2 · Patterns
What tends to come before what — and how sure the app is.
For every food and exposure, and every symptom, Digestly asks: in the hours before this symptom, how often was that item logged, compared with the times it was logged and nothing followed?
It asks that question across several windows — two, six, twenty-four, forty-eight and seventy-two hours — because reactions are not always immediate, and a diary that only looks at the last meal misses the delayed ones entirely. Each answer is put through an exact statistical test, then corrected for the number of pairings tried, so that testing hundreds of combinations does not manufacture a result by chance. Finally the same diary is re-run with its days rotated, and anything that survives that shuffle is thrown out as a coincidence of routine.
What remains is labelled Recurring or Possible, with its counts, the window that matched, and a note of any medicine, illness or poor sleep that overlapped. On generated diaries with no real relationship at all, 3% produce any pattern and 1% produce one at the stronger label. Those figures are part of the app’s own test suite, and the engine cannot change without re-earning them.
Every card opens into its evidence — each entry that counted and each that did not — and a “why am I seeing this?” sheet explains what was compared, in plain language, including what the pattern does not mean.