Training lab
The whole roster keeps learning.
The league rotates through every athlete automatically. This page is the public control-room view; anything the worker keeps writes straight into the career save the ring page fights from.
Current rotationMalik OkoroAthlete 13 of 36 · middleweight
- Record
- 7-1
- Rating
- 86
- Build
- 5'11" · 160 lb
- Episodes
- 0
power78
speed91
cardio84
defense88
Ring IQ92
reaction94
Track 01 · standard trainingLeague sparring
idleEvery four minutes, a durable worker runs sixteen short episodes of the same 2D bout the ring page shows, tests three policy candidates, keeps only a better result, then hands the next turn to another boxer. It stays armed after every browser is closed.
TargetOkoro
Loop cycles0
Physics steps0
Sparring episodes0
Policies tested0
Checkpoints kept0
Reward fitness0.0
Session gain+0.0
Observe→Act→Step→Reward↻
Next learning gate0:00Live-fight corners are skipped automatically. A checkpoint is kept only if it scores better in deterministic sparring.
Rules engineFairness monitor
VerifiedExploration happens between episodes. During an episode, submitted actions enter the same deterministic physics the live ring uses — there is no roll anywhere in a result.
Outcome randomnessNone
Decision makerNeural athlete
Working memory8 recurrent units
Slow strategyBetween rounds only
Motor authorityNeural policy
Agent interfaceObserve / act / reward
ObservationDelayed + quantized
Live ringArticulated 2D
Ground forceEstimated from the articulated stance
EvasionHead level + angles
Reaction floor179 ms
ContactPrecise collider + glancing
Impact modelEffective mass × velocity
KnockoutEquilibrium releases control
Judge cards3 sealed
Physics step10 Hz