I made a car during COVID, I probably should write that up a little But an unexpected part of "having a weird car" is "need weird shoes." What with the "need a weird wallet" / leather-making, I figured I should just crank out a set of driving shoes. How hard could it be, right?

  • first, what do you want to learn?
  • bootstrapping / "start with something that works"
  • automate the build where possible
  • Prototyping / "plan to throw one away"
  • A/B testing
  • production testing & deployment

A/B testing

How to tell if a new design is better?

In a software environment you'd run the new thing alongside the old thing in the same environment, and compare the results -- ask users, check metrics, whatever. The better one wins, iterate on that (and discard the other). At its core this is part of the scientific method: "A" is the control group / baseline, "B" is the experimental group / the thing that changes.

Turns out, you can do this with shoes also! The secret: make one at a time. Note: this may have a low WAF.

Wait, ONE shoe?

yeah, make one at a time. "Fail faster" (google it) -- there's not a lot of benefit to making two identical (or mirrored), untested shoes. Make one, try it. If it's perfect, make the other one. If it sucks, fix that on the next one / other side.

Since you make them super fast (as a side effect of making and discarding a LOT of singles) you'll only be in this state for like 1 day.

Make a single prototype (say, right foot). Wear it in your test environment(s): walking around, pressing very-closely-spaced pedals in a tiny pedalbox, whatever. Be sure to *also* wear the benchmark shoe on the other foot. Your shoes don't match, but nobody really cares. Also you're wearing hand-made shoes, "don't really match" is probably not very high on the list of priorities.

That's pretty much it. Wear the A + B shoes at the same time. If "B" is the preferred model, iterate on it. IF by some miracle "B" is the perfect shoe, mirror it and make the mate. But more likely you'll tweak another variable and make a slightly-changed mate shoe, then re-run the test with the "A" (formerly the test shoe on right foot) and the new "B" (new test shoe on left foot).

Rinse, repeat.

In shoes it's a little tricky, there are a lot of interacting variables:

  • material selection (soles, upper/lower, any mesh panels, etc)
  • size/shape
  • insole design / squishiness

Changing one of these may non-obviously interact with the others: added insole material (more squish) may make the shoe feel smaller, necessitating a change in size or shape. An outsole may change the shape and fit of the upper, or may obviate the need for insole. Etc. So tweak a small number of things and re-run the experiment.