Sunrise Runner

The problem. Couch-to-10K plans assume every week goes as written. Real beginners miss sessions, find a week too hard or feel a twinge, and a fixed calendar either pushes them into injury or makes them start over. Sunrise Runner is a bilingual Arabic and English coach that takes an absolute beginner to a continuous 10 km, with a plan that moves according to what the body actually did. Sign-in is required, through an email magic link.

The hard part. The coach is a deterministic engine, not an AI, so every decision can be explained and tested. It climbs a 24-rung ladder from eight rounds of 60 seconds running and 90 seconds walking up to 90 minutes continuous, earning effort credits along the way. Safety rails override the credits: at most one rung up a week, weekly time on feet capped at 10% over the four-week average, a deload every fourth week, and a pain flag that removes credits and books rest, and a second flag within 14 days freezes progression.

The approach. The engine lives in a framework-free package with no React, Next.js, Prisma or fetch, and the current time is always passed in as a parameter. Golden-scenario tests replay six named scenarios, each defending one rule against it. The session player is a shared reducer built to work offline, resumes from IndexedDB and queues its sync. Strava and Whoop adapters import outside activity, and Whoop recovery is shown as a hint only, and every chart has a data-table equivalent.

The outcome. The app is live as a PWA on Next.js 16 with Prisma on Neon Postgres and a Serwist service worker. Stylelint enforces logical CSS properties so the Arabic layout cannot regress, the end-to-end suite runs axe, RTL and mobile-layout specs, and the core package is held above 90% branch coverage by rule, within more than 1,500 tests over 171 commits. It is a training tool, not medical advice; onboarding requires a medical-disclaimer acknowledgement.