It learns to win the fight. Watch it.
Offense is where you find out if a brain is real. Ours plays shooters and tank games from the screen and the engine, gets better from its own losses, and takes orders like only the chainsaw or don’t fire as hard limits.
Nothing here has been in the field. It has been tested in games and simulation. Here is what that proves and what it does not.
Cross-compare: attack results, game by game
| game | our brain | what went against us |
|---|---|---|
| DOOM E1M1, Hurt Me Plenty | exit switch at 52 s, 4 kills | Nightmare: not cleared in 8 tries |
| DOOM — rules with no learning | 17.9 vs random play’s 3.2 | memory on top: even |
| DOOM E1M5 — learning | route progress more than doubled, 0.19 → 0.43 | no exit on E1M5 yet |
| Atari BattleZone (tank) | 27,000 vs random 3,000 | a dodge rule was rejected |
| BattleZone — a brain moved in from another tank game | 33,575 vs the game’s own rules 28,600; after pruning 48,588 vs 29,275 | a second export was only even |
| 3D tank arena (BZFlag) | net kills per minute −2.05 → −0.26 across versions | still loses to BZFlag’s own AI |
| Robot Tank (Atari) | 5.67 tanks destroyed vs random 2.33 | hand-written rules, not learning |
| GoldenEye 007, Dam | invented 2 tactics that beat the default | 0 of 72 attempts survived |
Aim is counted, not claimed: in the tank game the eye’s aim error is 0.55° at the 95th percentile.
Would you want a chatbot driving your tank?
We put the big language models in the same 3D tank arena, each tank with its own driver. One 10-minute match:
| driver | kills – deaths | note |
|---|---|---|
| BZFlag’s own AI (years of tuning) | 38 – 6 | beat us too |
| Peel rules | 20 – 20 | beat every language model |
| DeepSeek | 10 – 18 | 1.0 s per decision |
| Claude | 1 – 15 | 7.6 s per decision |
| Llama 3.2 3B, local | 0 – 10 | small local model |
On a written exam of 8 tank situations with no clock, Peel’s rules scored 8/8, Claude 6/8, DeepSeek 5/8. In simulated driving, every language model crawled at about 8–10 km/h. Honest caveats: one match, not a tournament; Claude ran through its command-line tool, so part of its 7.6 seconds is plumbing; and BZFlag’s own AI beat our rules.
Why not just put an AI model in charge?
Orders inside its authority, it carries out. Orders outside it, it refuses and tells you why. It does not argue, and it does not improvise new authority. A neural model still has a job: seeing and reading. Peel does the deciding.