TFCL AI

Dataset monitor

Connecting

DEMOS

— / —

Checking

TOTAL LEARNING TIME

— hours

—

TOTAL BRAIN + TRAINING DATA

—

Brain + verified training data

DATA ON DISK

—

DISK AVAILABLE

—

STORAGE BOX ARCHIVE

—

DATASETS—

DatasetMetadataValidatedStorage

Sixescp_process

———

BBallball maps

———

Ultiduoultiduo maps

———

Balooultiduo_baloo*

———

CURRENT PROCESS / TASKIdleNo active task reported

DEMO DOWNLOADS—

Waiting for download telemetry

DEMO-BASED POLICY PRETRAINING—

STATE—

TRAJECTORIES—

NEXT MILESTONE—

LAST TRAINED—

MOVE ACCURACY—

AIM ACCURACY—

Waiting for training telemetry

BALOO LIVE REINFORCEMENT LEARNING—

POLICY VERSION—

GENERATION—

EPISODE—

LEARNING TEAM—

EPISODE TIME—

SOLDIER REWARD—

MEDIC REWARD—

TRAINED EPISODES—

NEXT UPDATE—

NEXT CHECKPOINT—

TRAINING—

TRAINER ERRORS—

DEPLOYMENT—

CHECKPOINT SHA—

REPLAY SYNC—

REPLAY FILES—

Waiting for reinforcement-learning telemetry

CURRENT EPISODE OUTCOME (LIVE)—

KILLS—

DEATHS—

CAPTURES FOR—

CAPTURES AGAINST—

DAMAGE DEALT—

DAMAGE TAKEN—

Learning-team totals for the in-progress episode

TRAINING DATA QUALITY—

ELIGIBLE ON-POLICY—

ELIGIBLE RED / BLU—

COMPLETED EPISODES—

REJECTED (GENERATION)—

REJECTED (STALE POLICY)—

REJECTED (SIDE SCHEDULE)—

MALFORMED / PARTIAL—

SYNC ERRORS—

Trainer gating counters — rejects are healthy filtering, not failures

LEARNING PROGRESS (PER 10-EPISODE UPDATE)—

PolicyK / DCapsDamage +/-Trained at

Waiting for training history…

Learning-team combined objective per training batch · struck rows were rolled back

TACTICAL LEARNING (BANDIT TUNER)—

TUNER—

ACTIVE PROFILE—

BLOCK PROGRESS—

BLOCKS SCORED—

BEST PROFILE—

EXPLORATION ε—

PARAMS (RANGES/HP)—

LAST ERROR—

ProfileBlock K−DBlocksMean K−DScored at

Waiting for tuner history…

Scripted rules stay fixed — the tuner learns rule parameters (ε-greedy over bounded profiles, one switch per 10-episode block). Tactical bot names show the active profile as tN.

ROCKET JUMP LEARNING (TRIAL & ERROR)—

LEARNER—

PHASE—

BATCH PROGRESS—

ATTEMPTS / MOUNTS—

MOUNT RATE—

TRIALS (ADOPT/REVERT)—

BEST SCORE—

ACTIVE PARAMS—

LAST ERROR—

TrialMountsScoreDecisionScored at

Waiting for rocket jump trials…

The tactical Soldier attempts a point-mount rocket jump; each attempt is scored (mount, proximity, self-damage, deaths) and a hill-climber perturbs jump params (pitch, fire delay, burst, drive, trigger window) — adopting improvements, reverting regressions. Plugin clamps every param.

NAVIGATION LEARNING (WALL MEMORY)—

LEARNER—

PHASE—

BATCH PROGRESS—

ESCAPES / SUCCESSES—

ESCAPE SUCCESS—

STUCK EVENTS—

TRIALS (ADOPT/REVERT)—

BEST SCORE—

ACTIVE PARAMS—

LAST ERROR—

TrialEscapesStuck/minScoreDecision

Waiting for navigation trials…

Tactical bots remember where they got wedged (per team, shared between teammates), flip the escape side on failure, and pre-emptively curve around remembered spots. A hill-climber tunes stuck threshold, recovery burst, turn angle, avoid radius, spot memory TTL, and merge radius — scored on escape success minus stuck frequency.

GAME SERVER HEALTH—

CONTAINER—

PLUGIN VERSION—

LOAD AVERAGE—

MEMORY FREE—

LIVE HOST—

TRAINING HOST—

x86 game server + ARM training server reachability

LIVE ULTIDUO EXPERIMENT—

MODE—

BOTS ALIVE—

TELEMETRY TICKS—

ROCKET-JUMP CANDIDATES—

STUCK SAMPLES—

RECOVERIES—

POINT DISTANCE—

EPISODE LOG—

No live experiment telemetry reported