OZROM 1E: a third-party monitor ROM
The machine
Status: Solved — as tooling, and #81 is closed on that deliverable. It boots under MAME on all three Osborne 1 machines (measured, §3), its 58-page manual is OCR’d and indexed here (§4–§8), and its three engineering features — a software 52/80/104-column switch, a chip-level memory tester and a documented ROM interface — are available to us without hardware. The column switch’s mechanism is measured from the ROM as well as documented (§7): it drives the stock video PIA through the stock ROM’s own helper, and OZROM adds no cell of its own. Its companion Utilities disk, which surfaced in 2026-10, is catalogued in §10 — the directory and all twenty file bodies are complete. One residue is routed rather than carried: #167 (drive the column keys from the harness), §9. #168 — filed for a complete Utilities disk — is closed: the “290 of 400 sectors erased” reading that justified it was our own sector-skew bug (§10.2).
Source: research/roms/OZROM-1E/OZROM_1E.BIN (4096 bytes) and
OZROM_1E_Manual.pdf, both from BrettHallen’s Osborne_1 repository;
OZROM_1E_Manual.ocr.txt is the 300 dpi tesseract text layer (grep it for the
appendices — but not for exact code, it is OCR). Micro Management, Inc.,
Manual Rev. A 84/09/01. Serves #81, and touches #24 (video), #29 (validation),
#45/#47 (keyboard), #20 (floppy) and #7/#18 (clock).
1. What it is
A replacement for the 2732 monitor ROM at UD11, sold in 1984 by Micro
Management (Irving, TX). It is a ROM swap with capability, not a font
change: the manual’s feature list is mostly things the stock ROM lacks. The two
engineering features that matter to this project are the documented memory map
and jump table (§5–§6) — a second, independent description of the machine’s
interface to hold against rom144.asm (#72) — and the software column switch
(§7), which is on the design path for the video personality (#24).
2. How to run it under MAME
MAME cannot be told about a BIOS it does not define, but it will load a file whose bytes do not match the expected hash (it warns and carries on). So present OZROM under a stock BIOS filename.
Build a romset, do not just drop files in a directory. The simple
mkdir .../osborne1 && cp ... recipe in mame-emulation.md §5 works for
osborne1 and osborne1sp but fails for osborne1nv, because the Nuevo
machine needs character_generator_6-29-84.14, which MAME satisfies by CRC
fallback from 7a3007-00.ud15 only when the files are loose in the rompath root
or inside a zip — not when they sit in a plain osborne1/ directory. Build the
zip instead:
mkdir -p /tmp/ozbuild/osborne1
cd /tmp/ozbuild/osborne1
cp <repo>/tools/emulator-setup/roms/*.ud11 \
<repo>/tools/emulator-setup/roms/*.ud15 \
<repo>/tools/emulator-setup/roms/*.ua15 .
cp <repo>/research/roms/OZROM-1E/OZROM_1E.BIN 3a10082-00rev-e.ud11
cp 7a3007-00.ud15 character_generator_6-29-84.14 # satisfies the clone's chargen
mkdir -p /tmp/ozset && zip -q -X /tmp/ozset/osborne1.zip *
mame osborne1nv -verifyroms -rompath /tmp/ozset # -> romset ... is good
The character_generator_6-29-84.14 copy is not a hack: it is a byte-identical
duplicate of 7a3007-00.ud15 (docs/o1-rom-variants.md), and it is what makes
the hash fallback deterministic across all three machines.
3. Measured: it boots on all three machines
O1_ROMPATH=/tmp/ozset tools/research/mame-boot-probe.sh MACHINE \
tools/emulator-setup/floppies/52-blank.imd 120
| Machine | Screen | Columns | Banner | Result |
|---|---|---|---|---|
osborne1 |
416×240 | 52 | OZROM 1E #00187 / (c) 1984 Micro Management |
BOOTED to A> |
osborne1sp |
832×240 | 104 | same | BOOTED to A> |
osborne1nv |
640×240 | 80 | same | BOOTED to A> |
So OZROM’s window is the machine’s (52/104/80), exactly like every other disk
(mame-emulation.md §2), and it boots CP/M off the utility disk to A> on all
three. MAME prints “the machine might not run correctly” because the ROM hash
is not what it expects — that warning is the point, not a problem.
4. What it adds, and what it gives up
From the manual (page 1 feature list; Appendix C for the losses):
| Gained | Why it matters here |
|---|---|
| Keyboard-selectable 52 / 80 / 104-column display | A documented software video-mode switch — the mechanism the video personality (#24) is being built on. On an un-expanded machine it can also toggle the left/right 52 columns of an 80-column display. |
| System memory tester that names the failing chip | A ready-made second oracle for #29, alongside tools/check_o1_map.sh. |
| Software-redefinable keyboard, plus eight locally-redefinable function keys (up to 63 chars) | Relevant to #45/#47. |
Time-of-day clock (TIME, ALARM, TIMER, QUIKTIME) |
#7/#18 — see §8 and docs/rt60a-analysis.md “The third shape”. |
| Documented interfaces: interrupt hook, ROM jump table, memory map, keytable | An externally documented O1 ROM interface to hold against rom144.asm (#72). |
| 1793/1797 FDC notes; truer Televideo 912/920 emulation | #20 and terminal behaviour generally. |
| Removes the “Centronics printer not connected” lockout on DD systems | #19. |
| Given up | Consequence |
|---|---|
| IEEE-488 functions (Appendix C) | OZROM and Drive C cannot coexist out of the box — Drive C is an IEEE-488 device. The manual says “extra software would be required”, and that software is the driver whose source we hold in rom144.asm (#6/#22). |
| The stock 4 KB video window | OZROM uses 3 KB (F000–FBFF) for video and the last 1 KB (FC00–FFFF) for the redefinable keyboard and function keys, so it scrolls by moving the 23 lines above rather than the window. This is the 1 KB price of the redefinable keyboard, and it constrains the 2.0 ROM design (#83). |
The IEEE-488 loss is not passive: the jump-table appendix says the entries at
013F–0156 were the IEEE-488 routines and calling one now produces a
controlled system crash (no disk damage — the machine’s two top display lines
go “gangbusters hog-wild”), and “IEEE-488 AS IOBYTE DEVICE 3 NOW NULL DEVICE”.
The parallel/Centronics entries at 0193+ survive unchanged.
5. The OZROM memory map (Appendix C-5)
“Some of the memory locations reserved by the OZROM 1E for its own purposes.”
The OCR splits the address column from the description column, but the pairing
is unambiguous; the clock rows are confirmed independently against the OCC
source in docs/rt60a-analysis.md.
| Address | Purpose |
|---|---|
EF07 |
Parallel-printer control — 0 = normal, non-zero = null device (DD only?) |
EF57–EF58 |
Current BIOS starting address — how to find the start of the O1’s special BIOS regardless of MOVCPM/OS patches |
EF5C–EF5D |
KTBPTR — keyboard pointer. OZROM sets it to FC00; if (KTBPTR)+2 changes from its initial 0, OZROM repoints it at the ROM table |
EF5F |
Keyboard matrix value of the last key pressed |
EF66 |
Start of system clock — hours, 2-digit BCD |
EF67 |
Minutes, 2-digit BCD |
EF68 |
Seconds, 2-digit BCD |
EF69 |
Sixtieths of a second, 2-digit BCD |
EFDF |
Time-display toggle byte |
EFE0 |
Keyclick toggle byte |
EFE1 |
Keyclick counter |
EFE2 |
Blink toggle byte |
EFE3 |
Blink counter |
EFE4 |
Blink rate value — gives 30/X blinks/sec; default 20 (decimal) = 1.5 blinks/sec |
EFE5 |
KEYTYP — whether the current key is normal / shift / control / control-shift |
EFE6–EFE7 |
Temporary pointer used by the keyboard-redefinition routines |
EFE8 |
Clock position — F000 + ((value + 43) MOD 256), anywhere on the first two display lines |
EFE9 |
Temporary storage of the character at the cursor (used by the blink routine) |
EFF2–EFF4 |
Interrupt routine hook — preset with RET (0C9h); see §8 |
F000 |
Beginning of video RAM (3 KB, F000–FBFF) |
FC00 |
Start of the RAM keycode table: FC00–FC37 normal, FC38–FC6F shift, FC70–FCA7 control |
FCA8–FCBF |
Reserved for future use |
FCC0–FD7F |
Used by ALARM/TIMER; 192 bytes available to a programmer’s interrupt routine |
FD80–FDBF |
Temporary storage for the bottom video line (64 chars) during MyKey II reprogramming |
FDC0–FDCF |
MyKey II function-key pointers |
FDE0 |
MyKey II cyclic counter (next function-key space) |
FDE1–FDFF |
MyKey II stack space |
FE00–FFFF |
MyKey II function-key definition space (8 definitions) |
6. The ROM jump table (Appendix C-6)
Provided “for the programmer’s convenience … since actual ROM addresses called
by this table may vary according to which version of the OZROM 1E is being
used”. This is the same interface the stock ROM exposes at 0100+, so it is
directly comparable to rom144.asm’s vector table (#72).
| Addr | Entry | Note |
|---|---|---|
0100 |
JMP COLD_BOOT |
|
0103 |
JMP WARM_BOOT |
|
0106 |
JMP INPUT_STATUS+01 |
keyboard |
0109 |
JMP CONSOLE_INPUT+01 |
keyboard |
010C |
JMP CONSOLE_OUT+01 |
display |
010F |
JMP LIST |
serial output |
0112 |
JMP PUNCH |
serial output |
0115 |
JMP READER |
serial input |
0118 |
JMP DRIVE_RESET |
|
011B |
RET ! NOP ! NOP |
no ROM-level select-disk |
011E |
JMP READ |
|
0121 |
JMP WRITE |
|
0124 |
JMP READ_TRACK_INFO |
|
0127 |
JMP READ_SECTOR |
|
012A |
JMP WRITE_SECTOR |
|
012D |
JMP LIST_STATUS |
serial-port status |
0130 |
JMP SENSE_DISK_DENSITY |
same entry as the stock ROM’s SENDEN |
0133 |
JMP ROM_JUMP_1 |
exit from the interrupt routine for bank 1 |
0136 |
RET ! NOP ! NOP |
no ROM_JUMP_2 |
0139 |
JMP FORMAT |
|
013C |
JMP INIT_SERIAL |
|
013F–0156 |
(were IEEE-488 routines) | calling these crashes the machine |
0157–0186 |
same functions as O1 ROM 1.4 | |
0187 |
OR 0FFH ! RET |
IEEE printer input status → null device |
018A |
OR 0FFH ! RET |
IEEE printer output status → null |
018D |
XOR A ! RET ! RET |
IEEE printer input → null |
0190 |
RET ! RET ! RET |
IEEE printer output → null |
0193 |
JMP PARALLEL_INPUT_STATUS |
Centronics path survives |
0196 |
JMP PARALLEL_OUTPUT_STATUS |
|
0199 |
JMP PARALLEL_INPUT |
|
019C |
JMP PARALLEL_OUTPUT |
|
019F |
JMP READ/WRITE_SEC |
direct FDC command execution; needs RWCMD (EFD1) and SAVTYP set |
7. The keytable (Appendix C-7) and the column switch
OZROM’s normal, shift and control keytables are moved up into RAM at power-up or
RESET (FC00+ per §5); the control-shift table stays in ROM and cannot be
redefined. The codes are the machine’s own, not ASCII: 0x80–0x89 are the ten
function keys, 0x8A–0x8D the four arrow keys, and the control-shift table
carries the “process immediately” codes 0xF0–0xFC (screen positioning; the
time, keyclick, blink and auto-horizontal-scroll toggles; the Screen-Pac
52/80/104-column select keys; and the assign/reassign-parallel-printer keys).
Three codes are permanently assigned: 0xDD = ctrl-shift-Z, 0xDE =
ctrl-shift-X, 0xDF = ctrl-shift-R.
The column switch is the part on our design path (#24). OZROM’s own words (Appendix C): the ctrl-shift keys include “the Screen-Pac 52-, 80-, and 104-column select Keys”, and on a machine with no 80-column upgrade the switch “can also toggle between the left and right 52 columns of an 80-column display”.
Measured against the three machines (§3)
OZROM boots at the machine’s native width — 52 / 104 / 80 columns on
osborne1 / osborne1sp / osborne1nv — and reaches A>. That is the same
result every other disk gives (mame-emulation.md §2): the window is the
machine’s, the text layout is the software’s.
What is not yet measured is the switch itself in action: pressing
ctrl-shift-
The mechanism, from the ROM
The switch is not a new mechanism and it uses no new RAM cell: it drives the video PIA through the stock ROM’s own helper.
rom144.asm defines H.VIO = 02C00H ; Video memory controls and
PIAAD = 0EF61H, and carries this routine:
LDK A,4+3 ; PIA control: select the data register
STO A,H.VIO+1 ; 2C01 = 7
MOV A,C
STO A,PIAAD ; 0EF61 - the stock ROM's "last PIA-A data" cell
STO A,H.VIO ; 2C00 - send data
RET
Disassembling OZROM_1E.BIN finds the same routine, unchanged, at 016F
(ld a,007h / ld (02c01h),a / ld a,c / ld (0ef61h),a / ld (2c00h),a / ret), and
OZROM’s console path reads the same two registers back at 04EC and 0BE1. So
the column keys compute an offset and hand it to the stock video-PIA write: the
register that holds the mode is 2C00, with EF61 (PIAAD) as the
cached copy — which is why OZROM’s own memory map (Appendix C-5, §5) has no
column-mode byte in it. There is not one.
Two things this does not settle, and they are the honest boundary of the finding:
- Which field of the
2C00byte is the horizontal offset. That port also carries the drive-select and density (DDEN) lines (o1-memory-io-map.md§3), androm144.asm’s own note onPIAADis about bit 0 = density. The manual gives the effect precisely — control-up-arrow repositions the window to start at column 29, control-down-arrow to column 76, control-[homes it — so the values are known even though the bit layout is not. - The dispatch address for the control-shift codes
0xF0–0xFC. A linear sweep cannot follow it:z80dasmwarns about self-modifying code, and the codes are dispatched through a table rather than by direct calls. Locating it is worth doing only if we want to trace OZROM’s handler; for #24 we need the register and the values, and those are above.
8. The clock, the memory tester, and the error codes
Time-of-day clock. OZROM’s TIME HHMMSS sets it, ctrl-shift-T shows it (top
two display lines, position set by EFE8), and ALARM/TIMER/QUIKTIME ride
the interrupt hook. Its cells are the machine’s own EF66–EF69 (hours, minutes,
seconds, sixtieths, BCD) — not a new ABI. Accuracy is “better than three
minutes a day” with the drives idle; disk access momentarily stops the clock
(the manual is explicit that the clock “has no guarantee” under disk load). The
full treatment, and why it matters to our own clock contract, is in
docs/rt60a-analysis.md “The third shape”.
The interrupt hook. On the 60 Hz video interrupt, OZROM’s SYSINT saves the
stack pointer, PC and registers and then calls 0EFF2h; a bare RET there
continues normally, so a programmer puts a jump to their routine at EFF2 and
ends with RET — no register save/restore needed. 26 bytes of interrupt
stack (13 levels) and 192 bytes of routine RAM (FCC0–FD7F) are available. Note
this is not the RT-60A’s 0xEFF8 hook: different INTBL slot, different
calling convention (see rt60a-analysis.md).
The memory tester. Press TAB at power-up or after RESET; it cycles
0 → 0EFFFh, writing and checking 0, 0FFh, 0AAh and 55h at each
location and naming the failing chip. Video RAM is not tested. A ready-made
second oracle for #29.
Error codes (Appendix D): BOOT ERROR = “you are not using an OZROM 1E”;
NO SYS! = a formatted diskette with no system on it (SYSGEN it); OH, NO! =
same as BOOT ERROR (or a bad system track). Also worth knowing: OZROM enables
CRC checking on disk access where previous ROMs did not, so copy verification
now catches real read errors.
9. What is left, and where it goes
#81 is closed on its stated deliverable. The manual is indexed, the column switch is documented and its mechanism measured from the ROM (§7), the memory tester and the error codes are written down (§8), and the ROM interface is recorded — memory map, jump table, keytable, interrupt hook (§5–§6). Four residues were left at the close, and the table records where each one went; only the first is still live, and it is routed rather than carried because it needs code rather than reading:
| Residue | Where it goes |
|---|---|
| Drive the column switch from the harness (§7) — the one measurement still missing | #167 — the harness can only post plain characters, so ctrl-arrow and ctrl-shift are inexpressible. That is a tools capability, wanted by #24 as much as by this issue. |
| A complete copy of the Utilities disk (§10) | #168 — closed: we already have it. The “290 of 400 sectors erased” reading was a cpmfs.py sector-skew bug (§10.2); with the correct skew all twenty files extract whole. |
Diff the jump table against rom144.asm (§6) |
Done — see below. |
Reconcile the EF66–EF69 clock against ROM 1.44 |
Not OZROM’s question: it is about which stock revision dropped UPTIM. Recorded in docs/rt60a-analysis.md; belongs with #83/#18, not here. |
The jump-table diff, done
Entry for entry, the manual’s Appendix C-6 table (§6) and rom144.asm’s vector
table at 0100+ describe the same interface. Every address matches, including
the ones where the names differ: 0118 is the stock ROM’s JMP RDRV / HOME
and OZROM’s DRIVE_RESET; 0112 is the stock ROM’s second JMP LIST for punch
output, which OZROM calls PUNCH. The only structural differences are the two
already recorded — 013F–0156, where OZROM removed the IEEE-488 routines so
that calling them crashes the machine, and 0136, where the stock ROM’s JMP
ROMJP2 becomes RET / NOP / NOP. Two independent descriptions of the same
interface agreeing is the point of the exercise.
10. The OZROM 1E Utilities disk — the software half (#81)
research/roms/OZROM-1E/OZROM_1E_Utilities.IMD is the disk that goes with the
ROM: the utilities the manual documents, as a customer received them. We had the
ROM and the manual and not the software, so this is the missing third of the
package.
It is a data disk, not a system disk. It does not boot — the system tracks carry a stub whose message is “There is no CP/M system on this disk.” You boot your own CP/M and put this in drive B, which is why the package shipped as a bare set of files.
10.1 The directory — complete
Readable in full: 20 files, all in user 0, in creation order, allocation intact. Standard single-density O1 geometry (40 cyl × 10 × 256 B FM, 2 KiB blocks, 3 boot tracks, 64 directory entries).
| File | Bytes | File | Bytes |
|---|---|---|---|
ALARM.COM |
640 | OZPATCH.COM |
3328 |
CLKPATCH.COM |
640 | QUIKTIME.COM |
256 |
CLKPATCH.DOC |
1792 | QWERTY.COM |
1024 |
CRASH.COM |
256 | READ.ME |
1280 |
CRCKLIST.CRC |
512 | SET.COM |
2048 |
DVORAK.COM |
1024 | SHOW.COM |
512 |
DVORAK.DOC |
1024 | TD.COM |
896 |
FK.COM |
2816 | TD.DOC |
1024 |
MEMORIZE.COM |
1280 | TIME.COM |
640 |
MYKEYII.COM |
640 | TIMER.COM |
768 |
The set maps onto the manual’s feature list one for one: the column switch
(SHOW), the keyboard remaps (DVORAK, QWERTY, MYKEYII), the clock
(TIME, TIMER, QUIKTIME, CLKPATCH, ALARM), the function keys (FK,
MEMORIZE), and OZPATCH to patch the ROM image itself. Two things it settles
that the manual only implies:
- The utilities are separate
.COMfiles, not ROM calls. Nothing here is resident; the ROM supplies the interface and the disk supplies the programs. A reproduction does not need to ship this disk to be OZROM-compatible. CRCKLIST.CRCis a self-check file — a name/CRC list, so a 1984 owner could verify their copy. It is the vendor’s own integrity manifest, and the hashes are what a replacement would be checked against (§10.2 has the news: we do not need one).
10.2 Condition — intact: the file bodies are whole
The disk is whole. Its ~291 sectors of uniform 0xE5 are the format fill a
CP/M formatter writes across a disk — free blocks, not holes — and the
arithmetic agrees to within a track: twenty files totalling ~22 KB, plus the
directory and the system tracks, is ~110 sectors of the 400, which is what is not
0xE5. Re-extracted with the correct geometry, all twenty files are whole:
no file holds an interior run of 0xE5 longer than a handful of bytes (those are
ordinary data bytes), and every size matches the directory. READ.ME is complete
from its first byte, which the emulated machine confirms by printing it from the
start (§10.2a).
Measured how. The file bodies come out from sectors that were never written
if the read uses the wrong sector order. The osb1sssd preset in
tools/cpmfs.py needs the format’s sector skew (skew 2 — the Osborne SD
layout interleaves its ten sectors per track); without it, block pointers resolve
to the wrong physical sectors: the directory still reads, because the three
directory sectors in use happen to land inside block 0 in some order, while every
file body comes back as 0xE5. cpmtools’ own osb1sssd diskdef carries the
skew, and reading the same image through it returns real content — CRCKLIST.CRC,
for one, is intact text. osb1sssd carries skew=2 (#170), and cpmfs.py
generates the logical→physical table the same way cpmtools does (cpmfs.c):
for this format, 0 2 4 6 8 1 3 5 7 9. The directory read goes through the same
mapping, so there is no off-by-one phantom entry. Checked against
cpmcp -T raw -f osb1sssd on all three in-archive SD images — this disk,
DRIVE_C.IMD and OCC1_HARDDISK.IMD — where every file matches byte for
byte. Nothing else in the archive rests on the bad read: the two other
osb1sssd extractions (research/drive_c/extracted/,
research/harddisk/extracted/) were made with cpmtools, which always had the
skew, and are unchanged, and non-interleaved formats default to skew 1, so
kp2x and drivec output is unchanged. This disk was the only casualty.
10.2a Measured on the machine: the ROM makes no difference
The OZROM ROM and this disk are independent of each other, and running them
together changes nothing about what the disk holds. The same image under the
stock ROM and under OZROM (-flop1 52-blank.imd -flop2
OZROM_1E_Utilities.IMD), with the same commands posted to each:
| stock ROM | OZROM 1E | |
|---|---|---|
DIR B: |
20 files | the same 20 files |
TYPE B:CRCKLIST.CRC |
the full vendor CRC list | the same list |
TYPE B:READ.ME |
the TIMER syntax note, from its start | the same |
So the diskette does not read differently with the ROM installed. The ROM supplies the interface and the disk supplies data; nothing in OZROM’s disk path changes what the bytes are. OZROM does enable CRC checking on disk access (§8), but that can only make it stricter, not more permissive.
One behavioural difference is worth knowing, and it is not about reading:
OZROM’s console acts on its control-shift codes 0xF0–0xFC — the “process
immediately” set from §7 — so TYPEing a binary file such as CRCKLIST.CRC
makes OZROM execute whatever codes its bytes happen to contain. Under OZROM the
screen came up with the keyboard-redefinition display after that TYPE; the
stock ROM simply prints the bytes.
10.3 The boot sector is not vendor code
Track 0 sector 1 holds a stub, and it does not look like anything Micro Management shipped:
C3 5C D2 JP 0D25Ch
...
+20 "PUT THIS ON T0S1"
...
+80 "THIS IS A TEST OF THE AMERICAN BROADCAST NETWORK"
“PUT THIS ON T0S1” is an instruction addressed to whoever was writing the disk, and the Emergency Broadcast System test line is a fill/verification pattern. The pair reads like a placeholder written by whoever produced this image, not product. Two consequences: treat the boot sector as an imaging artefact rather than evidence about the product, and expect that this disk cannot be booted to check anything — that has to be done by copying the files onto a system disk.
10.4 A tooling bug this found
imdinfo.py --raw appended sectors in the order they appear in the .imd, which
equals logical order only when the numbering map is sorted. Single-density
Osborne disks are not: they number (2,3,…,10,1), so every track came out
rotated by one sector.
The failure was silent, and that is the part worth remembering. A rotated CP/M
disk still parses — the directory lands somewhere plausible and cpmfs.py
returns a file list — it is just the wrong list. Here it returned 12 files
instead of 20, with 8 of them carrying another file’s blocks. Nothing errors,
nothing looks malformed, and the output is plausible enough to archive.
Fixed in tools/research/imdinfo.py; the dump is now byte-identical for sorted
maps (checked against DRIVE_C.IMD) and correct for rotated ones. Two archived
images turn out to be rotated and were never convertible before the fix:
research/diagnostics/DIAG_2.04.IMD now yields the 15-file Osborne diagnostic
disk (DIAG.COM, FILETEST.COM, DDBIT.COM, SETUP.COM, SYSGEN.COM, …),
and research/source-code/OCCROM.IMD its 77-track FM/MFM disk.
A second bug of the same class does more damage. cpmfs.py’s osb1sssd
preset had no sector skew for the format it names, so every SD Osborne file
body it read came back from the wrong sectors — the read that made this disk look
“partially erased”. §10.2 has the mechanism, the fix (#170) and the check
against cpmtools. The lesson is the same for both bugs, and it is the reason
neither was caught by a test: both produced output that parsed, and the result
looked like physical damage rather than a misread. A reader that returns
plausible garbage is more dangerous than one that errors.
10.5 What it changes for #81
Less than the file count suggests: the ROM is the deliverable and this is the optional companion disk. What it does give us:
- The complete and authoritative file list of the utilities — and, as §10.2 records, the complete files themselves, not a holed copy.
- Confirmation that the utilities are external programs, not ROM entry points, which bounds the reproduction scope: OZROM compatibility means the ROM’s jump table and memory map, not this disk.
- A second, independent description of the feature set to hold against the manual (§4).
The hunt for a complete copy is closed. #168 was filed to find one and
closed the same day — the “290 of 400 sectors erased” that justified it was our
own skew bug (§10.2), and the disk we hold is whole. The one check still not
possible is against the vendor’s own manifest: CRCKLIST.CRC lists a two-byte
CRC per file, and none of the standard CRC-16 variants tried so far reproduces
them, so per-file verification against the manifest has to wait for someone to
identify the algorithm.
Sources
research/roms/OZROM-1E/OZROM_1E.BIN,OZROM_1E_Manual.pdf,OZROM_1E_Manual.ocr.txt— BrettHallenOsborne_1,ROM/; OCR per theresearch/README.mdconvention (300 dpi,tesseract).research/roms/OZROM-1E/OZROM_1E_Utilities.IMD— the companion utilities disk (§10). Owner-supplied local file, 2026-10-11; source URL not yet recorded. ImageDisk 1.19, labelozrom, 40 cyl × 10 × 256 B FM.research/source-code/BIOS-144-source/rom144.asmand the OCC CBIOS modules (occram25.asm,occbio15/45/95.asm) — the stock interface and RAM map the jump table and clock cells are compared against.docs/mame-emulation.md§2/§5,docs/o1-rom-variants.md,docs/rt60a-analysis.md— machine widths, ROM inventory, the clock.