One line, one day
Eleven hours lost to an auth redirect, solved by resetting to the last good commit and one ChatGPT snippet. The anatomy of a vibe-coding bad day.
// trace: where this idea came from
- ↳ video diary @ 3:45 (la última cosita, desde el backup)
- ↳ video diary @ 4:52 (era una pendejada)
- ↳ Seed 21-1: La basurita del código (la doctrina del reinicio, aplicada y funcionando)
The bug: Zenota’s Google sign-in popup said “firebase-something.com” instead of the app’s own domain, and fixing it meant teaching Netlify where to redirect an auth route. The cost: from ten in the morning until nine at night, modify here, modify there, doesn’t work, doesn’t work ▸ 2:31. The day was declared lost out loud.
Then the recovery, which is the teaching. One last attempt, and its first move was not another patch: it was going back to the last commit where everything worked ▸ 3:45. That’s seed 21-1’s doctrine in action: a day of layered failed fixes is sediment, and the cheapest path forward starts from clean ground. From there, with the logic finally understood instead of guessed at, ChatGPT produced a one-line snippet, the redirect reference Netlify needed ▸ 4:08. Something said it would work ▸ 4:46. It worked. “It was that? A pendejada. I searched everywhere, and it was one line” ▸ 4:52.
once horas, una línea →
fix size: 1 line
what changed the outcome: reset + understanding, not more patches
The ratio is the lesson. Days like this are not a sign the method is failing; they are its known tax, and the root cause was a three-service tangle (domain at Hostinger, hosting at Netlify, auth at Firebase) where nothing is any one vendor’s fault. What shortens them is a ritual, not heroics: when patches stack without progress, revert to known-good, re-derive the actual logic, then ask again. The one-line answer was always available. It only became findable from clean ground…