← What I found · the finding
Too much, in the days before
Not found Heavy exertion in the 4-day lead-up
What the guide says There is a personal activity threshold above which crash probability rises sharply; respect your lag (test exertion a few days before the crash, not the same day).
Catches ~8 of 10 crash lead-ups but also fires on ~3 of every 4 ordinary days. When it fires, a crash follows roughly 1 time in 44.
- Catches
- 82%
- of crash lead-ups it fired on (sensitivity)
- Stays quiet
- 23%
- of ordinary days it correctly stayed quiet (specificity)
- When it fires
- 2.3%
- of the days it fires are followed by a crash (precision)
- Lift
- 1.06×
- × more likely than knowing nothing
These describe how the signal behaves in daily life, framed against how rare crashes are, not whether the pattern is real. That verdict is the badge above; a usable-looking number doesn't make a signal that didn't hold real. Tier C (worth watching at best), and with only 29 crashes the margins are wide.
What showed up in the data
Heavy exertion in the four days before a crash first looked like the clearest recent tell anywhere in the data, a headline result. Then it turned out to be an artifact of how the baseline was computed: the 'normal' it compared against quietly included the very days being tested. Recomputed on a clean baseline that excludes them, the signal vanished, on both ways of drawing the comparison.
How we tested it
The original used a rolling baseline; the correction re-ran the identical test on a lagged baseline (a window that stops 30 days before each candidate day, so the comparison can't see the days under test). Crucially, the re-run was bundled symmetrically, the win was re-tested on the corrected baseline alongside everything else, not quietly kept while only the losers were re-checked.
What it means, and what it doesn't
- What it rules out
Once the measurement error is removed, heavy exertion in the days before a crash is not a precursor for my body, on either reference frame.
- What it does NOT mean
It doesn't mean activity is irrelevant to pacing in general, or that the 'respect your lag' idea is wrong. It means this particular signal, measured cleanly, doesn't separate crash-weeks from ordinary ones here.
- Why it's on the board at all
Because the honest thing to show isn't only what survived, but what didn't, and why. A baseline that includes the days you're testing is a common, easy mistake; catching it before it hardened into a headline is the system working as intended.
Go deeper
The short version is above. These go to the method and the raw research.