the workings · the road
Why we started simple
and only later reached for the complicated picture
When something as strange and stubborn as a long-COVID crash keeps happening to you, it is tempting to reach straight for a big, clever explanation. We did the opposite, on purpose: we began with the dullest, simplest questions we could ask, and only allowed a more complicated picture when the simple ones ran out of road. This note is about why that order matters, and what changes when you finally do reach for the complicated picture.
The razor: prefer the simple explanation
There is an old rule of thumb in science called Occam's razor: when two explanations both fit what you see, prefer the simpler one. If you hear hoofbeats, think horses, not zebras. This has nothing to do with simple things being prettier. Simple explanations carry three advantages that matter here.
- Easier to check You can actually test "my resting heart rate rises before a crash" in an afternoon. You cannot easily test "a twelve-part feedback network is quietly destabilising."
- Harder to fool yourself with A complicated enough story can be bent to fit almost any data, which means it isn't really explaining anything. A simple claim can fail, and a claim that can fail is worth something.
- More honest about what you actually know It is easy, especially with a mystery you badly want solved, to fall for a complicated story because it feels like it explains everything. Simplicity is the guard against that. It keeps you honest about the difference between a story that merely fits and a story that is actually true.
So the rule we set ourselves was this: climb only as high up the ladder of complexity as the data forces you to, and no higher.
The ladder we climbed
- 1
Just look
The first thing we did was not model anything. We took the daily numbers a smartwatch already gives you, average heart rate, average stress, sleep, body battery, and simply looked. Does the average go up before a bad patch? It is the cheapest, most transparent question there is, and if a plain daily average had predicted crashes, we would have been done.
It mostly didn't. The daily averages were quiet.
- 2
Test it properly
Where the looking hinted at something, we didn't just trust the hint. We tested it carefully: write the prediction down before peeking, set a fair bar in advance, ask whether it is more than chance, and check whether it still holds on data we hadn't looked at yet. This is still simple thinking, one clear claim, pass or fail, done with enough discipline that we couldn't quietly move the goalposts to make ourselves right.
Most of those tests came back empty too. But one thing survived, and it was telling. What set a crash apart wasn't the level of any number, it was how unsettled the numbers were. The one signal that held across the whole record was the jumpiness of my overnight stress line, not how high or low it sat. And a related thing showed up in the shape of a recovery: after a crash, how I felt came back in a few days, but the body's quieter, steadier night-time readings took weeks to settle. That second part is a description of the recovery, not a test that passed, and it is kept apart from the one that did.
- 3
Treat the body as a system
Only here, after the simpler pictures had run out of road, did a bigger possibility earn its place. Maybe the simpler pictures weren't just incomplete. Maybe they were the wrong shape. A daily average is a snapshot. But a crash isn't a snapshot event; it is something the whole body seems to fall into and then slowly climb back out of. That points, and "points" is as far as one body's data can honestly take it, toward something you read less as a tidy list of causes and effects and more as a system with states and tipping points.
This is a question the failures earned, not an answer they proved. It stays at the level of the next honest question.
We were pushed up this ladder by the failures, not pulled up by the appeal of a clever model. Occam's razor cuts both ways: don't add complexity you don't need, but don't refuse the complexity the data is clearly asking for either. The empty daily-average results are exactly what earned the right to reach for the complicated picture. We didn't start there because it was interesting. We ended up looking that way because nothing simpler would fit.
What the complicated picture asks of you
Most everyday statistics, the models behind "X is associated with Y," ask one kind of question: on average, which inputs move the output? That is a snapshot of fixed relationships, averaged across time. Thinking in terms of a system asks a different kind of question. It is less "which knobs move the dial" and more "understand the landscape the ball is rolling around in." In practice it asks you to hold several things at once that an ordinary average lets you ignore.
-
History matters
Where the system goes next depends on where it already is, not just on today's inputs. You can no more understand a crash from a single day than understand a wave from a single photograph.
-
Causes loop
In a feedback system, A affects B which affects A. There is no tidy arrow from cause to effect, there is a circle. A thermostat, a spiral of worry, a traffic jam: none has a single cause.
-
Small pushes can have huge effects, but only sometimes
Below a threshold a stressor does almost nothing; one step past it, the whole system tips. Water doesn't cool smoothly into ice, it holds, holds, then flips all at once. A crash looks more like that flip than like a dial slowly creeping down.
-
The same body can be in different states
Not just "more or less sick on a scale," but qualitatively different regimes, steady, wound-up, crashed, in which the very same action lands completely differently.
-
You are inside the system you're studying
Because I pace, resting when the numbers look bad, I change the very thing I am trying to measure. It is like studying a traffic jam while being one of the drivers braking because of it.
-
The rules themselves drift
Over years of illness and recovery, what triggers a crash, and how one even looks, slowly changes. There isn't one fixed law waiting to be found, there is a moving one.
And there is a humbling shift baked into all of this: the goal quietly changes from prediction to understanding. An association model dreams of "tell me tomorrow's crash." The systems view often concludes that day-ahead prediction is, for deep reasons, not really possible here, and aims instead at something more modest, and truer: what state am I in, and is my footing quietly draining? Not a crystal ball.
A weather report, not an alarm.
Why the discipline is the point
None of this makes the complicated picture "the answer." It makes it the honestly-earned next question. We climbed the ladder one rung at a time; we made ourselves show that each simpler rung had failed before taking the next; and we keep saying out loud where even the systems view can't reach. It describes one body, not everybody. It can't cleanly separate cause from the awkward fact that I rest whenever I feel bad. And it turns a crash into something to understand rather than something to reliably foresee.
Starting simple isn't timidity. It is the thing that lets you trust yourself when you finally do say something complicated.