i18n + Tarot + Compatibility in One Day: Constraint Design Is the Real Skill

I added AI tarot reading, rebuilt compatibility analysis, and overhauled translations across 8 languages — all in one day. 20 commits, 139 files changed. What made this possible was not coding speed. It was prompt decomposition. What I Was Building A saju (Korean astrology) analysis service built on Next.js + Supabase + OpenAI API. Saju uses Four Pillars of Destiny — year, month, day, and hour of birth — to generate personality and fortune readings. This session added three features: AI tarot reading (78-card deck) Compatibility analysis overhaul (birth time input, 6 sections, pricing) Translation fixes across 8 locales (ko, en, ja, zh, hi, th, id, vi) Doing all three manually in one day is not realistic. Doing all three with structured AI prompting is. Decompose or Fail Tarot reading requires card deck data, UI components, LLM interpretation prompts, and i18n strings. Asking Claude to build it in one prompt guarantees failure. Too many moving parts, too much left to the AI's discretio