The bench went on the grid

When I wrote about the models repo in August it held a dozen things, and the pitch was that the STL is an output you regenerate, not a mesh you’re stuck with. That’s still the pitch. What changed is the bench.

Over the summer the whole repair bench got reorganized onto Gridfinity — the magnet-free Clickfinity flavour, on shallow plates. Once the surface is a grid, every tool that used to sit on the bench wants a bin, a dock, a riser, or a plate that puts its footprint back into the grid. So the repo went from twelve models to thirty, and nearly everything on the bench that isn’t an instrument is now a print from it. The catalog has the full list; here are the ones with a story.

The shaft that is the spool

Solder wick comes on spools with a small bore. The downloaded holder I was using has a ⌀14.6 shaft, which is too fat for them. The obvious fix is a thinner shaft. The better fix is a shaft that is a spool — flats on the journal, a winder plate, a crank — so you wind wire or wick straight onto it and there’s no bore to match at all. Base and brackets unchanged. Now it holds anything.

Nine posts for ten nozzles

The hot-air rack has nine posts. The wand came with ten nozzles. That’s not a mistake: one nozzle is always on the wand, and a 7×1 rack that held all ten would be longer than the printer bed. Nine covers ten. Small design decisions like that are most of what a bench fixture is.

A mount that passes the mesh check and points at the wrong thing

The thermal camera mount is a two-part sandwich clamp on the ring light’s control tab, aimed down at the board. It went through more revisions than anything else in the repo, mostly because of one fact: a mount pointed at the microscope objective instead of the board is still a perfectly watertight, 2-manifold mesh. CI can’t see aim. So the model ships an aim-verification script that checks where the camera’s axis actually lands, because the failure mode was invisible to every other check.

Corner feet only

The filler tiles turn empty grid into a working surface — mouse, drink, wrist rest — and lift out to be storage again. The first instinct is to latch every cell. Don’t. A Clickfinity cell grips at roughly 12 N, so a 6×6 tile footed everywhere takes about 45 kgf to lift — a permanent fixture you’d break removing. Four corner feet cap release at about 5 kgf at any size, and underside ribs bear on the grid walls so the middle doesn’t sag. That rule now applies to every multi-cell part in the repo.

Measure, then print

Two parts in August came out wrong for the same reason: a number in the source that was assumed instead of measured. A syringe bore. A swab length. Both scrapped prints. The response wasn’t more careful guessing — it was gauges: print-first fit coupons that live in a coupons/ directory beside the model, get rendered and validated by CI like everything else, and get skipped by the release build so a release stays a set of parts you can print without reading anything first. Real measurements, no guesses.

The catalog is enforced now

Thirty directories is enough that the README table drifts if you let it. So it can’t. A catalog check runs on every PR and fails it if a model directory is missing its README, its preview, or its row in the table. No exemption list on purpose — a model that isn’t ready to be catalogued isn’t ready to be on main.

Everything’s on GitHub, each model with its own release, source under CC BY-NC and the library and tooling under MIT. If your bench lives on a grid, there’s probably something in there you can regenerate at your own dimensions.