Reinforcement Learning / 30x30 Tactical Arena
RL-Arena
Fight checkpoint-trained bots across four difficulty levels in a tight tactical arena.
Checkpoint provenance and evaluation
- Training revision
- Unknown for this historical checkpoint
- Environment version
- Historical version not recorded
- Evaluation conditions
- Historical opponent pool and sampling seed were not preserved. The recorded win rate is not a human-opponent benchmark.
- Recorded evaluation
- Loading checkpoint
Current code parity checks compare 25 controlled transitions: neutral habit, no manual dash direction, and a matched 100-tick clock. Training uses 100 ticks; interactive rounds use 180. Existing weights are retained, not retrained or retrospectively assigned a source version.
Round Ledger
Round 01Round 1. Selected checkpoint loaded.
You
0
Bot
0
Draw
0
No finished rounds yet. The first few rounds are where exploration is most visible.