The Methodology

Why N=1?

Population studies are built to find average effects across groups. The signal they surface is real — but it describes a statistical composite that may not resemble you. Average protein recommendations were derived from people who are not you. Average sleep targets were measured on people with different chronotypes, genetics, and stress loads.

The N=1 approach accepts this limitation and turns it into a feature. One subject means every measurement is directly relevant. One subject means no between-person variance to average out. One subject means you can run interventions and observe effects on a body you actually have to live in.

The trade-off is external validity — you cannot generalize my results to you. That’s honest, and it’s fine. The value is the framework, not the conclusions.

Correlation Engine

The core analytical layer is a rolling correlation engine that continuously calculates Pearson r coefficients across metric pairs. Rather than reporting static relationships, every correlation is recalculated over a 90-day sliding window — capturing how relationships evolve as the body and behaviors change.

Running 23 metric pairs simultaneously creates a multiple comparisons problem: the more tests you run, the more likely one produces a spurious positive by chance alone. The engine applies Benjamini-Hochberg false discovery rate correction to every batch — a method that controls the expected proportion of false positives among all reported discoveries, rather than protecting against any single false positive as Bonferroni does. It is less conservative and more statistically appropriate for exploratory research.

Methodology in Action

Here’s how one finding moved through the pipeline — from raw signal to validated insight.

// This is N=1 data. The threshold (7.5h) and effect size (77% vs 25%) are specific to this physiology. Your numbers will differ. The process is what’s portable.

What We Track

26 data sources. Each ingested daily by a dedicated AWS Lambda function on a fixed UTC cron schedule. Raw JSON archived to S3 permanently; normalized metrics written to DynamoDB single-table with gap-aware backfill.

Honest Limitations

Every framework has edges. These are the places where this one is weakest, stated plainly rather than footnoted away.

AI Advisory System

Every architecture decision, content strategy call, and health protocol change is reviewed by a system of 34 AI-generated personas organized into three boards:

When personas disagree, the throughline tiebreaker applies: “Does this help a visitor connect the story from any page to any other page?” If yes, it ships. If not, it waits.

See the full advisory roster →

Evidence Badge System

Every claim on this site carries a badge indicating how much data backs it. The thresholds follow the Henning Brandt standard: with N=1 research, observation count determines confidence.

Badges appear on pull-quotes, discovery cards, and protocol findings across the site. They’re a constant reminder that all findings on this platform are from a single person’s data.

The Value Isn’t in the Conclusions

If you’re reading this hoping to find a protocol you can copy, you will be disappointed — or at least you should be. My sleep responses are not yours. My glucose is not yours. My correlation between protein intake and HRV tells you almost nothing about your own body.

But the framework is portable. The question “what does my data actually show about my sleep?” is one you can ask about yourself. The practice of tracking something for 90 days before drawing conclusions, running FDR correction to avoid over-interpreting noise, requiring a minimum observation count before reporting — these habits of statistical hygiene cost nothing and apply universally.