For the last two and a half years I've kept a record of everything sweet I ate. I attempted to be accurate, but sometimes I have just had to give up and vibe it, a day of conference snacks will defeat any data collector.
I had Claude generate graphs and extrapolations from my data. The results are below, with some lovely editorial work from yours truly.
- Period covered
- 19 Feb 2024 – 22 Aug 2026
- Recorded days
- 774
84% of the span - Days with chocolate
- 575
74% - Total mass of sweets
- 79.4 kg
25.5 kg chocolate - Estimated energy
- 355,112 kcal
459/day - Estimated sugar
- 32.0 kg
41 g/day - Median day
- 90 g
mean 103 g - Longest choc streak
- 39 days
longest gap 8
01 / The habit
How often each thing showed up
Chocolate appears on 74% of recorded days and biscuits on 64%. Nothing else clears 30%. On 94 days chocolate was the only thing recorded, and chocolate alone supplies 38% of the calories in the whole record — with biscuits, two-thirds.
Every recorded day
One row per year, one stripe per day, brighter for heavier. The pale block through spring 2024 is the opening period, when the diary was newest and the entries largest.
The long dark stretches are not gaps in the record. They are deliberate: late May to mid June 2025, and the three weeks from 22 November to 11 December 2025, are periods of eating no sugar at all. Shorter dark patches of a day or two are more likely to be days I forgot to write anything down.
The whole record, day by day
Every day as a grey spike, with a 30-day mean over the top. The shaded columns are the no-sugar periods. The opening three months average well over 150 g/day; from summer 2024 onwards the mean settles into a band between roughly 75 and 115 g and stays there for two years. Overall mean 103 g/day.
A typical day
Right-skewed, as intake data tends to be. The median day is 90 g and the mean 103 g; the gap is a thin tail of heavy days pulling the average up.
Five days clear 400 g, and all five fall in 2024: 26 April (480 g), 25 April (465 g), 14 June (444 g), 2 March (425 g) and 14 April (420 g). The two April days are back to back.
02 / Two results
Neither was obvious before the data was assembled.
A heavy day predicts a heavier one
The obvious hypothesis is compensation: overdo it, then ease off. The data says the opposite. After a day over 100 g, the next day averages 119 g — above the overall mean of 103 g. After a day under 50 g, it averages 91 g. Day-to-day correlation is +0.32. Bars show standard error.
It is also not permanent. Split by period, the effect belongs almost entirely to the years of the PhD, and has since gone — see below.
Only the staples are sticky
The same correlation, per category. The pale marks are the baseline: the correlation you get from the same days shuffled into random order, which for every category sits near 0.07. Anything at or below that mark is indistinguishable from noise.
Ice cream (0.08) and event snacks (0.09) are at the baseline. Chocolate (0.38) and biscuits (0.35) are five times it. Habits carry from one day to the next; occasions do not, which is what makes them occasions.
03 / Seasons
Monthly averages from June 2024 onward. The opening period was heavy enough to invent a spring peak that later years do not show, so it is excluded.
Chocolate and ice cream have real seasons
These two have stories that can be told. Chocolate runs near 30 g/day for most of the year and then hits 52 in December — going home for Christmas, plus the run-up. April is the second peak at 32, and the drop to 18 in May is the sharpest month-on-month fall anywhere in the series: Easter eggs, and then no more Easter eggs.
Ice cream is the cleaner signal and the more obvious cause. It peaks at 14 g/day in August against 1 g in January, roughly fourteen-fold.
Bars show the standard error of each monthly mean. They cover sampling noise only — the correlated 15 to 20% uncertainty on absolute values described below sits on top, and would move every bar in a panel together rather than independently. December and August clear their neighbours comfortably; most of the six smaller panels do not clear their own error bars at all.
Heavy days are a different kind of day
Categories present on an average day in each band. A day under 50 g is chocolate and biscuits and very little else. What appears as days get heavier is cakes, desserts and pastries — things that arrive in fixed portions, at events, and are not really a decision about quantity. Heavy days look like occasions rather than larger helpings.
04 / Jobs
I changed jobs in 2025: unemployed from February, PhD defence in May, then a new country and a new job on 1 June. I changed again in July 2026. Splitting the record at those dates changes what the earlier findings mean.
Three things moved when the job did
Bars are bootstrap 95% intervals. On the left, how much: 117 g/day through the PhD, dropping to about 90 when I stopped working and staying there. The February 2025 step is real — the intervals barely overlap.
In the middle, the weekend. During the PhD, weekends ran 17 g above weekdays, which survives a permutation test that shuffles day labels within each week (p = 0.037). Afterwards it is nothing. Testing four periods weakens that considerably; call it suggestive rather than settled.
On the right, the momentum from the previous section. It was +0.40 during the PhD, with an interval clear of the noise floor. In the new job it is +0.08, an interval straddling it. The habit had inertia, and moving country broke it.
The mix moved too. Chocolate falls from 37% of mass under the PhD to 28%, then 21%, while desserts more than double to 13.6%. Sticky toffee pudding and Gü pots appear where lava cake and mochi used to be, which is roughly what emigrating to Britain should look like.
05 / What it adds up to
355,112 kcal and 32 kg of sugar
Calories accumulating on the left, total sugar by category on the right. Chocolate contributes 135,289 kcal and 14.0 kg of sugar on its own — more than the other seven categories combined in both cases. Biscuits add another 100,839 kcal.
Across 774 recorded days that is 459 kcal and about 41 g of sugar a day. Both figures are estimates built on category averages, and the note below explains how much to trust them.
06 / Method
What these numbers are worth
Per-day error bars are sound. Aggregate ones are not. Combining them in
quadrature across 774 days implies ±0.7% on the total, which assumes the
errors are independent and cancel. They are not: if the assumed 15 g biscuit is
wrong, it is wrong on all 497 biscuit days. Treat the headline totals as
±15–20%.
Sugar and calories are softer still, because one category density stands in for every item inside it. The sugar figure counts total sugars, including the milk sugars in chocolate and ice cream, so it is not comparable to the free-sugar guidelines usually quoted.
Every average here is taken over recorded days. The no-sugar periods are genuine zeros rather than absences, and counting the 45 days in the five long abstentions would pull the daily mean from 103 g down to about 97 g.
How a vibe became a number
Bare entries were given standard portions: ice cream 60 g, cake slice 90 g, biscuit 15 g, doughnut 65 g, Gü pot 85 g, Percy Pig 5.5 g, KitKat 41.5 g. Travel and event entries were scored at 75 g, Christmas and Easter at 100 g. Explicit weights carry ±7%; bare named items ±20–30%; vague entries ±35–45%.
Nineteen date headers were repaired. Nine were out of sequence in a strictly
reverse-chronological document and were fixed against their neighbours; ten
more were partial or malformed and would otherwise have been absorbed into the
previous day. Three entries that looked like counts were in fact grams
— 47.4 choc is 47.4 grams, not 47 bars. One header,
27-29, was a date range and became three days.