Insights
Written from the day's own figures
Summaries, trends, a month at a glance, and correlations that are tested before they are shown. Where a relationship is weak, the page says weak.
01 · Daily summary
A paragraph made of the day's numbers
The summary states what you logged, against what limit, and how the day sat relative to the rest of the week. It names the items behind the observation.
Notes are grouped as could be better and well chosen, each attached to a specific food and a specific figure — not to you.
02 · Week and month
Compared against the window before it
Weekly and monthly summaries are the same arithmetic over a longer window, with the previous window shown as the delta. Days you did not log are counted as not logged, never as zero-calorie days.
That distinction is why "3 days logged" appears next to the averages instead of being folded into them.
03 · Correlations
Seven questions, asked the same way every time
Run seven tests on noise and one will look significant about a third of the time. Every threshold below exists to subtract that.
Tested over 41 paired days. The rank correlation is 0.11, below the 0.30 floor, and the permutation p is 0.34.
Shown, not hidden. "We looked and there was nothing there" is a result.
On the 22 days with a logged activity you ate 214 kcal more on average than on the 46 without. Hedges' g 0.62, p 0.0015.
An association between two things you logged. Not a cause.
18 paired days of the 30 this test needs. A gap in the log is treated as a gap, so two days either side of a missed day are not paired.
The card stays on screen. A missing card would read as a bug.
Some obvious comparisons are deliberately never asked. Logging completeness against over-limit rate is guaranteed by construction — fewer logged meals means lower recorded intake whether or not anything changed. Adding it would publish an artifact as an insight, and raise the bar on every other finding.
The seed is a constant, never the clock, so the same logged days produce the same p-value every time the tab is opened. A finding does not flicker between significant and absent because you looked twice.
04 · Month and year
A month, graded on three steps
Each day takes its colour from where it landed against that day's limit: under, slightly over, or over. Three steps, no gradient, no score.
Cheat days are outlined rather than recoloured — they were planned, so they are not a miss, but their calories are still counted.
Days with nothing logged stay blank. They are not drawn as zero-calorie days.
Supporting
Reminders
A time you choose, for meals you choose. No nudge for a missed day, no message about a streak.
Cheat days
Set which weekdays get a higher limit. Marked on the calendar as planned, counted in every total, and excluded from the comparison they would distort.