MATCH RESULT — STRATEGY LOG
Units deployed. Resources allocated. Objective: outlast, outmaneuver, win.
Overview
Battlecode is MIT's annual programming competition where teams build
autonomous bots to fight for control of a simulated battlefield — no human
input once the match starts. That year's game had units called rats and cats.
The strategy space is enormous: when to build, when to attack, when to hold,
when to spend resources and when to hoard them.
My teammate handled pathfinding and cheese — an exploitative early-game
strategy designed to catch opponents off-guard. I owned strategy design:
specifically two modes I developed, Kingbuilder and Kamikaze.
Kingbuilder, Kamikaze, and What Actually Worked
Kingbuilder: I discovered you could spawn multiple kings.
My instinct was that more kings meant more power — so I designed a mode
built around building as many as possible. It was wrong. Multiple kings
tripled resource costs and starved the rest of the army. The theory was
sound on paper. The math killed it in practice. I learned more from
that failure than from most things that worked.
Kamikaze: This one worked. The baby rats would relentlessly
swarm the cats — no retreat, just pure aggression on target. Combined with
my teammate's cheese keeping opponents pinned high up on the map, the
kamikazes had room to work. At our peak they were dealing over 3,000 damage
to the cats in a single engagement.
I designed the code with tweakable knobs throughout — adjustable parameters
that let me dial in thresholds and find the sweet spots without rewriting
logic every iteration. Resource management was a constant tension across
both modes: the Kingbuilder needed too much, the Kamikaze mode needed just
enough to keep the swarm alive long enough to do damage.
Outcome
A competition entry with two distinct strategic modes — one that worked and
one that taught us why it didn't. The project built real fluency in game
AI, resource constraint reasoning, and iterative tuning. It also confirmed
that the most interesting engineering problems are the ones where your
intuition is wrong and the math tells you why.
Source & Documentation
Code and match replays available upon request.