Documentation & Architecture
System architecture, statistical methodologies, cryptographic privacy, and data ingestion specifications.
Local-First WebAssembly Architecture
Client-Side Pyodide Worker
All heavy statistical modeling, matrix factorizations, and linear regressions execute entirely within dedicated background WebWorkers running Pyodide (CPython compiled to WebAssembly). Raw biometric and environmental data never leave your browser for analysis.
Dexie IndexedDB Storage
Measurements and studies are cached in browser-local IndexedDB with optimistic offline mutations. Changes synchronize bidirectionally with optional backend endpoints or peer browser nodes via WebRTC.
Statistical Analytics & Algorithms
Correlation Models
Pearson linear (r) and Spearman rank (ρ) coefficients with two-tailed p-value hypothesis testing (statMath.ts).
Time-Lag Windows
Continuous lead-lag cross-correlation offsets ranging from -48h to +48h to identify delayed physiological or environmental responses.
Decomposition & Causal
Loess STL time-series decomposition into trend, seasonal, and residual components, plus Granger vector autoregression for causality scoring.
Methodological Guardrails:
Studies report Benjamini-Hochberg False Discovery Rate (FDR) corrections for multi-stream hypotheses to prevent p-hacking and spurious associations.
Cryptographic Privacy & Federated Research
AES-256 BYOK Storage
All credentials, OAuth tokens, and patient identifiers are encrypted client-side using Sodium / Web Crypto AES-256-GCM prior to storage.
SMPC Masking
Federated studies utilize Secure Multi-Party Computation with zero-sum peer noise vectors. When aggregated across the quorum, noise masks sum to zero.
Differential Privacy
Calibrated Laplace and Gaussian perturbation mechanisms ensure rigorous (ε, δ)-differential privacy guarantees for multi-center research cohorts.