The peak-end rule: why your quietest good months vanish
In 1993, four researchers asked people to hold a hand in cold water twice and then choose which trial to repeat.
The short trial was 60 seconds at 14°C. The long trial was the same 60 seconds, followed by 30 more seconds during which the temperature was allowed to drift up to 15°C. Slightly less unpleasant at the end, but half a minute longer, and strictly more total discomfort.
A significant majority chose to repeat the long trial. Given a free choice, people preferred more pain, because it ended better.
The paper, by Daniel Kahneman, Barbara Fredrickson, Charles Schreiber and Donald Redelmeier, is called “When More Pain Is Preferred to Less: Adding a Better End”, published in Psychological Science. Its conclusion was that retrospective evaluations of an episode are dominated by two moments: the worst one and the last one.
It holds outside the lab
Three years later, Redelmeier and Kahneman ran the same question through actual medical procedures. In a 1996 paper in Pain, they recorded moment-by-moment pain intensity during colonoscopy (154 patients) and lithotripsy (133 patients), then asked patients afterwards to rate the total pain of the experience.
Those retrospective ratings correlated strongly with the peak intensity and with the intensity during the final three minutes. Duration barely registered. Procedures that took substantially longer were not remembered as substantially worse.
That is the peak-end rule, and the second half of it has a name: duration neglect. When you evaluate a stretch of experience after the fact, the amount of time it occupied is nearly absent from the calculation.
What this does to a day
Apply it to something ordinary, like a Wednesday.
Nine hours, mostly fine. Two hours of decent work in the morning, a meeting that went well, lunch outside, an afternoon that moved. Then at 5:40pm an email arrives that is going to ruin your Thursday, and you read it three times, and you leave.
Ask about Wednesday on Friday. It was a bad day. Not “it was seven-eighths fine and then ended badly.” Just bad. The peak (the email) and the end (the email) are the two inputs, and they happen to be the same event, which gives that one event nearly total control over the summary.
Nothing about that is a mistake you made. Evaluating an episode by its extreme and its ending is fast, and for deciding whether to repeat something it is often good enough. It is simply not a description of the day.
The part that costs you more
The same mechanism runs in the other direction, and this is the version nobody talks about.
Consider a genuinely good month. Not a dramatic one. No trip, no promotion, no crisis resolved. Just a month where work was interesting enough, you slept properly, you saw people you like, the weather was reasonable, and nothing went wrong.
Peak-end evaluation asks: what was the most intense moment, and how did it end? A calm month has no peak. Its ending is another ordinary Tuesday. Duration neglect ensures that thirty consecutive good days count for almost nothing on their own, because duration is not an input.
So the month leaves nearly no trace. Asked about it a year later you have no assessment at all, or you fall back on whatever you can reconstruct from surrounding context, which is usually not much. A single dramatic week beats it in memory easily, even a bad one.
The recall system is optimised for extremes. Ordinary contentment produces no extremes, so it does not get recorded, so it does not get counted when you form a view of how your life is going. Which is a fairly serious accounting error to be making about your own experience.
Why writing at the end of the day does not fix it
The obvious response is to write a diary entry each evening. This does not solve the problem, and it is worth understanding why not.
A retrospective entry written at 11pm is produced by exactly the system under discussion. You sit down and ask yourself what the day was like, and what comes back is the peak and the ending, so that is what gets written. The entry is a faithful record of your evaluation rather than a record of the day.
Which is also why a month of calm produces a month of very short entries, or none. There was nothing to report, according to the faculty deciding what counts as something to report.
You can partly beat this with structure. Rate the day hour by hour, keep a running log during the day rather than a summary after it, use interstitial notes between tasks. All of these work, and all of them cost attention during the day, which is why they get abandoned.
What a record does differently
A factual record has no peak and no ending. It has no evaluation step at all, which is precisely why it can act as a check on one.
An entry assembled from what your tools logged represents every hour on the same terms. The two good hours of morning work appear at the same weight as the 5:40pm email, because neither one was scored. The four decent days in the middle of the calm month are as present as the interesting one, since nothing decided that interesting days deserve more lines.
This is not a better summary. It is deliberately not a summary. It is the material a summary would have been made from, kept so that the summary can be checked.
The practical effect shows up in re-reading. Open that Wednesday six months later and you find eight hours of an ordinary decent day with one bad email at the end of it. Your memory said bad day. The record says something more accurate and less flattering to your memory, and you cannot get to it from the inside, because from the inside the evaluation is all there is.
What the record cannot do
It cannot tell you how anything felt. The entry knows the email arrived. It does not know you read it three times.
It also cannot restore the calm month’s emotional texture. What it gives you back is the content: who you saw, what you worked on, what you were listening to, where you went. The contentment itself was never written down by anything, and a factual record does not conjure it. What it does is stop the month from being blank, which is the difference between having a chance at recovering it and having none.
And peak-end evaluation is not a defect to be corrected. It will keep running. Memory also softens negative feelings faster than positive ones over time, so your reading of a period keeps moving after the fact regardless of what you do. The point is not to replace the evaluation. It is to have a second source that was never asked for an opinion.
The month you would not have thought to keep
The peak-end rule predicts something uncomfortable: the periods most at risk of disappearing are the good, unremarkable ones. No peak to anchor them, no ending to weight them, and duration, the only thing they had going for them, not counted.
deariary generates an entry for each day out of what your connected services logged, whether or not the day contained anything your evaluation system would classify as an event. Which is the whole point. The days that produce no summary are exactly the ones with nothing else keeping them.
