Open questions
The project · living document
Status: Solved — as a status review: every open research question is tracked
here with its evidence and what is left, cross-checked against the
research-labelled issues and against each document’s own status line. Last
reconciled with the documents at the docs/ readability sweep (#200).
This is the single place that says, for every open research question, whether it is answered, by what evidence, and what is left. Where this file and a document disagree, the document is the authority — it holds the measurement — and this file is the one that is wrong.
Phase 0 reads 22 of its 23 sub-issues closed, counted from the issue tree (§F); #39 is the last one open.
Status words (same vocabulary as docs/README.md):
| Status | Means |
|---|---|
| Solved | Answered, and the answer is checkable from a doc |
| Partially solved | The main shape is known; a named sub-question is not |
| Open | Not answered |
| Superseded | Stopped mattering — a better route was found, or a decision was taken |
A gap is only Solved if the evidence is written down. “We know this” is not a status.
A gap and the document that carries it answer different questions, so their words can differ without disagreeing: a gap is Solved when the question is answered even where the doc is Partially solved because a second, smaller sub-question is still open (gap 8’s RT-60A is the worked example — the register map is answered, the doc is not). Where the two differ, the What’s left column names the missing thing, and that is what reconciles them.
Summary
| Count | |
|---|---|
| Solved | 9 of the original 15 |
| Partially solved | 3 |
| Open | 1 |
| Superseded | 2 |
What actually blocks the next phase — everything else can wait:
- #39 — the identity of the three target machines (user action; blocks gap 12).
Drive C’s command set— closed 2026-10-11 (#6). The transport and the device contract are measured from the disassembly and a trace of the real software, and a model of the box makes the 1983 loader complete and PIP write toC:. The DPB came out ofDCL.COM(#145), andDFD.SPRturned out to be a replacement CCP rather than the transport, so the disassembly it was waiting on is off every critical path. What is left is #84’s, not research’s: how the box resolves banks 0–2 (hardware-gated) and whether Drive C, the printer at device 3 and the RT-60A can share one bus. The build spec isdocs/drive-c-design.md.- The hard-disk personality (#179) — the controller decision is closed:
emulate the Adaptec ACB-4000 inside the Media Distributing MD-10, not the
ACT (#8), and #178 is closed with it — the drive table’s layout and the
DPB’s values are both read (
harddisk-decision.md§7.2–§7.3,md10-system.md§9). The personality is built and host-tested, and the MD-10 system it needs boots to a prompt with three partitions (md10-system.md§1, §7). What is left is who copies the static DPB and descriptor tables into the running system (harddisk-decision.md§10 item 1,md10-system.md§9.1) and pointing the MAME harness at this personality rather than at MAME’s generic SCSI model. The RT-60A ABI (#7) — decide the contract, then implement it in #18.Done 2026-10-11. The ABI is written as a contract inrt60a-analysis.md(pointer block, sentinel/invalidation rules) and #18 implements it. The hardware handshake is retired as a known unknown.Does the DD upgrade read foreign formats? (#95)Solved and closed 2026-10-11. The answer is in the OCC CBIOS source — a DPB + sector-skew table (occbio95.asm) whose capacities match the manual’s list exactly, selected by a monitor-ROM density probe — and Osborne SD is confirmed end-to-end in MAME. The four foreign formats are taken as written, not synthesised: they are only needed to read original foreign media on the O1, and foreign media is converted on the modern side instead.dd-upgrade-formats.md.OZROM 1E as tooling (#81)Solved and closed:ozrom-1e.mdindexes the manual, measures it on all three machines, and records the column switch’s mechanism from the ROM. Of its four residues one is live — #167, driving the column keys from the harness — and #168 (a complete Utilities disk) is closed: we already had one, and the “290 of 400 sectors erased” that justified the hunt was acpmfs.pysector-skew bug (#170).
Everything else is either done, or is research that can proceed in parallel. #79 is closed and no longer blocking: the decision taken is not to chase the Nuevo path, and to build the video personality from the ScreenPac design plus OZROM’s soft 52/80/104-column switch.
A. The original gaps
| # | Gap | Status | Evidence | Issues | What’s left |
|---|---|---|---|---|---|
| 1 | Exact memory & I/O map, free decode windows | Solved | o1-memory-io-map.md §2–§7 (schematic), confirmed from the ROM source in o1-rom-source.md, measured on the running machine by tools/check_o1_map.sh |
#40, #62, #64, #72 | Decode granularity is still schematic-level — the chip selects check A11 plus one of A8/A9/A10 (o1-mainboard-schematic.md §7, §7a). Of the ranges this row used to call free, only three page-aligned blocks are selected by nothing: 2000–20FF, 2400–24FF and 2800–28FF; 2800–28FF is the recommended virtual-peripheral base (o1-memory-io-map.md §7) |
| 2 | NMOS vs CMOS Z80 across the target machines | Superseded | Resolved by history: the CMOS Z84C00 shipped in 1985, O1 production ended in 1983 — every O1 is NMOS. Interposer standardised on 74AHCT/HCT | #39 | #39’s remainder: machine IDs and date codes, for the record |
| 3 | Z80 socket and char-gen socket pinout | Solved | Z80: datasheet, schematic sheet 3 of 9, tools/check_schematics.py. UA15: read at 1200 dpi off sheet 4 of 9 (not the video sheet) — plain Intel 2716 pinout, /CE+/OE grounded, pin 21 strapped +5 V; address = (scan<<7) \| char; outputs serialised by UA14 (74166). o1-mainboard-schematic.md §4a, o1-memory-io-map.md §6, asserted in tools/check_schematics.py |
#62, #24, #64 | — |
| 4 | CoPower-88 monitor ROM dump | Superseded | The SWP co-processor ROM is in the repo: research/roms/swp-p88.rom, 8088 code, from Maslin’s archive. The board and interposer are common across hosts; only the driver varies |
#3, #4, #5, #70 | The dump is held; the doorbell is read off the schematic (copower88-schematics.md, Protocol synthesis). What is left is gap 6 — the 8088-side command set, still a disassembly away |
| 5 | Power budget | Solved | Decision taken: the interposer gets its own 5 VDC feed near the Z80 socket, sized for RP2354B bursts (~300 mA+) | #12 | — |
| 6 | CoPower-88 protocol | Partially solved | copower88-protocol.md — the port pair, the handshake and the block transfer, read out of the two RAMDISK.COM builds and corroborated by the board’s own schematic; copower88-schematics.md — SWP drawing 611-0003: an 8088 in minimum mode and a switch-set port decode into a two-port mailbox with a doorbell; plus the 8088 ROM |
#3, #4, #5, #23 | The 8088-side command set — research/roms/swp-p88.rom is that monitor and has not been disassembled for it; what drives status bit 7, the read-side flag (the schematic accounts for bit 0 only); and the Osborne variant’s port pair (the drawing sets the base on the board in units of 2 — Kaypro default 0x7E/0x7F — while the one photographed board reads as a hardwired comparator). The two docs already agree, so there is no separate spec left to consolidate |
| 7 | Drive C IEEE-488 protocol | Solved | drive-c-protocol.md; DRIVE_C.IMD extracted and disassembled (research/drive_c/disasm/); the ROM’s IEEE-488 implementation gives the PIA programming, the GPIB command bytes and the “always ready” stub behaviour (o1-rom-source.md); and the byte transport is now read from DCL — PIA init, the 128-byte strobed transfer, the operation bytes 88h/81h, port B bit 6 as the device status input. The transport is in DCL, not DFD.SPR (which makes no PIA references at all); and the port’s own semantics are now documented from the vendor’s side — o1-ieee488-port.md (Companion #8): eight low-level ROM drivers, 15 devices on the bus (8 if parallel poll is used), the machine’s own address is 0, and the driver table is the eight 3-byte jumps at 013F–0156 |
#6, #22, #84 | — #6 closed 2026-10-11. The two residues are #84’s now: bank resolution for n = 0,1,2 (hardware-gated) and the GPIB device-address question. The manual settles one thing: real GPIB protocol is not supported on the Drive C connector |
| 8 | RT-60A register map | Solved (as a contract) | rt60a-analysis.md, reconstructed from the manual |
#7, #18 | Decision taken 2026-10-11 (#7): implement the ABI, not the hardware. The manual documents the memory contract (0x0040 pointer, the block, the 0xEFF8 vector) and does not document the IEEE handshake — so faithful-hardware emulation is blocked on a CLOCK.COM we cannot read, and buys nothing the ABI does not. Left: |
| 9 | OCC1 hard disk | Solved | occ1-harddisk-analysis.md (product analysis; the ACT’s ports are read out of its own BIOS image); harddisk-decision.md (the decision, the plan, and §7.2–§7.3 for the drive table and the DPB); md10-system.md §9 (the tables read out of HARD.COM); acb4000-firmware.md, acb4000-host-adapter.md, acb4000-personality.md |
#8, #178, #179, #191 | #8 is closed on the decision: emulate the Adaptec ACB-4000 — the controller inside the Media Distributing MD-10 — not the ACT. Its command set is documented (the OEM manual is archived, harddisk-decision.md §9.1) and the ACT’s is recorded as unobtainable. #178 is closed too, both halves read: the drive table’s record layout and byte→field mapping are in harddisk-decision.md §7.2, and the DPB’s values turned out to be static data inside HARD.COM — type bytes at 0xDD5E / file 0x998, the 24-byte descriptors at 0xDD6E / 0x9A8, the DPB at 0xDDB6 / 0x9F0 (§7.3, md10-system.md §9). No peek is needed for any of it, and the system it belongs to boots to a prompt with three partitions (md10-system.md §1, §7). Left: who copies the static tables into a running system — a peek at 0xDD5E/0xDD6E on the booted system settles it (harddisk-decision.md §10 item 1, md10-system.md §9.1) — the CoPower-88 drive-letter policy (harddisk-decision.md §8, #183), and the four other open items in harddisk-decision.md §10 |
| 10 | 6850 ACIA location for the WiFi modem | Solved | o1-memory-io-map.md §3: 2A00/2A01, UC4 — confirmed independently by the ROM source |
#21 | — |
| 11 | Floppy controller, port map, DD changes | Partially solved | Ports 2100–2103 and the full WD179x command set from the ROM source; MB8877 ≡ WD1793, read off the schematic (o1-memory-io-map.md §3, §8); drive/density selects are PIA-driven, not a dedicated latch; 1.44 is the DD ROM; DD block = 1K; the Nuevo and Osmosis DD halves are documented (dd-upgrade-formats.md) |
#20, #95, #73 | The DD upgrade’s own board — which machine carries one, its part number, and how DDEN is wired — is not in the archive; it is on the photo shot list (photo-survey.md, item 8) |
| 12 | Which three machines, which board revisions | Open | — | #39 | User action: machine IDs (one is 24187A) and Z80 markings |
| 13 | A known-good boot disk per machine | Solved | Measured 3×3 matrix in mame-emulation.md §4: all three -blank.imd system disks boot on all three MAME machines; diagnostics disks also archived |
#29, #70 | The Nuevo system disk (OS1NUEVO.IMD) does not boot — #79, closed as not needed — but the gap as stated is met |
| 14 | Consolidate into per-device protocol specs | Partially solved | Done: copower88-protocol.md + copower88-schematics.md (the CoPower-88 pair), drive-c-protocol.md + drive-c-design.md + drive-c-executive.md, rt60a-analysis.md, o1-rom-source.md, o1-rom-variants.md, mame-emulation.md, o1-ieee488-port.md, o1-memory-io-map.md + o1-mainboard-schematic.md, harddisk-decision.md + md10-system.md + the three acb4000-* docs (hard disk), dd-upgrade-formats.md (floppy formats) |
#75 | Missing: a printer/parallel spec — o1-ieee488-port.md §6.2 maps P3 to Centronics, but no device spec exists — and a keyboard spec, where o1-memory-io-map.md §3 has the matrix and not the protocol. #75 (the book) is the final consolidation |
| 15 | Memory-resident vs pure I/O-port for virtual devices | Solved | Recommendation stands: pure I/O-port, in the free windows from gap 1 (o1-memory-io-map.md §7) |
#12, #16 | Same decode-granularity caveat as gap 1 |
B. Questions that appeared since the original review
B1. Hard-disk products — two in the archive, and seven routes advertised in total
The original gap 9 assumed a single OCC1 hard disk. The archive drop (#71) turned up a second, unrelated one. Confirmed from their own strings:
| ACT hard disk | Media Distributing / Adaptec | |
|---|---|---|
| Source | research/harddisk/OCC1_HARDDISK.IMD |
research/adaptec/*.img |
| Vendor string | AUSTRALIAN COMPUTER & TELECOMMUNICATIONS COPYRIGHT (C) 1982 |
Copyright (c) September 1983, Media Distributing |
| Versions | Version 31.05.82, Version 9-02-81 |
HARD version 1.93, PORTCHNG version 1.92, PREP version 1.93 |
| Controller | its own; Enter disk controller address ( |
Adaptec ACB-4000; Enter starting port address in hex, must be on a 4 port boundary |
| Utilities | BOOTHD, LOADBIOS, HARDBIOS.HEX, MOVCPM10/5/F, DISKEDIT, DISKTEST, RESTORE, SAVEFILE |
MD10.COM, MD20.COM, HARD.COM, HARDBIOS.SPR, HARDBDOS.SPR, HARDCCP.SPR, HD00/HD04BIOS.SPR, PREP.COM, PORTCHNG.COM, DRIVETBL.DAT |
So the three routes represented in the archive are the ACT hard disk, the Media Distributing/Adaptec hard disk, and Drive C (IEEE-488). Both hard disks use a port-relocatable controller — which matters, because the O1’s I/O decode is sloppy, so “4 port boundary” is the same two-address-line decode the rest of the machine uses.
Four more routes appear only in the magazine archive, with no software or media of ours behind them — recorded because a survey of “what storage existed” is incomplete without them, and because none is buildable from what we hold:
- Corvus hard disks, reached through a link program —
CLINK2, built fromCLINK2K2.ASMand assembled by the owner withASM.COMandLOAD.COM. 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 for exactly that program, needed because DD leaves less memory for it (docs/dd-upgrade-formats.md). No Corvus software or media held. - Osmosis OSMOS 7, advertised as “Winchester disks … 5¼, 10½, 15¾ & 21 MByte
units - a 42 Megabyte max capacity”, $2475–$3650 — on the same page as the
OSMOS 5 80-column board, which itself “includes a Hard Disk interface for
future plug-in expansion” (
docs/o1-upgrade-boards.md). No media held. - Trantor TSL-5 / TSL-10 — The Portable Companion #5 (February/March 1983),
PDF p. 28: “unpack your disk, plug it into the IEEE port, run a simple 5
minute installation procedure”, 5 MB $2995 / 10 MB $3395, supplied with BIOS,
back-up and retrieve utilities and cable
(
research/technical-articles/TPC-Issue05-1983-pp28-trantor-tsl-hard-disk-ad.pdf). This is the linedocs/drive-c-protocol.mdrecords the Drive C vendor as reselling at a discount; the advertisement is what pins its interface to the IEEE bus. No media held. - Helix (Micro Computers of New Orleans) — The Portable Companion #5,
PDF p. 98, “Winchesters Now Available For The Osborne 1”, 5/10/15 MB at
$2500/$3000/$3500, a “Z-80 adaptor card [that] leaves external ports free”
and a ribbon cable that “disconnects for portability”
(
research/technical-articles/TPC-Issue05-1983-pp98-tra-80-100-column-video-and-winchester-ads.pdf). No media held.
One of the two archived products now has a name. The Mega-byte Inc.
advertisement in The Portable Companion #5 (PDF p. 25) sells the MD-10 —
10 MB formatted for $2295, “20 and 44 megabyte Winchesters also available” — and
its footnote reads “MD-10 is a registered trademark of Media Distributing.”
That is the same Media Distributing whose September-1983 disk sits in
research/adaptec/, and that disk ships MD10.COM/MD20.COM, matching the 10
and 20 MB models. So the archive’s “Media Distributing / Adaptec” entry is the
MD-10, with an Adaptec ACB-4000 controller inside
(docs/occ1-harddisk-analysis.md). The ad also fixes the architecture: the
MD-10 “does not tie up existing OSBORNE ports” and is “significantly faster than
other winchester subsystems which use the IEEE port” — i.e. it is not an
IEEE-488 device, unlike Trantor, Drive C and the RT-60A. Taken with the ACT
unit’s Z80-socket interposer and Helix’s “Z-80 adaptor card”, that is a pattern
worth recording for #84: several vendors reached the Osborne 1’s storage
through the CPU socket or a port of their own, not through IEEE.
So the honest count is seven advertised storage routes beyond the floppies —
ACT, Media Distributing/Adaptec (the MD-10), Drive C, Corvus, OSMOS 7,
Trantor TSL-5/TSL-10 and Helix — of which two are buildable from what this
project holds: Drive C (#84), and the MD-10, whose ACB-4000 personality is
built and host-tested (acb4000-personality.md) and whose vendor-produced system
boots on the emulated drive (md10-system.md).
B2. Character-generator ROM variants — closed, and the two fonts measured
#44 asked for the early-font and Nuevo char-gen variants. Both are captured and
catalogued (o1-rom-variants.md): 5297c109 (early, paired with BIOS ≤ 1.4) and
6c1eab0d (late, paired with 1.43/1.44 and used by ScreenPac and Nuevo alike).
They really are different typefaces, and that is measured, not assumed. Read
glyph by glyph out of the two images (tools/romfont.py decodes both): 93 of
128 glyphs differ — 0 of the 32 graphics characters, 90 of the 95 printable
ASCII, and 3 of the rest — so the late font carries the graphics set over
unchanged and changes almost everything else. Only space, the apostrophe and the
backtick are shared (o1-rom-variants.md, The two fonts really are different).
Which font is fitted is therefore a visible difference, not a subtle one.
B3. The Nuevo system disk does not boot under MAME — #79, closed
OS1NUEVO.IMD carries the Nuevo 80-column CBIOS 1.5, loads, prints its banner,
and stops. The pristine upstream copy behaves identically, so it is not a bad
conversion. The three plain DD system disks do boot the Nuevo machine, at 80
columns (mame-emulation.md §2, §4).
Decision: not pursued. Nothing in the Nuevo design has turned up a feature
worth the validation cost, so the video personality (#24) is built from the
ScreenPac design plus OZROM’s soft 52/80/104-column switch, and the Nuevo
board stays prior art (#73). The residual question — MAME fidelity gap, or a
physical-card dependency — is recorded in mame-emulation.md §4 and only
matters if the Nuevo path is ever reopened.
B4. OZROM 1E — a third-party ROM usable as tooling — #81
It boots under MAME (via BIOS substitution) and its manual documents a software
52/80/104-column switch, a chip-level memory tester, a redefinable keyboard, and
its own ROM jump table and memory map. It also removes the IEEE-488 routines
entirely — the vectors at 013F–0156 now deliberately crash the machine, and
“IEEE-488 AS IOBYTE DEVICE 3 NOW NULL DEVICE” — so OZROM and Drive C are
mutually exclusive without reimplementing the driver. It also shrinks the video
window to 3 KB (F000–FBFF) to make room for its keyboard features, which
constrains any 2.0 ROM (#83).
Its four closing residues are routed in ozrom-1e.md §9. One is live — #167,
driving the column switch from the harness, which the harness cannot post because
it types plain characters. #168 (a complete Utilities disk) is closed: the
disk we hold is whole, and the “290 of 400 sectors erased” was our own
cpmfs.py sector-skew bug (§10.2). The jump-table diff against rom144.asm is
done (§6). One residue is new and unowned: the Utilities disk’s own
CRCKLIST.CRC lists a two-byte CRC per file, and no standard CRC-16 variant
reproduces them, so per-file verification against the vendor’s manifest waits
on identifying the algorithm (ozrom-1e.md §10.5).
B5. Other open research items
| Item | Issue | State |
|---|---|---|
| Nuevo + Osmosis upgrade boards as prior art (two products each; neither matches OCC’s implementation) | #73 | Closed. Both halves of both products are written up in o1-upgrade-boards.md — the video side there, the DD side in dd-upgrade-formats.md — with the prior-art note in research/inspired-by/README.md. Its residue, #173 (the Nuevo BIOS 1.5 presence check), is closed too: the check is a read of a port the Osborne 1 does not decode. Two live residues are recorded in o1-upgrade-boards.md §Open and neither has an issue: one measurement on a real Nuevo board — IN A,(04H) with the card fitted — and Osmosis’s paid technical manual, which the archive does not hold |
| Advent Products’ Osbaud baud-rate generator (#21) | #21 | The four connection points are decoded from the review + schematic (6850 pins 4/12, 74LS32 and 74LS04 pin 7); the internal crystal/divider topology is not known — needs the instruction sheet or a photo. o1-modem-serial.md, Advent Products “Osbaud” |
| Archive provenance TODOs (2 items) | #74 | Closed |
cpmtools misreads Kaypro DSDD images |
#35 | misc. Upstream identified: the primary site is moria.de (Michael Haardt, 2.23, email patch — no GitHub, no tracker), and the practical PR venue is the unofficial lipro-cpm4l/cpmtools fork (3 open PRs, 2.21-based). Closes when filed, not when merged |
| G2 multi-format diskette read/write | #30 | Closed — folded into #95, which is itself closed: the G2 set is more rows in the same DPB+skew table, and no rows get written, because the O1 does not need to recognise foreign formats |
Case + power research (the next-integrated board and the O1NEXT enclosure) |
#52 | Phase 5 (#49). A design task, not Phase-0 research. The archive does hold the O1’s own power and video hardware: the Astec AC8151 PSU repair manual (test set-up, the 115/230 V input-select wire and the +5/+12/−12 rail table, research/osborne1/o1astecpsac8151.pdf) and the MD1000-190 monitor specification, both added by #245 — what is missing is a supply schematic and board-level photographs |
| Stock MAME corrupts a guest file written to a floppy | #281 | research + misc + MAME upstream. Deterministic, unreported, and worked around; it has damaged two of this project’s builds (tools/research/rom13-reassemble.sh, rom2asm.sh) and is not caused by anything in this repo — osborne1nx is not involved |
Contribute OZROM 1E to MAME (BIOS slot or an osborne1oz machine) |
#82 | misc. The machine definition may be worth more than the dump; a physical OZROM 1E is available to read, which would at least give honest provenance |
| A local Jekyll environment | #106 | misc. site/ has no Gemfile, so this is partly creating the manifest, not just resolving it |
Osborne1nx.ttf — a complete modern font |
#107 | misc. Extrapolate the O1’s 8x10 forms into a full modern face with accents |
| The complete Osborne 1 reference (“the book”) | #75 | Ungated — a rolling build; pieces tracked as #108–#115 |
C. The research-labelled issues, and what closes each
After the 2026-10-11 triage the label means reverse engineering and
documentation again. Five issues left it: #35 and #82 (misc — upstream
contributions), #97 (misc — site and font), #84 (a build, now
hardware/drive-c-repro/) and #52 (Phase 5). #83 was never
research-labelled and has been removed from this table. #75 is not
research-labelled either — it is documentation/deliverable, the book — and it
stays only because it is where these specs are finally consolidated.
| Issue | Title | State | What closes it |
|---|---|---|---|
| #1 | Phase 0 umbrella | Open | The sub-issue tree (§F) — 22 of 23 closed; #39 is the last one open |
| #6 | Drive C → GPIB protocol spec | Delivered as docs/drive-c-protocol.md and validated end to end. The residue moved to #84 — bank resolution for n = 0,1,2, and the GPIB device-address question |
|
| #7 | RT-60A → a time ABI, not the hardware | The ABI is the deliverable and it is written in rt60a-analysis.md; #18 implements it |
|
| #8 | Choose which hard-disk controller to emulate | Closed | The written recommendation, taken: the Adaptec ACB-4000, one personality rather than two (harddisk-decision.md §1–§2, §9.5). Its build is #179 |
| #178 | Read the MD-10 DPB and DRIVETBL.DAT |
Closed — both halves read | The drive-table half: the file’s structure (13-byte header, 126 11-byte name records, sentinel, 113 128-byte description records) and the byte→field mapping — the nine geometry bytes are the ACB-4000’s own MODE SELECT drive-parameter block — are in harddisk-decision.md §7.2, read from the shipped file and confirmed by running MD10.COM’s editor in MAME. The DPB half: the values are in the archive, because 0x055E/0x056E are offsets from the MD-10 BIOS base 0xD800, not low-RAM addresses — the type bytes are at 0xDD5E / file 0x998, the descriptors at 0xDD6E / 0x9A8 and the DPB at 0xDDB6 / 0x9F0 (§7.3, md10-system.md §9), and DPB+3 = 31 matches the BLM the driver’s code reads. The mechanism stays as §7.3 recorded it (SELDSK → DPH at 0x056E + type*24 + 8; the DPB pointer at DPH+10; 128-byte records, 4 to a 512-byte sector). Left as #179’s: which copy a boot uses |
| #73 | Nuevo + Osmosis prior art | Closed | Written up in o1-upgrade-boards.md (both products, both halves, and how each is detected) plus the prior-art note in research/inspired-by/README.md. Residue #173 closed with it; two unowned residues remain, named in §B5 |
| #173 | The Nuevo BIOS 1.5 presence check | Closed | Read out of the two CBIOS images: IN A,(04H) at the head of the cold boot, guarding the 80-column configuration, with the factory DD disk O1CPM.IMD as the 1.4 partner. The one step a board would add — that the read is zero with the card fitted — is stated as such in o1-upgrade-boards.md |
| #75 | The book | Ungated | The pieces in #108–#115 |
| #81 | OZROM as tooling | Closed | The manual is indexed, the 52/80/104-column switch’s mechanism is measured from the ROM (it drives the stock video PIA through the stock ROM’s helper — 2C00, cached in EF61), and the jump table now diffs clean against rom144.asm. Residue routed: #167 (harness keystrokes). #168 (a complete Utilities disk) closed as unnecessary — we had it; the “erased sectors” were a cpmfs.py skew bug (#170) |
| #95 | DD multi-format (now also the G2, absorbed from #30) | Closed | Taken as written: the manual’s list and the CBIOS DPB+skew table agree field for field, and Osborne SD is confirmed end to end in MAME. The four foreign skew tables are deliberately left unverified — the O1 does not need to recognise them, because foreign media is converted on the modern side |
| #281 | Stock MAME corrupts a guest file written to a floppy | Open | A characterisation, not a fix: the defect is deterministic, reproducible on the stock osborne1 machine, and unreported upstream. It closes when the report is filed and answered — or when the workaround stops being needed |
Closed, but still carrying the label — the work that got Phase 0 to where it is: #2, #3, #4, #5, #6, #7, #8, #9, #10, #30, #44, #64, #70, #71, #72, #73, #74, #79, #81, #95, #165, #168, #173, #178, #198, #245. Most are attached to #1 as sub-issues, so the parent’s progress bar counts them.
D. Triage decisions taken 2026-10-11
The 2026-10-11 triage produced decisions rather than close-outs — #3, #4, #5, #44, #64 and #72 were all closed on it — and the decisions are recorded here because they were reached by discussion and are not re-derivable from a source:
- #30 folded into #95. Same question one product apart: the DD half is already
answered in
dd-upgrade-formats.mdfrom the DD upgrade manual, and the G2 format table moves to that doc. - #35 and #82 stay open, as
misc. They are upstream contributions, not reverse engineering — and #35 is a genuine courtesy fix now thatcpmfs.pygives us a known-good oracle to diff against. - #7 re-scoped to an ABI. Implement the documented memory contract, not the undocumented IEEE handshake (see §A gap 8).
- #8 re-scoped to a decision. Which controller to emulate — one personality, not two.
- #97 keeps the machine’s own repertoire — the graphics-set Unicode mapping
and the recovered logo art. The modern font is split out as #107
(
Osborne1nx.ttf). - #84 is a build, filed beside the other standalones (#47, #56) in
hardware/drive-c-repro/, not under the interposer. - #52 belongs to Phase 5 (#49).
- The book is ungated. #75 no longer waits on every research issue; it is a rolling build (#108–#115), with Part I and the appendices buildable today.
E. What is left before hardware design starts
Ordered by what blocks what:
- #39 — machine identities, board revisions and ROM markings (user action; unblocks gap 12 — the issue lists the three things actually worth reading).
Finish Drive C’s command set fromDone 2026-10-11 (#6 closed) — the protocol, the DPB and the capacity ceiling are inDCL/DFD.SPR.drive-c-protocol.md, and the P3 pinout came out of the Technical Manual already in the archive (o1-ieee488-port.md§6). TheDFD.SPRdisassembly came off the list rather than onto it: it is a replacement CCP, which a reproduction inherits. The bank-resolution question is hardware-gated and now sits on #84.Decide the hard-disk controller (#8): characterise both and pick one to emulate.Closed — the Adaptec ACB-4000 inside the Media Distributing MD-10, not the ACT (harddisk-decision.md). Its two sub-issues: #178 is closed — the drive table’s layout is read (§7.2) and the DPB’s values turned out to be static data insideHARD.COM(§7.3,md10-system.md§9) — and #179 is open: the personality is built and host-tested, the MD-10 system boots to a prompt with three partitions, and what is left is who copies the static tables into the running system and pointing the MAME harness at the personality rather than at MAME’s generic SCSI model.Write the RT-60A ABI contract (#7), then implement it in #18.Written 2026-10-11 (rt60a-analysis.md); #18 implements it.#73 — the DD halves of the Nuevo and Osmosis boards.Closed: written up ino1-upgrade-boards.mdanddd-upgrade-formats.md, with the Nuevo presence check (#173) closed with it — it is a read of a port the Osborne 1 does not decode. Two residues remain and neither has an issue: one measurement on a real Nuevo board (IN A,(04H)with the card fitted) and Osmosis’s paid technical manual, which the archive does not hold (o1-upgrade-boards.md§Open).
The book (#75) is no longer gated on all of this: it is a rolling build whose pieces are tracked as #108–#115, and Part I plus the appendices are buildable now. #11/#12 (interposer design) have everything they need.
F. The issue tree (added 2026-10-11)
Phase 0 was an umbrella with no children, so “how close is Phase 0?” was unanswerable from the issue list. Every phase parent now carries a sub-issue tree, closed issues included, so each one renders a progress bar:
| Phase parent | Children | Closed |
|---|---|---|
| #1 — Phase 0: research & reverse engineering | 23 | 22 (#39 open) |
| #11 — Phase 1: interposer hardware (+ the standalone adapters #47, #56, #84) | 8 | 2 (#13, #15) |
| #16 — Phase 2: firmware personalities | 12 | 3 (#83, #86, #155) |
| #25 — Phase 3: Osborne-side software | 4 | 0 |
| #29 — Phase 4: on-hardware validation | 0 | — |
| #49 — Phase 5: the reproduction | 3 (#52, #125, #128) | 1 (#125) |
| #75 — the book | 8 (#108–#115) | 0 |
Two conventions came out of this, and both are worth keeping:
- Closed work stays attached. #2, #3, #4, #5, #6, #7, #8, #9, #10, #30, #40, #44, #64, #70, #71, #72, #73, #74, #79, #81, #95 and #168 are children of #1. The progress bar is the point: Phase 0 is 22 of 23 and should read that way.
miscis the label for work that is not a phase — upstream contributions (#35, #82), the site (#97), build tooling (#106) and the modern font (#107). Everything else belongs to a phase, and to a parent.miscis the label for work that is not a phase — upstream contributions (#35, #82), the site (#97), build tooling (#106) and the modern font (#107). Everything else belongs to a phase, and to a parent.
The rule that keeps it honest: when an issue is opened, closed or re-scoped, the tree and this table move in the same PR — the same way a doc row does.
Appendix — decisions already taken
Kept here because they were reached by discussion and are not re-derivable from a source:
- All Osborne 1s are NMOS. The CMOS Z84C00 shipped in 1985; O1 production ended in 1983. The interposer uses 74AHCT/HCT; the replica uses a modern CMOS Z84C00 deliberately.
- The interposer gets its own 5 VDC feed near the Z80 socket (~300 mA+).
- Bank switching cannot be latched.
ROM MODE*is re-derived on every instruction fetch; a bank-2 access is only meaningful while the CPU is actually in bank 2 (o1-memory-io-map.md§4). - The interposer must never drive I/O addresses
0x00–0x03— anyOUTthere toggles the bank latch. The ROM never does it, and there is no software bank switch to model. - The
52-/80-/104-floppy prefix is a label, not a bootability gate. The display width is the machine’s; the text layout is the CBIOS’s (mame-emulation.md§4).