Starting from a design that was already finished
The project began with three briefs and an approved design canvas, which is an unusual place to start and made the first real decision an easy one: build what was drawn, and don't redesign it on the way.
Most side projects begin with a blank page. This one began with a masterdoc, a design brief, a build brief and an exported design canvas — six games' worth of researched content, a dark shell, Cormorant Garamond over Lora with JetBrains Mono for figures, and a per-game accent that runs white to grey in release order.
That made the first decision less about taste than about discipline. The artboards carry exact tokens, so every colour, size and radius in the build is lifted from them; a literal hex or a bare 14px anywhere in a component is drift, not judgement. Three things were added that the design does not have, and each is written down with the brief that required it: search in the header's empty third column, the scope line under the homepage hero, and the "how to play it today" section on each hub.
One reversal was inherited rather than made. The design pass had tried a centralised structure — an engine hub, a file format reference, a community directory — and rejected it for pulling attention away from the game a visitor actually arrived for. Six self-contained game hubs won. That decision has held through everything since, and it is why the format specs, buildings, campaigns and mechanics pages all hang off hubs rather than sitting in an index.