COGNITIVE FACILITY
Systematic cognitive training grounded in 2020–2026 neuroscience: Relational Integration, Desirable Difficulties, Continuous Calibration, and Empirical Transfer Auditing.
Intra-Individual Response Time Variability (IIV) Vigilance Monitor
Tracks micro-lapses in prefrontal executive control to abort training before neural exhaustion.
Recommended 45-Minute Daily Training Protocol
Rapid motor inspection to calibrate today's baseline latency variance.
Multi-premise relational integration (R1 to R4+) to exercise frontoparietal binding.
Cold reconstructive recall under interference. Synaptic consolidation.
Continuous probability credences (0%–100%) and pre-feedback error hypothesis.
Audit Transfer Index τ to prove genuine generalized cognitive gains.
7-Domain Latent Ability Profile (IRT Ability Parameter θ)
Separated psychometric constructs preventing the “single brain score” fallacy.
[ COGNITIVE TRAINING ARENAS ]
Select Arena to EnterProcedural Training Engine
Unified catalog of 10 validated experimental paradigms (Stroop, Flanker, N-Back with lures, OSPAN, Task Switch, WCST, Raven Matrices, Bayesian, Choice RT). Procedural generation with IRT difficulty vectors.
Relational Integration Gym
Multi-premise relational binding (Wang et al., 2025). Coordinate 2 to 4+ simultaneous premises across numerical, causal, and verbal modalities.
Reconstructive Retrieval
Desirable difficulties engine. Learn dense technical concepts, survive cognitive distraction, and execute cold reconstruction under exponential decay.
Metacognitive Calibration
Assess subjective probability on continuous 0%–100% slider. Predict failure modes before feedback. Minimize quadratic Brier Score penalty.
Transfer Verification Audit
Audit true cross-domain generalization. Calculate mathematical Transfer Index τ = (0.6·Near + 0.4·Far) / Trained to eliminate task-specialization illusions.
Feynman Epistemic Arena
Deconstruct complex scientific concepts into zero-jargon definitions, analogies, boundary condition stress-tests, and counter-intuitive edge cases.
Calibration Lab (Baseline 9)
Our original 9-test psychometric battery: Reaction Time, Simon Span, Ayumu Chimp Test, Visual Memory, Aim Trainer, Typing, and Reading.
HumanEval does not make marketing claims that repetitive brain-game puzzles increase biological IQ. Modern preregistered double-blind RCTs (2025) systematically demonstrate that isolated N-Back gains fail to produce statistically significant far-transfer. HumanEval conditions core relational binding (Wang et al., 2025), frictional reconstructive retrieval (Karpicke & Roediger), and metacognitive calibration (Tetlock), while explicitly measuring the mathematical Transfer Index (τ) against untrained tasks.