Overview
The Florent Code League is a Nordic bot-programming competition run by Florent Venture Partners. Teams write an agent that plays on its own — securing resources, expanding territory, and fighting for the arena — and every entry is ranked on one live ladder across the region.
The 2026 edition ran three weeks online through August, with the top 32 teams qualifying for the live finals in Stockholm and a €20K prize pool. I competed solo from Helsinki, writing the bot in Python.
Result
- Top 10 overall, against competitors with prior MIT and Cambridge Battlecode experience
- First place on the ladder for several days
- Qualified for the live finals in Sweden — the top 32 of the region
Key Features
- A single Python agent running both economy and army each turn
- Harvesters, conveyor networks, builders, and turrets under one decision loop
- Per-map routes and terrain precomputed offline to save the turn budget
- Champion-versus-challenger workflow: nothing shipped before beating the live bot
- Replay analysis tooling to explain why a change worked, not just whether it won
Tech Stack
Challenges & Learnings
You never debug the bot directly — you debug thousands of turns against opponents you cannot see, on a ladder that also moves when someone else improves. Small samples lie, so the habit that paid off was writing down what would falsify a change before running it, and measuring the mechanism it touched rather than the win count.
A per-turn CPU budget also means an elegant plan that misses the deadline scores nothing. Most decisions came down to what was affordable rather than what was optimal — precompute what you can, keep the hot path cheap, and prefer the simple behaviour that never fails.