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 as a design. Every decision this design needed is made (§8), the Floppy Adapter’s Rev A schematic is captured, wired, footprinted and checked, and the board is ready for PCB floorplan and routing; nothing is built and nothing is measured on hardware.

Design spec for issue #62. Tracks the floppy interposer personality (#20); the disk-format position is settled in docs/dd-upgrade-formats.md (#95). Derived from research/osborne1/floppy-adapter/ (Wayne Visser passive adapter + provenance), Richard Loxley’s Osborne Restoration parts 5 and 12–15, and hardware/floppy-adapter/README.md.

The switch is solid state — three silicon parts — decided 2026-10-22. The case has no height for relay blocks, so DS-A is switched by an NC7SB3157P6X, the Gotek’s +5 V by a TPS22918 and its ground return by an AO3400A (§4.3, §4.4). The fail-safe survives: the switch’s own truth table parks the machine on the physical drive. Rev A KiCad capture lives in hardware/floppy-adapter/ (floppy-adapter.kicad_sch, generated by gen_floppy_adapter_sch.py).

Vocabulary: Rev A / Rev B in this document are the Floppy Adapter’s revisions. That is an unrelated board to the interposer’s Rev A / Rev B, so the letters always carry this board’s name — see hardware/README.md.

1. Purpose

Let the user choose, from a front-panel switch, whether the drive A position is served by the physical drive or by a Gotek (FlashFloppy) emulator, with drive B always present in both modes. The card sits between the O1 mainboard/DD-board floppy header and the OEM floppy cable; nothing in the machine or its cable is cut.

Prior art, deliberately not copied:

  • Wayne Visser’s Osborne1FloppyAdapter is a passive tap — it puts a Gotek on the bus alongside the physical drives and switches nothing. We keep its pin/rail mapping as the O1 interface reference.
  • Richard Loxley’s hand-wired build (parts 12–15) proved the concept but needed a DPDT switch + a 5 V relay soldered onto the Gotek, plus hand-cut Dupont cable. This design moves all of that onto a PCB at the connector end of the cable.

2. Osborne 1 floppy interface (34-pin, Shugart-like)

The O1 uses a Shugart-like 34-pin interface, but with +5 V and +12 V routed to the drives over otherwise-unused cable pins — there are no separate drive power connectors. The pin functions below are read from the O1 mainboard schematic itself — DWG 1A2011-00 Rev E, sheet 9 of 9 (DISC CONTROLLER), the P8 symbol; the same pinout is extracted in docs/o1-mainboard-schematic.md §5 — and they agree with the archived Wayne Visser adapter (research/osborne1/floppy-adapter/FloppyAdapter.sch), Loxley’s signal audit and the standard Shugart ordering for the shared signals.

Pin(s) Signal Notes
1,2,3,5,9,19,27,29,31,33 GND pin 7 is GND on the drive-end drawing instead; neither 2 nor 7 is a driven signal, so both are tied to GND here
4 TG43 not GND — FDC-driven; the drive’s write-current-reduce input on 96 tpi drives
6 EARLY not GND — FDC-driven
7 GND (drive-end drawing) see note on pin 1 row
8 INDEX  
10 DS-A (drive select A) the switched line — Gotek replaces drive A
11,13,15,17 +12 V power-over-cable (physical drives)
12 DS-B (drive select B) not switched — drive B stays present
14 (not drawn on P8)  
16 4 MHz O1-specific clock; not used by Gotek
18 DIRECTION  
20 STEP  
21,23,25 +5 V power-over-cable (physical drives; tapped for Gotek)
22 WRITE DATA  
24 WRITE GATE  
26 TRACK 0  
28 WRITE PROTECT  
30 READ DATA  
32 SIDE SELECT  
34 LATE O1-specific

Pins 4 and 6 carry live FDC outputs, so the card passes them straight through J1 → J2 rather than grounding them. This card switches the OEM cable through to the physical drives, and clamping TG43 low would assert reduced write current permanently; a Gotek-only adapter that grounds 2, 4, 6 and 7 does not have that problem, which is the whole difference between the two.

The Gotek needs only the static/step/data lines plus its drive-select (Loxley’s 11-line audit). The card routes that subset to the Gotek header and leaves 4 MHz and the spare power pins off it.

3. Topology

flowchart LR
  MB["O1 mainboard<br/>34-pin floppy header"]
  subgraph CARD["floppy-adapter daughter card"]
    PLUG["J1  2x17 socket (bottom)"]
    CABLE["J2  2x17 header (top)<br/>to OEM cable"]
    GOT["J3  2x17 header (top)<br/>to Gotek signals"]
    PWR["J4  Gotek power<br/>(switched +5V, GND)"]
    SW["J5  panel-switch lead"]
    U1["U1  2:1 switch<br/>(DS-A)"]
    U2["U2  load switch<br/>(Gotek +5V)"]
    Q1["Q1  N-MOSFET<br/>(Gotek GND)"]
  end
  DA["Physical drive A<br/>(150 ohm term pack)"]
  DB["Physical drive B<br/>(always present)"]
  GK["Gotek / FlashFloppy<br/>(as drive A)"]

  MB -- "J1 mates header" --> PLUG
  PLUG -- "pass-through (all pins)" --> CABLE
  CABLE --> DA
  CABLE --> DB
  PLUG -- "signals only" --> GOT
  GOT --> GK
  U2 -- "+5V (when Gotek)" --> PWR
  Q1 -- "GND (when Gotek)" --> PWR
  PWR --> GK
  SW -- "logic only" --> U1
  SW -- "logic only" --> U2
  SW -- "logic only" --> Q1
  PLUG -. "DS-A (pin 10) in" .-> U1
  U1 -. "DS-A to physical A (S low)" .-> CABLE
  U1 -. "DS-A to Gotek (S high)" .-> GOT

Everything between J1 and J2 is a straight pass-through except pin 10 (DS-A), which is intercepted. The physical drives therefore see an unmodified cable; the only difference is which device the machine’s DS-A reaches.

4. Switching design

4.1 Drive-select routing (pin 10)

DS-A from the mainboard (J1.10) is U1’s common. Control low connects it to J2.10 → the physical A drive; control high connects it to J3.10 → the Gotek. Never both — and U1’s break-before-make guarantees that even during the changeover — so the two devices cannot both answer as drive A. DS-B (pin 12) is untouched, so drive B stays present and selectable in both modes.

4.2 Gotek power switching

Loxley’s field notes (part 15) are decisive and are the reason this card exists:

  • Switching only +5 V does not turn the Gotek off — the O1’s pulled-up signal lines parasitically power the board through its input protection diodes.
  • Switching only GND makes the other drive unreliable — the Gotek’s signal lines are still referenced to the shared 5 V.
  • Switching both +5 V and GND works. That is what this card does.

Consequence: the Gotek’s GND and +5 V must arrive only through J4, not through the J3 header, otherwise switching GND has no effect. J3 carries signals only.

4.3 Switch element — DECIDED 2026-10-22: solid state, because there is no height for a relay

The decision, and the constraint that decides it. On 2026-10-22 the maintainer settled the switch element as solid state, with a constraint this card cannot argue with:

There is not enough height in the case for relay blocks.

That is decisive, and it is not about price or availability. A DIP-14 signal relay is ~10 mm tall; an industrial 4PDT on a socket is 35 mm. The card is a mezzanine between the logic board and the floppy harness, and the machine has no room for either. Rev A is solid state — three SMD packages, ~1.5 mm.

The distribution check agrees, independently: there is no 4PDT DIP signal relay in production to design around. Omron G6A-434P / G6A-474P were discontinued with a last-buy date of 2010-06-29, and the 4-pole DIP class went with them — Panasonic TQ4-5V and DS4E-S-DC5V are broker-only at $20–186 a piece.

  relay — the option rejected solid state — decided (Rev A)
height ~10 mm DIP-14; 35 mm socketed three SMD packages, ~1.5 mm
what the panel lead carries 56–73 mA of coil current microamps
DS-A a contact pair NC7SB3157P6X, ≤10 Ω, break-before-make
the two rails two more contact pairs TPS22918 (52 mΩ) + AO3400A (28 mΩ)
fail-safe de-energised → NC contact S low → B0 → the physical drive
parts 1 (unobtainable) or 2 3 + 4 passives
cost $4–12 for two DPDT ≈ $3
click, and contact life yes; 10⁵–10⁷ operations none; no wear-out

4.4 The three switches

Three parts, one control node (CTRL), and the same fail-safe the relay’s NC contact gave:

Ref Part What it does
U1 NC7SB3157P6X — single 2:1 analog/bus switch, SC70-6 picks which device owns DS-A. A is the machine’s DS-A; B0 is the physical drive, B1 is the Gotek
U2 TPS22918 — 5.5 V / 2 A load switch, 52 mΩ, SOT-23-6 switches the Gotek’s +5 V; VIN is the cable’s 5 V, VOUT is +5V_SW
Q1 AO3400A — N-MOSFET, 28 mΩ at Vgs = 4.5 V, SOT-23 switches the Gotek’s GND return (GND_SW)

The truth table is the fail-safe, and it is the part’s own. The NC7SB3157’s datasheet says S low → B0 connected to A; S high → B1 connected to A, so wiring B0 to the physical drive means a control that is low, floating, or has lost its lead parks the machine on the real drive. R1 (10 kΩ) pulls CTRL down for the same reason the relay design pulled the coil node down. The part also has break-before-make select circuitry, so the two devices are never both connected to DS-A even for an instant — the property the relay gave for free, and the one that must not be lost in a solid-state version.

Why these parts, and not the others.

  • A bus switch on DS-A, not a 74HC4053 or CD4053. The DS-A line is pulled up by 1 kΩ on the Gotek side; a 70–120 Ω switch in series with that puts the low level at 0.7–0.9 V against a 0.8 V TTL threshold — a part that works on paper and fails on a machine. The NC7SB3157 is a 5 V rail-to-rail part at ≤10 Ω, and its control input tolerates 5 V independent of VCC.
  • A MOSFET on the ground return, not a second PhotoMOS. An AQY212GS is 0.7 Ω, which at the Gotek’s ~0.3–0.5 A is 0.2–0.35 V of ground offset between the Gotek and the machine’s TTL signals — half the noise margin, on the read-data line. The AO3400A is 28 mΩ, which is 14 mV at half an amp. (Two AQY212GS would have been the simplest possible build — two parts, no 5 V-level care — and this is the number that rules it out.)
  • A load switch on +5 V, not a discrete P-FET pair. The TPS22918 is one part with a logic-level ON pin, 52 mΩ, and a configurable rise time; CT and QOD are left open (open CT is the fastest rise, open QOD disables the discharge). If the bench ever shows the Gotek lingering on power-down, QOD is the pin to come back to.
  • No flyback diode, and no coil. The card has no inductive load, so there is no D1.

Both rails are switched, and for the reason §4.2 gives: +5 V alone leaves the Gotek parasitically powered through the pulled-up signal lines, and GND alone upsets the other drive. The relay-versus-silicon decision does not touch that — it decides what switches the rails, not which rails are switched.

4.5 The panel switch, and the one lead that leaves the board

The panel end is a flip switch: UP = the physical drives, DOWN = the Gotek. It is a plain SPST in series between the cable’s +5 V and the control node, so:

Switch Lead CTRL DS-A goes to
UP (open) floating, R1 pulls down low the physical drive
DOWN (closed) tied to +5 V high the Gotek

Up being the fail-safe position is not a coincidence of wiring — it is the same property the relay’s de-energised contact had, and it is why R1 exists. The lead carries microamps (three CMOS inputs and a gate), where the relay’s coil wanted 73 mA, and the two wires are interchangeable: reversing them changes nothing, because an SPST in series with a gate input is symmetric.

R5 (1 kΩ) is in series in that lead, and it is fitted, not deferred. The lead leaves the board and runs to the front panel, so it is the one net here that can arrive carrying something the board never sent — a transient, or a short to something else in the machine. 1 kΩ limits what that can push into three CMOS gates and a FET gate; the FET’s ~630 pF gate then turns in about a microsecond, which is irrelevant for a switch a hand moves. R1 sits on the control side of R5, so the fail-safe is unchanged: an unplugged lead is still a low.

The card-end connector is DECIDED (2026-10-22): a bare 2-pin 2.54 mm header, with a female Dupont housing on a twisted pair of leads to the panel switch. The maintainer’s reasoning, recorded because it is the reason: “just a pin header is fine — an appropriate switch with a twisted pair dupont connector wire is trivial to implement.” A series SPST has no polarity to key, the two wires are interchangeable, and a lead that works loose fails to the safe position; twisting the pair is what keeps the lead quiet next to the floppy cables.

Considered and rejected, recorded so it is not re-litigated: a shrouded JST PH 2.00 mm (S2B-PH-SM4-TB), the part the WiFi COMM PAC uses for its keyboard-adapter power port, where the shroud and polarising slot are what remove the reverse-voltage case. It would buy retention and no exposed 5 V pin, and cost PH housings and a crimp — for a two-wire series switch that does not need either.

4.6 Termination and pull-ups (why drive A matters)

The O1 terminates the floppy bus with a 150 Ω resistor pack on every signal line, located in physical drive A — the drive at the far end of the cable. Drive B has no termination (Loxley parts 5 and 12). Two consequences:

  • The shared signal lines stay terminated in both modes, because only DS-A is switched: drive A remains connected to every other line, so its 150 Ω pack stays on the bus even when the Gotek is selected.
  • DS-A loses its pull-up in Gotek mode. The 150 Ω pull-up lives at drive A; routing DS-A to the Gotek disconnects it. The Gotek has 1 kΩ pull-ups on most lines but none on the drive-select lines (Loxley part 5), so a floating DS-A is exactly the failure Loxley chased for days.

→ Add an on-card pull-up on the Gotek-side DS-A net (U1’s B1 → J3.10): 1 kΩ default, with a 150 Ω option (jumper or second footprint) to match the O1’s own termination. This is the single most important addition to the design, and it exists only because the Gotek replaces drive A rather than drive B. Loxley tried both 1 kΩ and 150 Ω on the select lines.

5. Reference schematic (Rev A, netlist level)

Components:

Ref Part Notes
J1 2×17 socket, 2.54 mm bottom, mates O1 mainboard header
J2 2×17 header, 2.54 mm top, OEM cable → physical drives
J3 2×17 header, 2.54 mm top, Gotek (signals only)
J4 4-pin header (floppy power) switched +5 V / GND to Gotek
J5 2-pin header front-panel switch lead — bare 2.54 mm pins for a Dupont housing; JST PH alternative in §4.5
U1 NC7SB3157P6X, SC70-6 DS-A 2:1 switch (§4.4)
U2 TPS22918, SOT-23-6 Gotek +5 V switch; CT/QOD open
Q1 AO3400A, SOT-23 Gotek GND switch
R1 10 kΩ control-node pull-down (fail-safe)
R5 1 kΩ in series in the switch lead (§4.5)
R3 1 kΩ DS-A pull-up, Gotek side, default (§4.5)
R4 150 Ω (DNP) DS-A pull-up alternative — fit R3 or R4 (§4.5)
R2, D2 1 kΩ + LED (optional) front-panel “Gotek active” indicator

Connections (pin n of J1 unless stated):

  • Pass-through: J1.n → J2.n for every pin except 10. This carries the signals and the +5 V / +12 V / GND rails to the physical drives unchanged.
  • DS-A: J1.10 → U1 A (the common). U1 B0 → J2.10 (physical A, the S-low throw). U1 B1 → J3.10 (Gotek, the S-high throw).
  • DS-A pull-up: R3 (1 kΩ default; 150 Ω option) from the J3.10 net to +5 V, on the Gotek side of the switch (§4.5).
  • Gotek signals: J1 pins 8, 18, 20, 22, 24, 26, 28, 30 (and optionally 32, 34) → the matching pins of J3. J3 pins 12, 14, 16 and all GND/power pins are NC.
  • Gotek +5 V: J1 +5 V (21/23/25) → U2 VIN; U2 VOUT → J4 (+5 V).
  • Gotek GND: board GND → Q1 source; Q1 drain → J4 (GND).
  • Control: J1 +5 V → J5.1 → (panel switch) → J5.2 → R5 → CTRL, which drives U1 S, U2 ON and Q1’s gate. R1 pulls CTRL to GND. UP = physical drives, DOWN = Gotek (§4.5). Optional R2+D2 from CTRL to GND for the indicator.

6. BOM (Rev A)

Qty Item Notes
1 NC7SB3157P6X 2:1 analog/bus switch, SC70-6 DS-A; ≤10 Ω, break-before-make. ~$0.35–0.60
1 TPS22918 load switch, SOT-23-6 Gotek +5 V; 52 mΩ, 2 A. $1.60 @1, $0.51 @5
1 AO3400A N-MOSFET, SOT-23 Gotek GND return; 28 mΩ at Vgs = 4.5 V. ~$0.10
1 2×17 socket, 2.54 mm, vertical bottom mezzanine
2 2×17 header, 2.54 mm, vertical top (cable + Gotek)
1 4-pin header (floppy power) Gotek power, right-angle
1 2-pin header panel switch
1 10 kΩ TH control-node pull-down (R1)
1 1 kΩ TH series resistor in the switch lead (R5)
1 1 kΩ TH DS-A pull-up (R3, default)
1 150 Ω TH (DNP) DS-A pull-up alternative (R4)
1 1 kΩ + LED (DNP) optional indicator
1 PCB, 2-layer, 1.6 mm ENIG preferred (mating, hand-soldered SMD)

7. Physical design brief

Board (PCB).

  • 2-layer, 1.6 mm FR4; ground plane on the bottom layer; +5 V/GND traces ≥ 0.5 mm (Gotek ~0.3–0.5 A). The Gotek’s return path now runs through Q1, so the GND_SW trace and Q1’s pads carry that current too.
  • Envelope to be fixed against the mainboard keep-outs (research/osborne1/OCC1_1A2011-00_Schem_RevE.pdf) and host photos (research/pictures/); target a compact card that clears the DD board. Height is the binding constraint, not area — it is what decided the switch (§4.3), and the three SMD packages are ~1.5 mm against a relay’s 10 mm.
  • J1 (socket) on the bottom face, positioned to engage the mainboard 34-pin header; add 2 mounting holes/standoffs so the mezzanine cannot rock.
  • J2 and J3 on the top face; orient both so the OEM ribbon and the Gotek ribbon exit without fouling. Silk-label each connector, mark pin 1, and label “O1 IN”, “→ DRIVES”, “GOTEK”, “GOTEK 5V”, “SWITCH”.
  • Assembly is mixed: the connectors and the four passives are through-hole, the three switches are SOT-23/SOT-23-6/SOT-363. Order the boards with paste and a stencil or expect to hand-solder SC70 — it is the smallest part on the card.

Gotek mount (3D-printed part).

  • The Gotek (top cover removed) fits a floppy storage pocket; with its top headers fitted it is too tall, so connect it by a slim pigtail/direct-wire pads rather than plugging a header on top.
  • Design a printed bracket/faceplate that holds the Gotek in the pocket at the right depth, exposes the USB port, and carries the panel switch. This is the “Modification to support Gotek drives with 3D-printed floppy-disk faceplate” write-up planned in site/README.md. Cable egress from the floppy pocket — two accepted routes.

The card itself sits at the logic-board connector, inside the machine. What has to leave is the Gotek run (signals + switched power) and the 2-wire switch lead, and the pocket’s ventilation slots are the constraint. Both of these are accepted, and the maintainer’s guidance is to pick one:

  1. Break the 34-pin run into parts that fit the ventilation slots — Loxley’s approach. The full-size ribbon is split into narrower runs (or pulled down to the reduced 12-line signal set of §5 plus the two power wires) so each part passes a slot, and they are rejoined inside the pocket. Nothing on the machine is cut. At the card end that is still a 34-way IDC socket onto J3, with only the needed conductors fitted — the unused positions simply carry no wire.
  2. Cut the case to pass a full-size 34-pin cable, alongside the two switch wires. More work, irreversible, and it gives a clean unbroken ribbon with no joints to go intermittent.

The switch lead is two wires and passes either way; it is the ribbon that decides which route is chosen. Both routes need the same card, so this does not change the PCB — it changes what the builder does at the pocket, and it is why J3 is a full 2×17 header rather than a 12-pin one: the cable that arrives may be either.

Panel switch.

  • Plain panel-mount SPST flip switch: UP = physical drives, DOWN = Gotek (§4.5). 2-wire lead to J5 through R5; the lead carries microamps, never drive power. Mount next to the Gotek pocket so the up/down reading matches the machine’s own “which drive” convention.

Reversibility. No cable or machine modification: unplug the card and the machine is stock.

8. Decisions, and what each one rests on

All eight are closed, and two of them — item 3 (the switch) and item 8 (the switch connector) — were settled by the maintainer rather than by a session. Each carries the evidence it rests on; the two that remain bench-verifiable are flagged as such.

  1. Drive topology — CONFIRMED (user, 2026-10-10). The switch selects drive A (physical A ↔ Gotek); drive B stays present in both modes.

  2. Connector genders — CONFIRMED. The O1 logic board’s floppy connector is P8, a 34-pin MALE header. The Field Service Manual (2F00040) instructs “Attach the disk harness to the 34-pin connection on the logic board … Be careful not to bend any pins”, and the double-density adaptor’s “sockets … fit onto the logic board disk harness connector (P8)”. The harness is a female IDC socket and is not keyed — “the RED stripe on the harness must be to the RIGHT” fixes pin 1. Therefore J1 = female 2x17 socket (card bottom) and J2 = male 2x17 header (OEM cable). Pin order is confirmed in §2.

  3. Switch element — SETTLED BY THE MAINTAINER, 2026-10-22: solid state. The decision, and the constraint that decides it:

    There is not enough height in the case for relay blocks.

    ~10 mm for a DIP-14 signal relay, 35 mm for a socketed industrial one, on a card that is a mezzanine between the logic board and the floppy harness. So: three silicon parts (§4.4) — NC7SB3157P6X for DS-A, TPS22918 for the Gotek’s +5 V, AO3400A for its ground return — ~$3, no coil, no click, no height. The fail-safe survives: the switch’s own truth table is S low → B0, and B0 is the physical drive. The distribution check agrees: Omron G6A-434P / G6A-474P were discontinued in 2010, and the whole 4-pole DIP class with it (Panasonic TQ4-5V and DS4E-S-DC5V are broker-only at $20–190).

  4. Isolation depth — DECIDED: Loxley’s proven minimum. Switch DS-A + both power rails and nothing else. The shared signal lines are not additionally gated: doing so adds series elements to a 4 MHz-class bus for no proven benefit, and the failure Loxley actually chased (a Gotek that would not power down) is fixed by the dual-rail switch alone. Full signal gating is recorded as a Rev B experiment, not Rev A scope.

  5. DS-A pull-up — CONFIRMED 1 kΩ default, 150 Ω option. The FSM settles the termination location: “The A drive has an 8 pin 150 OHM Terminator resistor pack. B DRIVE DOES NOT.” So when DS-A is routed to the Gotek it loses the O1’s 150 Ω pull-up, and the Gotek has no pull-up on its drive-select lines (Loxley part 5). The card therefore fits R3 = 1 kΩ as the default and R4 = 150 Ω as an alternative footprint (fit exactly one) on the Gotek-side DS-A net.

  6. Gotek drive-select jumper — SETTLED (bench-verify). The Gotek must answer the O1’s drive-A select, which is cable pin 10. The Gotek manual labels the jumpers S0 = Drive Select 0 / S1 = Drive Select 1, and FlashFloppy’s generic guidance for a Shugart host is “unit 0 → place the selection jumper at S0”. The O1, however, straps its drive A as DS1 (FSM 8.1.4.11: “Drive A has a jumper at positions HM and DS1”), and the proven O1 adapter (WayneVisser/Osborne1FloppyAdapter) notes “Jumper only S1 on the GoTek”. Start with S1; if the Gotek is not seen, move to S0. This is a 30-second bench check and the card is agnostic to it.

  7. Gotek cable route / bracket — DECIDED. Gotek in the right-hand floppy storage pocket (Loxley part 14), driven from a reduced ~12-line pigtail that passes the pocket vent slots (a full 34-pin ribbon does not fit), with a 3D-printed faceplate/bracket carrying the Gotek and the panel switch. This is the “Modification to support Gotek drives with 3D-printed floppy-disk faceplate” write-up planned in site/README.md (§7). The two accepted egress routes — break the 34-pin run into parts that pass the slots, or cut the case for a full-size cable — are in §7, and the card is the same either way.

  8. Panel-switch connector — DECIDED BY THE MAINTAINER, 2026-10-22: a bare 2-pin 2.54 mm pin header. “just a pin header is fine — an appropriate switch with a twisted pair dupont connector wire is trivial to implement.” The lead is a series SPST (§4.5), so there is no polarity to key and the two wires are interchangeable; the shrouded JST PH alternative is recorded in §4.5 as considered-and-rejected rather than left open.

The evidence §8 rests on

  • O1 Field Service Manual 2F00040 (research/osborne1/): 150 Ω terminator is on drive A only; logic-board connector P8 is male; harness not keyed, red stripe = pin 1; drive A strapped DS1.
  • O1 mainboard schematic, DWG 1A2011-00 Rev E sheet 9 of 9 (DISC CONTROLLER) (research/osborne1/OCC1_1A2011-00_Schem_RevE.pdf, PDF page 12): the P8 34-pin header — GND = 1,2,3,5,9,19,27,29,31,33; +12 V = 11,13,15,17; +5 V = 21,23,25; pin 4 = TG43 and pin 6 = EARLY, neither is GND; pins 7 and 14 are not drawn at all; DS-A/DS-B on 10/12. Read at 600–1200 dpi; see docs/o1-mainboard-schematic.md §5 (#62).
  • O1 disk-electronics schematic, DWG 1A3004-00/01 Rev C sheets 1–3 (PDF pages 1–3 of the same file — a different drawing, the drive’s own board): head/drive connectors P1/P2; P2 lists GND = 1,3,5,7,9,19,27,29,31,33 and CASE = 32. It disagrees with P8 on pin 2 versus pin 7, which is harmless — neither is a driven signal on the O1, so the card ties both to GND.
  • Gotek SFR1M44-U100K user manual: jumper table S0 = Drive Select 0, S1 = Drive Select 1, MO = Motor On, JA = READY on 34-pin interface.
  • FlashFloppy wiki — Initial Setup / Gotek Models: Shugart hosts expect unit 0 → jumper S0.
  • Omron G6A-434P datasheet: 4PDT DIP-14 through-hole relay; 5 VDC coil = 72.5 mA / 69 Ω / ~360 mW; contacts 1 A 30 VDC (ST-US). The part is obsolete — G6A-434P and G6A-474P were discontinued with a last-buy date of 2010-06-29, and the 4-pole DIP class went with them (§4.3).
  • Maintainer, 2026-10-22 — the constraint that decides the switch: “There is not enough height in the case for relay blocks.” Recorded verbatim because it is the load-bearing fact: the card is a mezzanine, and a DIP-14 relay is ~10 mm and a socketed industrial 4PDT is 35 mm. It settles §4.3 without reference to price or availability.
  • NC7SB3157 / FSA3157 datasheet (Fairchild, Rev 1.0.4 — read as text, not as a PDF figure): function table “Logic Level Low → B0 connected to A; Logic Level High → B1 connected to A”; pin assignments in SC70-6 1 = B1, 2 = GND, 3 = B0, 4 = A, 5 = VCC, 6 = S; 1.65–5.5 V VCC, <10 Ω typical on-resistance, rail-to-rail signal handling, control input tolerant to 7 V, and break-before-make select circuitry. The truth table is what makes B0 the fail-safe throw, and the break-before-make is what keeps the two devices from ever sharing DS-A.
  • TPS22918 datasheet and product page (TI SLVSD76C): 1–5.5 V in, 2 A, 52 mΩ at 5 V, N-channel switch with a charge pump, adjustable rise time on CT, and a configurable quick-output-discharge QOD. The DBV-6 pinout used here (1 VIN, 2 GND, 3 ON, 4 CT, 5 QOD, 6 VOUT) is the one KiCad’s own library carries for TPS22917DBV, the same package and family.
  • Distribution check, 2026-10-22 (DigiKey / Newark / LCSC / TME via their product pages and the Findchips and Octopart aggregators): G6A-434P-ST-US and G6A-474P-ST-US obsolete; TQ4-5V broker-only $20–31 CAD; DS4E-S-DC5V broker-only $21–186. In production and stocked: AQY212GS (Newark $3.05 @1, $2.22 @100); TPS22918 ($1.60 @1, $0.51 @5, $0.15 @6k); SN74CBT3257 ($1.13 @1); and, for the relay routes that were considered and rejected, G6S-2 DC5 (Newark $6.00 @1, $4.65 @100), HFD2/005-S-L2 (Newark $2.20 @1, $1.66 @100), FTR-B4CA005Z (Electro Sonic $2.37 @50), MY4N-DC12-S (TME $7.78 @12) and Finder 55.34 (TME $13.59 @1).

  • Richard Loxley, Osborne Restoration parts 5 and 12–15: 150 Ω pack at the far drive; Gotek has 1 kΩ on most lines but not on drive select; switching only +5 V leaves the Gotek parasitically powered, only GND upsets the other drive — switch both.

9. Rev A schematic capture

Captured 2026-10-10 as hardware/floppy-adapter/floppy-adapter.kicad_sch, generated by gen_floppy_adapter_sch.py. Rewired, footprinted and re-verified 2026-10-22 — the same treatment the WiFi COMM PAC got (#347) — and then re-captured the same day for the solid-state switch (§4.3, §4.4).

  • J1 female 2x17 socket (O1 logic-board P8), J2 2x17 header (OEM cable), J3 2x17 header (Gotek signals), J4 1x04 header (switched Gotek power), J5 1x02 header (panel switch), U1 NC7SB3157P6X (DS-A), U2 TPS22918 (+5 V), Q1 AO3400A (GND), R1 control pull-down, R3/R4 DS-A pull-up, R2+D2 optional indicator (R2, D2 and R4 are dnp — on the board, not in the BOM).
  • Every J1 pin except DS-A (10) is drawn straight through to J2. DS-A goes to U1’s common; its B0 throw returns to J2.10 and its B1 throw goes to J3.10. +5V_SW and GND_SW reach the Gotek only via U2/Q1 → J4; the J3 header is signals-only, and +12 V, 4 MHz and pin 14 never reach the Gotek.

The capture’s guarantees, and the hazards it removes

the hazard the capture
a net label on every pin, and no wire between any two parts every net drawn as orthogonal segments on a 1.27 mm grid, routed by the Manhattan A* in gen_floppy_adapter_sch.py
103 labels doing the connecting one label per signal net (21), so the names survive and a missing wire breaks the netlist instead of being hidden
GND/+5V/+12V as labels routed like signals, with one power symbol per rail — a symbol per pin would stack 3.81 mm glyphs 2.54 mm apart on J1’s eleven ground pins — plus a PWR_FLAG per rail
every Footprint field empty 13 of 13 assigned, from KiCad’s own libraries
a 4PDT relay, K1, with no footprint and logical pin numbers three silicon switches whose pin numbers are the parts’ own (§4.3, §4.4)
nothing regenerated or compared tools/check_schematics.py now asserts the pinout of J1 and J2, which throw of U1 is which device, the control node’s three consumers, the connector and switch footprints, and that the committed file matches its generator

Verified, not asserted:

  • kicad-cli sch export netlist matches §5’s net table; kicad-cli sch erc --severity-all is 0 errors / 19 violations, all 19 the one kind (embedded local: symbols, one per placed symbol).
  • kicad-cli sch upgrade — the operation that detached 14 pins on the WiFi board — leaves the netlist identical. 0 overlapping and 0 duplicate collinear segments, because the generator normalises its own geometry.
  • python3 tools/check_schematics.py --board floppy-adapter and python3 tools/check_boards.py pass.

Three defects came out of the capture, all found by comparing netlists rather than by looking at the sheet:

  1. A wire ended on a no-connect pin. The router blocked other nets’ pins but not the pins marked no-connect (they carry no net, so they are not in the net-holding table), and the first wired netlist grew J3.14 and K1.11 onto FLP_DS_A_GOTEK. A no-connect marker says the pin goes nowhere; the netlist believes the wire. The sibling generator in hardware/wifi-comm-pac/ had the same latent hazard and was given the same fix.
  2. tools/check_schematics.py was carrying a pinout table that contradicted the design — it read “pins 2/4/6 are GND” against §2’s “pins 4 and 6 are TG43 and EARLY, never ground”. Nothing called the table, so nothing noticed; it is now the check for J1 and J2, which is what makes it true or false.
  3. The relay’s pin numbers were placeholders and its footprint was missing, which is what the switch decision above finally closed — with the parts’ own pin numbers, taken from their datasheets and cross-checked against KiCad’s library symbols.

State: the schematic is complete — the next step is the PCB

Every decision this design needed is made (§8), every net is drawn, every part has a footprint, and the two things that were left hanging — the switch part and the switch connector — are closed. The board is ready for PCB floorplan and routing, which is the next unit of work: place the connectors on their faces, set the keep-outs against the mainboard and the pocket, and route it.

The one thing the floorplan must respect is the reason the switch is silicon: height is the binding constraint, and the three switch packages are ~1.5 mm against a relay’s 10 mm.

10. References