The Science | NeuroCrush

📐 What We Measure

NeuroCrush measures Working Memory (WM) — the cognitive system that temporarily holds and manipulates information. WM is one of the best-documented predictors of learning speed, job performance, academic achievement, and early cognitive decline.

We isolate WM using task paradigms with strong construct validity: complex-span tasks (Operation Span, Symmetry Span), n-back variants (position, audio, dual), and paced-serial addition (PASAT). Each game targets the phonological loop, visuospatial sketchpad, or central executive sub-systems described in Baddeley's (1986, 2000) model.

Our WM Composite Index is a weighted average of percentile scores across all games played, normed against published age-stratified reference data. Weights are set based on each task's loading onto the unitary WM factor from confirmatory factor analyses in published work (Conway et al., 2005; Shipstead et al., 2012).

🎮 Game-by-Game Evidence

⚙️ Adaptive Scoring Model

NeuroCrush uses a parametric adaptive model inspired by Item Response Theory (IRT). Difficulty adjusts continuously based on accuracy and response time, targeting each user's Zone of Proximal Development.

P(correct | θ, a, b) = 1 / (1 + exp(−a·(θ − b)))

θEstimated latent ability (WM capacity) for this user aItem discrimination parameter (how well the item separates ability levels) bItem difficulty parameter (the ability level where P(correct) = 50%)

In practice, we use a simplified Staircase Adaptive Procedure: N-level promotes after 2 consecutive sessions ≥80% accuracy and drops after 2 consecutive sessions <50%. Stimulus timing adapts separately at 0.1s steps between 0.5–10s. This is equivalent to a 2-up/1-down adaptive tracking procedure with an asymptotic threshold convergence.

Response times are logged at frame-rate precision (no network round-trips — all timing is done client-side using performance.now() for sub-millisecond accuracy).

📊 Normative Database

Scores are compared against published normative data stratified by age. The table below shows the mean and standard deviation for the N-Back composite (nLevel × accuracy) used to compute your percentile.

Age Band	Mean N-Back Composite	Std Dev	Source
18–25	2.8	±0.70	Jaeggi et al., 2010
26–35	2.6	±0.65	Jaeggi et al., 2010
36–45	2.3	±0.70	Redick et al., 2011
46–55	2.0	±0.75	Wechsler MAS Standardisation, 2009
56–65	1.7	±0.80	Wechsler MAS Standardisation, 2009
66+	1.5	±0.85	Park et al., 2002 (MIDUS study)

Digit Span norms: Wechsler (2009) WAIS-IV. Operation Span norms: Conway et al. (2005). PASAT norms: Gronwall (1977) updated Tombaugh (2006). Stroop norms: Golden (1978) updated Strauss et al. (2006).

🔄 Reliability & Practice Effects

Published test-retest reliability (ICC) for our core tasks:

N-BackICC = 0.78–0.85 (Jaeggi 2010, Schmiedek 2010)
Operation SpanICC = 0.82–0.88 (Conway 2005)
Digit SpanICC = 0.84–0.90 (Wechsler 2009)
PASATICC = 0.76–0.84 (Tombaugh 2006)
StroopICC = 0.80–0.86 (Strauss 2006)

Practice effects are mitigated by alternate puzzle forms (different stimulus pools across sessions), adaptive difficulty re-baseline on session start, and excluding the first 3 trials from scoring (warm-up exclusion). We track longitudinal score trajectories to distinguish genuine WM gains from learning-to-take-the-test effects.

🔒 Data & Compliance

NeuroCrush is designed for responsible data handling whether you're a consumer user or a clinical/HR professional.

🇪🇺

GDPR Ready

Data minimisation, right to deletion, no third-party analytics

🏥

HIPAA Aware

B2B accounts use separate session tokens; PHI never leaves server

🔐

Local-First

All timing & scoring runs in browser. No raw RT data leaves device

📋

Wellness Label

NeuroCrush is a cognitive wellness tool, not an FDA-cleared medical device

Important: NeuroCrush scores are not diagnostic. They are normative wellness indicators. For clinical diagnosis of MCI, dementia, or ADHD, consult a licensed neuropsychologist using validated clinical batteries (e.g., MoCA, RBANS, BRIEF-2).