How the Numbers Feel

Four years, four chapters.

It's tempting to treat the record as one long, even stretch of data. It isn't. The years fall into four chapters, and (crucially) the chapter breaks were set by the illness, not chosen by us to make a story work.

The four chapters

  1. I

    Before: a healthy baseline

    Aug 2021 – Mar 2022

    Watch data from a healthy body, no illness yet. About seven months, a single winter, so season and health are tangled here by construction, and any comparison to it has to say so. Still, it's the closest thing to a 'normal' reference the record has.

  2. II

    The infection

    Mar – Apr 2022

    A two-week acute window, labelled from the lived event. Too short to analyse on its own, but real, and the hinge between the two bodies the record holds.

  3. III

    Long COVID begins

    Apr – Sep 2022

    The body is now ill, but the daily felt-state log hasn't started yet. Watch data only, so no crashes can be labelled here, and the precursor tests can't run.

  4. IV

    The four years that carry the study

    Sep 2022 – now

    The felt-state score, the crash labels, and the watch data finally coexist. Almost every test on this site lives in this chapter, because only here do all three exist at once.

The clearest picture of how different those states are isn't a lab number; it's what I simply stopped being able to do. The recorded training tells it plainly:

What I stopped being able to do Recorded training, before Long COVID and after: the fit-then-collapsed calendar
Runs (per era)
  • Before 82
  • After · 2022–23 4
  • After · 2024+ 0
Kilometres per week
  • Before 55 km
  • After · 2022–23 7.5 km
  • After · 2024+ 0.7 km
Days active
  • Before 45.6%
  • After · 2022–23 33.4%
  • After · 2024+ 4.4%
Structured endurance training all but vanished after Long COVID onset: 82 runs to 4 to none, 55 km a week to well under one. (The middle era's session count stays up only because 231 walks replaced running, a gentler activity, in step with pacing.) This is the recorded-activity shape, not a claim about capacity; what it did to the body is the deconditioning half of the driver ledger. Two honest limits: the recording starts in 2021, so “before” is the last ~7 months of a longer training history, a floor, not the peak, and casual walks are logged unevenly, so the walk counts are a lower bound.

Why the chapters are hard lines

You cannot pool a healthy body, an acute infection, and a chronic illness into one average; they're different states, and a number mixed across them means nothing. So the boundaries between chapters are treated as walls, not knobs: they sit on clinical events, not on dates we tuned to flatter a result.

Inside the last chapter: a recovery told in phases

The last chapter (the four years that carry the study) isn't flat either. I lived it in stages: the early months when the acute symptoms hardened into chronic patterns, before I knew to pace; then learning to pace, through ergotherapy; then the long stretch once pacing was a habit; and finally the time on medication. Counting the healthy baseline and the infection, that's a six-phase recovery, and the watch turns out to see most of those boundaries too, without ever being shown them. See the recovery, in six phases.

Crucially, those phase boundaries come from lived experience, each anchored to a real event in my own recovery, not from an algorithm hunting the data for a convenient place to cut. That's exactly what makes them a trustworthy lens: a boundary tied to a remembered event can't be quietly slid around to flatter a result, the way a data-drawn one can.

The lines that come from life (the chapters, the recovery phases) we trust precisely because we didn't choose them to fit; a boundary tied to a remembered event can't be slid around to flatter a result. What else moved across these years is in the driver ledger.

The full segmentation, in the research repo: lc_era_temporal_segmentation.md ↗

Open full page ↗