Roughly, vibe-prototyped

Roughly is playable ▸ 5:57. The thesis got a sharper phrasing on the way: everyday math isn’t exact, it’s the skill of knowing “si me tumbaron, que no me tumbaron mucho” ▸ 6:22, and no app he can find teaches approximation, everything grades exactness ▸ 7:37. Julia contributes a use case from every standardized test: estimation is how you discard three of four multiple-choice options in a second ▸ 8:00.

The build story is the era in miniature: he handed Gemini’s Canvas the one-pager, and the model interpreted the idea, adding the countdown timer he had never specified, a design decision made by the tool that he ended up liking ▸ 13:54. Canvas hosts the mini-front itself and hands out a shareable link ▸ 9:51, prototype to distribution with zero deployment. Best signal available: its own maker keeps playing it longer than he expected ▸ 9:31.

los errores de mil usuarios son el mapa de dificultad →

The open problems are game design, not code. A three-digit sum on question two frightens off the beginner ▸ 10:23, but he refuses both fixes on offer: hand-tiered difficulty modes and predefined levels violate his standing architecture creed, minimal structure so the user cuts their own path and the maintainer maintains almost nothing ▸ 11:47. The target feel is Candy Crush, whose level 36 is the same game as level 1 and addictive anyway ▸ 12:08. His candidate: let the crowd do the tiering, aggregate a thousand users’ hit rates per exercise type and the difficulty map writes itself ▸ 12:44. Meanwhile Reisi’s prototype is done and waits only for Julia’s trailer ▸ 14:36, deliberately kept offline, since the validation needs viewers to get the idea, not testers to find the edge cases ▸ 15:04…