I/O

An instrument for the measured life.

Aggregate every signal. Mine your own data. Run your own trials.

Under Construction
Instrumentation

What I/O does

Most biohackers track dozens of signals — sleep, HRV, supplements, workouts, mood, blood markers — across a fragmented stack of apps and wearables. The data exists, but the connections between input and outcome remain hidden in the noise.

I/O aggregates every signal you already collect, runs rigorous associative-rules mining on-device, and surfaces input→outcome hypotheses with full statistical transparency. When a hypothesis looks promising, I/O converts it into an N-of-1 crossover trial — the same validation standard used in clinical research, applied to your own data.

The Engine

Methodology

I/O's mining engine doesn't guess. It applies a battery of statistical methods directly to your local data — no cloud processing, no black-box LLM inferences. Every surfaced rule carries support, confidence, lift, FDR-corrected p-value, and sample size.

Triple Dependency

Pearson r, Spearman ρ, and distance correlation (Székely et al.) computed for every input-outcome pair. Reports the strongest with full transparency.

Confounder Conditioning

Auto-generates candidate confounder sets and computes partial correlation. Flags pairs where conditioning reduces effect size by ≥40%.

Seasonality Residualization

Decomposes outcomes into day-of-week effect, baseline trend, and residual before any correlation is computed.

Block Permutation

Respects AR(1) autocorrelation via 7-day block permutation (Politis & Romano, 1994). No naive shuffling that inflates false positives.

FDR Correction

Benjamini-Hochberg exact step-up on all mined rules. Only associations with pFDR < 0.1 surface by default.

Robustness Scoring

Bootstrap stability + leave-2-out replication. Rules below 0.7 carry a warning; ≥0.85 earns a "stable" chip.

Validation

N-of-1 Trials

Association is not causation. When the mining engine surfaces a promising rule, I/O can spin up a randomized crossover trial with AB, ABA, or multiple-baseline design — randomized schedule, pre-registered hypothesis, effect-size tracking with confidence intervals, and a verdict computed at completion.

You are n = 1, but the methodology is the same standard used in precision medicine. No placebo effect hand-waving. No "I think it works." Just protocol and statistical verdict.

Architecture

On-device, always

Your health data never leaves your device. I/O stores everything in a local SQLite database and runs the mining engine in an isolated compute thread. No analytics, no telemetry, no cloud dependency for core features.

  • No analytics, no telemetry, no crash reporting
  • No third-party data brokers
  • Native connectors for HealthKit, Health Connect, Oura, Hevy, and more
  • Voice logging with on-device parsing for gaps in your stack
  • Premium subscription unlocks cloud-LLM parsing and literature RAG
Family

Built to work together

I/O is part of a tightly integrated family of instruments. Each product stands alone, but together they cover the full loop: track → model → mine → validate.