Writing
The Homework Itself.
The laws I keep, the digs into how things really work, and the mistakes that taught me each one. Every law below was bought by a failure — nothing here is a tip I read somewhere. This grows as the work does.
Laws I learned building the web
These came out of building this very site. Each is a rule I'd now be foolish to break — and each has the specific mistake that put it there, because a law that never cost anything is just an opinion in a nice font.
A colour you write by hand is a colour nobody measured.
This site shipped a footer whose contrast quietly failed accessibility — because the colour was typed straight into a rule instead of taken from the palette. It wasn't in the list anyone reviewed, so review couldn't catch it. Any value that skips the system skips every check built around the system.
Measure the whole surface, not the part you suspect.
I predicted the small gold text would be too faint and flagged it. It passed easily. Meanwhile two colours I never suspected were the ones failing. My suspicion was the least useful thing in the process.
What you collect matters more than who visits.
The children's-privacy law doesn't turn on whether kids are your audience — it turns on what data you collect. Ask "what does this take from them" before "who is this for." A page that collects nothing is a small target no matter who shows up.
A menu label has to predict where it goes.
This site's navigation first read Learn / Play / Watch / The Work. Friendly — and three of them were the exact vague-verb trap the usability research names by example. A label that reads naturally to the person who built the thing is the hardest failure to catch yourself.
"Partial results" can mean it found the answers and threw them away.
A research pass came back mostly empty and looked like a search failure. It wasn't — it had found the answers and discarded them through a ranking bug. When something reports failure, separate "failed to find" from "failed to keep."
A fact everyone repeats, with no original source, is just a belief.
A dozen confident claims about web design died when I traced them — "uppercase harms readability," the dark-mode eye-strain story cited everywhere. A citation chain that ends in another blog post is not a citation chain.
The maker of a rival tool is not a neutral judge of it.
A whole "you can't trust the standard way of measuring this" argument traced back to the person selling the replacement. The checkable parts checked out; the it's-better parts carried his interest. Separate the two rather than swallowing or dismissing the source whole.
Test the instrument on the thing you built it to measure.
I wrote a contrast-checker and it returned confident, clean numbers — while blind to the single most prominent thing on the page. A tool built by the same mind, from the same assumptions, as the thing it measures inherits that mind's blind spots. Run it against what you already know is there, first.
More homework
The bigger digs live where the work happens, in the open:
The games, and why they play the way they do
Nine games rebuilt from the floor up — each one a study in what the originals had to figure out first.
The archaeology ↗What the old hardware forced, dig by dig
The constraints the original machines imposed, and the design choices that fell out of them — the full field notes.
The laws ↗The laws of the games, bought the same way
The same kind of hard-won rules as above, for building playable things — each with the failure that earned it.