The seven limits.
Every finding on this site inherits these. They aren't fine print at the bottom of a page; they're the shape of what a four-year study of one body can, and cannot, tell you. Read the findings through them.
- 1
It's one person
Everything here is confirmed for exactly one body: mine. It may not be true for yours, and a single person can never tell us which findings would generalise and which are personal.
- 2
Everything changed at once
Across four years the virus, the medication, the seasons, the pacing, and the body itself all shifted together. Pulling apart which one moved a signal is often impossible with a sample of one. See the driver ledger.
- 3
The watch has blind spots
A 2019 Forerunner records no heart-rate variability at all, and some signals only exist in the recent years. Where a measurement was missing we used a proxy or went without, and said so.
- 4
The analyst is the subject
The person running the analysis is the person living it. That coupling invites bias; pre-registration and fresh-session audits hold it back, but they cannot remove it entirely. See how a question gets locked.
- 5
Notes only count what was written
A symptom written down on a given day is real evidence it happened. A symptom not written down is not evidence it didn't. The note record is presence-only, so absence can never be read as proof.
- 6
The score is a feeling
The daily felt-state is an honest self-report, not a lab value. It's the ground truth the whole site rests on, and it is unavoidably subjective.
- 7
We only see the crashes that happened
We can't study the crashes that were avoided, including any that good pacing prevented. That quietly biases every estimate of how well a signal 'predicts': the misses we'd most want are invisible. See pacing, in the driver ledger.
None of this is an apology. A study can be small and honest at the same time, and naming its limits plainly is what lets you trust the parts that hold.
The full limitations register, in the research repo: research_line_limitations.md ↗