks that are either single or double density and double-sided disks that are either 48 TPI or 96 TPI. You can read or write, even boot, any of these discs automatically. Directory Capacity 256 File Names.

So Nuevo claims four formats beyond the six in OCC’s DD table — 96 TPI, both densities, both sidednesses — and claims to boot them. Two things make that credible rather than just advertising:

  • The stock O1 interface already carries what it needs. P8 pin 4 is TG43, the drive’s write-current-reduce input for 96 tpi media, and pin 32 is SIDE SELECT — both documented in docs/floppy-adapter-design.md from the schematic and the OEM cable. The controller is not the limit; the drives and the format table are.
  • The format table is the CBIOS’s job, and Nuevo replaces it. The installers on OS1NUEVO.IMD are DDT patches inserting BIOS15.OBJ at 1F00 and BIOS15BM.OBJ at 29E0 (docs/o1-upgrade-boards.md), and the advertised directory capacities — 128 and 256 entries — are disk parameter block values, exactly what such a patch changes. The 2 x 2 and 4 x 4 prices include new drives, so the media capability is hardware and the recognition is the BIOS’s.

Consequence for #95 and #30: the multi-format requirement is not an OCC quirk. Two vendors shipped it, with overlapping but different format lists, and the O1’s own floppy connector carries the signals for both. That argues for a floppy personality (#20) that is geometry-agnostic — replay what the image says — rather than one that hard-codes a single format. It does not argue for implementing every format: the images we mount are converted to native geometries first, as the decision above records.

Nuevo’s own BIOS 1.4 → 1.5 list: the formats are disk parameter blocks

Nuevo’s DD manual has a passage that settles the mechanism behind all of this. Its BIOS 1.5 differs from the factory BIOS 1.4 in ways it enumerates itself:

2) For BIOS Version 1.5 the disk parameter blocks for IBM, Xerox and DEC formats have been removed to provide space for code to allow auto configuration at boot time via NEWSETUP.COM. Space remains for two more DPB’s to be inserted.

4) BIOS Version 1.5 checks for the presence of the Nuevo 80 Column Upgrade. If not present, the configuration routine for the 80 Column Upgrade will be skipped. Thus, the Version 1.5 BIOS will boot and run correctly on machines that don’t have the Nuevo 80 Column Upgrade.

6) The jump table configuration and the jump table start address (E100h) is the same in BIOS Version 1.5 as in BIOS Version 1.4. — Nuevo Double Density Upgrade Manual, “differences between the two BIOS versions” (OCR: research/nuevo/Nuevo_Electronics_DD_Upgrade_Manual.ocr.txt)

Five things follow:

  • “BIOS 1.4” and “BIOS 1.5” here are the CBIOS on the system tracks, not the monitor ROM. Nuevo’s installers load BIOS15.OBJ at 1F00 and BIOS15BM.OBJ at 29E0 (docs/o1-upgrade-boards.md), and the manual compares 1.5 against “Version 1.4, the most recent factory version” — so the DPB table it describes lives in the bootable CBIOS image, and that is the file to read for #95, not the 4 KiB monitor ROM.
  • The monitor ROM contains no literal DPBs, and the factory BIOS does not need them. A structural scan of all four ROMs the project holds — factory 1.4, 1.43, 1.44 and Nuevo 1.51 — finds zero CP/M DPB-shaped structures. The reason is in the 1.44 source: the loader “will load single or double density and any number of sectors per track or bytes per sector”, and SENSE THE DENSITY probes — it reads an address mark at one density, flips the density bit, and retries, with a second pass after a home command (rom144.asm lines 3666–3712). Geometry is discovered by trying, not by table lookup.
  • Nuevo is not a superset of the factory BIOS. Nuevo removed IBM, Xerox and DEC to make room for its own code and re-added formats through auto-configuration at boot via NEWSETUP.COM. Two vendors, two mechanisms, overlapping lists — the same conclusion as above, now with the reason.
  • “Space remains for two more DPB’s to be inserted” — the table is extensible, which is how the 2 x 2 and 4 x 4 geometries fit into the same scheme.
  • Nuevo’s BIOS checks for its own video board’s presence and skips the 80-column configuration if it is absent — the same intent as the SCREEN-PAC’s 2400H flag, reached by a different mechanism (docs/o1-upgrade-boards.md §”How software finds the board”), and the reason its 1.5 ROM still boots on a machine with no Nuevo hardware. The jump table sits at E100h, unchanged from 1.4, which is the region the SCREEN-PAC’s width variable (E1A5H, docs/o1-memory-io-map.md §2a) lives in.

The 1.4 → 1.5 list, checked against two images (#173)

The list above is the vendor’s. It can now be checked, because the archive holds both sides: research/shipped-software/O1CPM.IMD is a factory DD boot disk whose system image is the same 8 KB CP/M build as Nuevo’s base, and research/nuevo/OS1NUEVO.IMD carries CBIOS 1.5 both as the installer’s DDT objects and as the system it boots. Five of the six items are confirmed in the bytes:

Manual item Where it shows
1 — the 59k sign-on removed Osborne Computer System / 59k CP/M vers 2.2 becomes Osborne 1 Computer / CP/M vers 2.2 / CBIOS vers 1.5
2 — IBM/Xerox/DEC DPBs removed at system offset 172C the factory carries three sector-translation tables and Nuevo keeps the first and zeros the other two, with a pointer table (E236, E24A, E93C, E95C) where the first table’s header was
3 — autostart names 7 → 8 characters the field moves: 1.4 keeps it at CB10–CB17 (8 spaces, unused), 1.5 at CB87–CB90 (06 "AUTOST" 00 + padding)
5 — the “Unformatted disk” message deleted 1.4 selects between two messages on bit 0 and patches the drive letter into both (E38C); 1.5 has one message and one patch site (E340)
6 — jump table at E100h, unchanged 19 JP entries at E100 in both images; every target moved
4 — the presence check not a register: IN A,(04H) at the head of the cold boot, guarding the 80-column configuration — docs/o1-upgrade-boards.md §”How software finds the board”

Item 3 is worth its own line, because it explains an installer step that otherwise looks arbitrary: SYS-80C.SUB’s FA10,A17,00 is not “clearing the width/variable block”, it is erasing the 1.4 autostart field — 1.5 keeps the name somewhere else, and 0010–0017 of the image is exactly the 8 bytes the fill clears.

A third vendor sold the format list itself: Osmosis OSMOS 4

Osmosis advertised the capability as a standalone product, in The Portable Companion #11 (October 1983), PDF p. 32 = printed p. 30:

OSMOS 4 — Disk Formatting & Conversion: Allowing your Osborne to read, write and format over 20 other manufactures CP/M formats. $225

That is the strongest form of the claim in the archive: not a capability bundled into a density upgrade (OCC), and not a side effect of a BIOS rewrite (Nuevo), but a $225 product sold on its own, priced above the $195 double-density board it was usually bought with. The same page bundles them — “OSMOS 1, 4 AND 5 ALL FOR $385”, $670 separately — which is a fair measure of how much of the value was seen to be in formats rather than in density.

Its density line, for the record: OSMOS 1 184k per drive on existing drives ($195), OSMOS 1.5 single-sided DD now, 370k later with new drives ($335), OSMOS 2 double-sided DD, 370k ($895), OSMOS 3 double-sided DD double tracking, 750k per drive, “an incredible 1.5MB” ($1050). The full product table is in docs/o1-upgrade-boards.md.

The tally is four independent format mechanisms: OCC’s DPB table, Nuevo’s boot-time auto-configuration, Osmosis’s separate conversion program, and the factory BIOS’s density/sector probing. A floppy personality (#20) that knows only the stock geometry fails on any of them.

A user note in The Portable Companion #11 (October 1983), PDF p. 54 = printed p. 52 — “Doubling hard disk memory” by Bob Gras, reprinted from the OSUG newsletter (Anchorage) — is a patch to Corvus’s link program, and it says plainly why it is needed:

If you had a single-density Osborne computer, but now have a double-density and are using the Corvus hard disk drives, this one is for you. … With the new double density system, more memory is required for managing the Osborne floppy drives, so this value has to be reset to accommodate the double density version of the Osborne.

The value is a CP/M offset in CLINK2K2.ASM (the source for the CLINK2 program that links the machine to the Corvus drive):

IF NOT (ABOVE)      ;
DELTA  EQU  500H    ; OFFSET FROM STANDARD CP/M *

and the fix is to “type a 900H in place of the 500H”, then assemble and link with ASM.COM and LOAD.COM (the article points at the Osborne manual’s assembly instructions, pp. 253–255 of the “Rev: 2/28/82” edition). So the constant grows by 400H — 1 KB — between the single- and double-density versions of the same machine, and the author’s stated reason is the DD system’s larger floppy- management code. The DD upgrade is not memory-neutral, which matters for anything that assumes a fixed TPA base, and it is the kind of third-party consequence the DD documentation itself does not mention (#95).

The acceptance test, run in MAME (#95)

The manual’s claim is a capability, and the capability is testable without hardware because the recognition is in the computer. The test: boot ROM 1.4+ with a DD CBIOS, put a disk in drive B, move files on and off it, and read the result back byte-for-byte.

Tooling

Two small stdlib helpers were added for this (tools/research/):

  • imdmake.py — writes a blank Osborne 1 disk in ImageDisk form (the writer half of imdinfo.py): --dd = 250 kbit MFM, 40×5×1024; --sd = 250 kbit FM, 40×10×256, filled with 0xE5. It emits the IMD 1.x: version line first, because disk-analyse refuses a file that does not start with one.
  • o1disk-copy.sh + o1disk-copy.lua — boot a disk in -flop1, put another in -flop2, wait for A>, type a CP/M command, and report the screen. It is the keyboard-injecting counterpart to mame-boot-probe.sh.

Results (measured)

Test Command Result
ROM 1.4+ boots DD mame osborne1 -bios ver14/ver143/ver144 on 52-blank.imd all three boot to A>
Content onto B: PIP B:=A:MBASIC.COM B: image gains MBASIC.COM, byte-identical to A:’s (24320 B)
Content off B: PIP A:BACK.COM=B:MBASIC.COM A: image gains BACK.COM, byte-identical — full round trip
Foreign density read DD system in A:, Osborne SD disk in B:, DIR B: lists B:’s files — the DD CBIOS auto-recognised the single-density disk

The last row is the important one for this issue: it exercises the real mechanism end to end. The monitor ROM’s SENDEN read the SD disk’s address marks (FM, 256-byte sectors, 20/track), the DD CBIOS matched DPBS1, and DIR read a foreign-density directory. Osborne SD is one of the manual’s six formats, confirmed on the emulated machine.

A MAME gotcha worth writing down

os.exit(0) in the Lua script silently discards guest disk writes. MAME flushes a modified image when the device is torn down at a clean shutdown; os.exit() skips that. A SAVE 1 HELLO.TXT under os.exit() left the image byte-identical; the same command with the script simply ending (MAME running on to -seconds_to_run) persisted. o1disk-copy.lua therefore does not call os.exit() — the opposite of mame-boot-probe.lua, which only reads.

What the test does not yet cover

  • The other four formats (Xerox 820, Cromemco, IBM CP/M-86, DEC VT-180) — we hold no images for them, so the DPB-table proof above stands on the source rather than on a run. The way to close it is to synthesise an image per DPB entry with imdmake.py (or disk-analyse) and DIR it; that is the next unit of work, and it is cheap now that the geometries are known.
  • Real hardware — see below.

IMD → HFE for the real machine

disk-analyse converts both ways, and the round trip is lossless on the raw sector data:

disk-analyse blank-dd.imd blank-dd.hfe      # write to a Gotek/HxC
disk-analyse blank-dd.hfe blank-dd-back.imd # read it back
tools/research/imdinfo.py --raw a.raw blank-dd.imd
tools/research/imdinfo.py --raw b.raw blank-dd-back.imd
cmp a.raw b.raw                             # identical

Measured: a --dd blank and the 52-blank.imd system disk both convert to HFE (2 039 296 bytes each), and IMD→HFE→IMD reproduces the raw sector image exactly. disk-analyse’s HFE output is what a hardware floppy emulator wants, so the images built and validated above are directly testable on a real Osborne 1.

Sweep of the archive (measured 2026-10-11). Every .imd we hold converts — 22 of 22, no failures — including the ones that are not Osborne-format: the double-density system disks, all fourteen Dunfield shipped-software images (SD and DD), DRIVE_C.IMD, OS1NUEVO.IMD, OCC1_HARDDISK.IMD and the two diagnostic disks. A spot round trip on the single-density 80-column OS1XUTLS.IMD reproduces the raw sector image byte-for-byte. So the “foreign” images in the collection are ready to write to a hardware emulator.

for f in research/shipped-software/*.IMD research/drive_c/DRIVE_C.IMD; do
    disk-analyse "$f" "$(basename "${f%.*}.hfe")"
done

Sources