TODO
The web page at blinkingcomputer.org/web is the place to iterate: it can change in minutes, a board cannot. So the boards are decided on the page first, and this list is what the page has to answer before anything is sent out. Started 22 Nov 2026; the newest items are added where they belong, not at the end.
Two rules for the list, 22 Nov 2026. Tony is the first tester: if he cannot read a block from the page and say what it does, nobody in a class will, so nothing is finished until he has read it. And the boards are meant to make people say "wow" and then understand, so a track that is longer than it needs to be, or a label that cannot be read, is a defect even when the circuit is right.
Before committing to boards
- Feel the column pitch: 7.3 mm against 7.85 mm. (Tony.) The page has a pitch selector in its header and a 1:1 button that draws the close-up life size on a 96 dpi screen, with a scale bar to check against a ruler or a credit card (85.6 mm). What the two numbers mean on the bench: a slot is one part's room, 7.3 by 6.5 mm, less than half the area of a fingertip; a TO-92 is 4.6 to 5.2 mm across, so at 7.3 the bodies in a row are 2.1 to 2.7 mm apart and at 7.85, 2.65 to 3.25 mm, a visible gap either way; a 3 mm LED and a 1/8 W resistor (3.3 by 1.7 mm body, which is what a 5.08 mm pad pitch allows flat; a 1/4 W body is 6.3 mm and would have to stand up) sit easily in either; the pads are 1.6 mm round on a 5.08 mm pitch in both. The difference is the gap between the pads of neighbouring columns, 0.62 mm at 7.3, 1.17 at 7.85 and 1.72 at 8.4. A 0.25 mm track with 0.15 mm clear on each side (JLCPCB's minimum is 0.127) needs 0.4 mm of gap, so 7.3 holds one track, 7.85 two and 8.4 three; 7.85 holds three only with 1.4 mm pads at the bare minimum. (Corrected 23 Nov 2026: the page had claimed three at 7.85, by measuring to the pad centre instead of its edge. The page now derives the lane count from the pitch and the test checks the clearance.) The more lanes a gap has, the more of the slices' wiring stays on the top side where it can be read, but two tracks 0.6 mm apart also read as one line from a metre away, which is the other half of the question. A slice is 60 by 299 mm at 7.3, 63 at 7.85 and 67 at 8.4; the panel 1,040, 1,088 or 1,152 mm wide. Decide by looking, then fix the pitch in the design doc.
- The control board's wiring. At either pitch the control board needs two to three vertical tracks per column gap across a cut, and decode and the line stages more, because one control line reaches sixteen cells. The choices are the bottom layer for the control board's verticals (the top stays readable for the slices, less so for control), a wider panel, or taller, narrower control blocks. The test prints the residue (pairs of different-net track segments drawn over each other) after every change.
- The function strip on one-column tiles. A single transistor's tile is 7.3 mm wide and its line (
SUM: bus4 = 0 when ns4ยท~SUM_EN) is 15 mm at 1 mm letters. Run it up the tile's side, abbreviate it to the base term, or give those tiles two columns.
- Resistors stood upright? An axial resistor on end takes a 2.54 mm footprint, half a slot. The 547 LED loads are a tenth of the slots; all 2,000-odd resistors upright would be a fifth. Costs readability of the colour bands. Measure on the page before deciding.
- The rail. 23 control lines at 0.48 mm pitch is 11 mm at the connector, the one tight place on the slice; the real board tapers it and may put part of it on the bottom layer.
- First KiCad tile. Draw one register cell and one gate tile with real footprints and the silkscreen names from
web/leds.tsv; it calibrates every slot and area number above.
- Read the tracks, as the first tester. (Tony.) Since the evening of 22 Nov transistors face the row their resistors are in and a pin can escape into the channel above or below, so a flip-flop's cross-coupling is one straight run between its two rows instead of up, across and down; the function strip's words moved to the top of the strip and the lanes prefer to stay below it. What is left: one crossing per flip-flop is inherent (the two cross-coupled nets interleave along the channel, the same crossing the schematic has), and the test counts every place two nets still cross or lie on top of each other. Report the places that still read wrongly.
- Smaller things on the page. The strip text and power symbols have not been checked in a browser by the model that wrote them; report anything that collides. The LED names at 1.5 mm and the per-block words are the silkscreen; the page draws them from 2.2 px/mm.
The explanation
- Why, what, how for every block. (Tony, in his own words.) One or two sentences per block so that a pointer and a pause in front of the boards can say why it is needed, what it is doing and how. Shown on the page when the pointer is over the block; later printed on the board edge. The design doc has the material; a draft can be cut from it on request.
- The example programs.
const (a constant into A) is the first. Next: a copy, an add with carry seen crossing slices, a conditional jump taken and not taken (and P stepping over the word either way), a loop, then the Forth words.
- The assembler's two forms. The table of short forms and their expansions goes in the design doc as the assembler's definition, and the program pane shows the short form beside the full one.
- The software list. software.md has Tony's candidates, from Pong to Doom, each with what it costs in instructions and a verdict. Pick the first three.
The host
- Real-machine stream from the Pi into the page in place of the twin, and a camera view beside the drawn boards.
- Parallel or serial between the Pi and the machine: still open.
- The host program on the Pi Zero.