DiRT Tracker

The challenge

Shared spreadsheets fell apart fast: nobody remembered which stage, which car or which conditions a time was set under, so comparing was pointless. It needed a model that captured each run's context and a way to log someone else's time without handing over the session.

Implementation

• The game's real catalog, seeded

  The seed loads 13 locations, 156 stages and 79 cars with their data, and it's idempotent: rerunning it duplicates nothing and loses no already-recorded times.

• Times with full context

  Every entry stores weather, time of day, stage direction and input device (wheel or pad) alongside the car. Comparing a wet run on a wheel against a dry one on a pad stops being a pub argument.

• Millisecond precision

  Times are stored as whole milliseconds and formatted to MM:SS.mmm in the presentation layer, so leaderboards sort correctly and never drag floating-point errors around.

• Mutations as Server Actions

  Locations, cars, times, users and passwords are all modified through Server Actions validated with Zod and protected by session and admin-role helpers instead of per-route checks.

• Forced password change on first login

  Auth.js v5 with credentials and a JWT session; an edge-safe middleware redirects to the password change screen before letting anyone reach any other page.

• Same schema locally and in production

  PostgreSQL 16 in Docker for development and Neon in production, with the same Prisma migrations applied to both.