The first half: line up the right pair
The delay is the whole problem, and it is easy to fix.
Post-exertional malaise, or PEM, is what follows when the day went past the body's available capacity. The point of this is not that you did too much; it is that the ceiling is real, it moves, and it is usually invisible until you are already past it — which is exactly why knowing where it sits is worth the trouble. The crash does not usually arrive the same day. Twenty-four to forty-eight hours is the range people describe most often, and longer is common. So a table with one row per day, activity in one column and symptoms in the next, is asking whether Wednesday's walk made Wednesday worse. It usually did not. Monday's did, and the table never asked.
The fix is mechanical. Take your activity column and move it down one row, so each day's symptoms sit beside the previous day's activity. Look at the relationship. Then move it down two rows and look again. Then three. In a spreadsheet this is one formula and four copies of it. Each shift is called a lag, and you are testing lag 1, lag 2 and lag 3 separately, because a pattern at two days is invisible at zero.
Do this with ranks rather than raw numbers if you can — sort each column and compare positions instead of values. One catastrophic day with a symptom score of 10 can drag a raw correlation on its own; ranks stop a single extreme day from inventing a pattern.
The second half: the pattern that vanishes next week
This is the part almost nobody does, and it is the part that matters.
Once you are checking four or five columns at four delays, you are running twenty comparisons at once. Some of them will look striking for no reason at all. That is not a flaw in your data; it is what happens whenever you check a lot of things. It is why people describe seeing a clear pattern one week and finding it gone the next, and it is why a lot of tracking ends in the drawer.
The way to tell the difference is to ask what coincidence alone would have produced with your own numbers. Take your symptom column and shuffle it, then measure the relationship again. Do that a couple of thousand times. If your real arrangement is not clearly stronger than the shuffled ones, what you saw was chance, and the honest conclusion is that nothing stands out.
One detail matters for this illness in particular. Do not shuffle day by day. Chronic illness runs in good weeks and bad weeks, so neighbouring days resemble each other, and shuffling single days destroys that structure and makes almost everything look significant. Shuffle in blocks of consecutive days instead, so the runs survive the shuffle. Then correct for how many comparisons you ran — if you tested twenty things, the bar for any one of them has to be higher.
What the trackers do and do not do
Showing a correlation and testing one are different jobs.
Some trackers can look at the days after a factor — Bearable's factor report goes out to seven days — so it is not true that nothing looks back. What is generally missing is the second half. Correlation grids tend to display a strength without saying whether it clears chance, without correcting for how many pairings were checked, and sometimes without showing the number at all. Users say so in their own reviews: that the conclusions can be misleading, and that a single overlooked entry can produce a correlation that makes no sense.
None of that makes those apps bad at what they are for. Logging every day when you are ill is the hard part, and they are good at it. It does mean that the answer to "is this real?" usually is not in the app, and that exporting your own CSV and testing it yourself is a reasonable thing to want to do.
What this cannot tell you
The honest limits, because they change what the answer is worth.
Two things moving together is not one causing the other. A delayed association between hours upright and a worse day is a reason to look closer, not proof of a cause, and something you never wrote down may be behind both. This narrows the list. It does not close it.
You also need enough days. Around twenty is the floor for any of this to mean anything, and it works properly from about three months. Below that the honest answer is that there is not enough yet — which is a real answer, not a failure.
And a result that says nothing stands out is worth having. It is not a wasted month. It means the thing you suspected is not visible at this size in this data, which is information you can act on.
If you would rather not do it by hand
There is a free page that does exactly this.
Everything above can be done in a spreadsheet, and if you want to, you should — it is your data and the method is not a secret. If you would rather not, this free page runs it on a CSV you already keep. It checks lags of zero to three days, compares ranks, shuffles your own days in blocks at least two thousand times, corrects for every pairing it checked, and says plainly when nothing clears the bar. It reads a Bearable export directly. There is no account, no email address and no payment, and your file is read by your own browser and never uploaded — there is no server to send it to.
It does not diagnose and it makes no claim about any condition. It reports associations in numbers you already recorded.
Something wrong here? Tell me: [email protected]