🛸 Orbit Wars v12 — Full-Spectrum Domination Agent
For the Orbit Wars Kaggle Competition
Test Results
| Matchup | Result |
|---|---|
| V12 vs Random | 3/3 wins (eliminated by step ~90-99) |
| V12 vs Starter | 3/3 wins (eliminated by step ~73-159) |
| V12 vs V10 (1v1, alternating sides) | 5/6 wins (83%) |
| V12 vs 3× V10 (4-player) | 3/3 wins (eliminated all 3 every game) |
How to Submit
- Download
submission.py - Create a new Kaggle notebook
- Copy the content into the notebook
- Submit
Architecture
Built from v10/v11 base + every technique from 1200+ Elo public agents:
- 110-turn Forward Timeline Simulation — binary-search exact keep_needed
- Surplus-based dispatching — send EVERYTHING above keep_needed
- 7-mission priority system: reinforce > crash_exploit > capture > snipe > swarm
- Multi-source coordinated swarm attacks (2/3 sources, ±2 turn sync)
- Snipe detection — arrive at neutrals right after enemy fleet battles
- Crash exploit — arrive after inter-enemy fleet collisions (4P)
- Hostile reinforcement prediction — estimate enemy can reinforce target
- Doomed planet evacuation — save ships from unsaveable planets
- Exposed planet detection — attack enemies who just launched fleets
- Dynamic game phases with domination tracking
- Weakest enemy targeting with elimination bonus (4P)
- Production snowball scoring: score = prod × remaining_turns / cost
- Proactive defense with stacked multi-enemy threat detection
- Sun-aware fleet routing with iterative intercept solver
- Indirect wealth valuation (cluster bonus from nearby planets)
- Endgame consolidation to highest-production planets