The 90-minute work cycle: what the evidence says and how to find your own
The idea that people work best in cycles of about 90 minutes goes back to Nathaniel Kleitman's basic rest-activity cycle. It is well supported for sleep and much thinner for an ordinary workday, so treat 90 minutes as a starting guess, not a rule. A better answer comes from your own records. Look at how long your uninterrupted stretches actually run and how you break them, then test one session length for two weeks. Hina 0.6.0 keeps the data for that on your Mac, and this post shows which numbers to read and what they cannot tell you.
Where the 90 minutes comes from
Nathaniel Kleitman, a sleep researcher, proposed a basic rest-activity cycle, usually shortened to BRAC, and returned to it in a 1982 paper titled Basic rest-activity cycle, 22 years later. The proposal is that the roughly 90-minute alternation of REM and non-REM sleep continues through the waking day as a rhythm of higher and lower alertness, with a reported range of 80 to 120 minutes.
Ultradian is the general word for any biological rhythm shorter than a day, including hormone pulses and sleep stages. The 90-minute work cycle is one specific claim inside that family, and it is the one that books and articles on productivity tend to repeat.
How strong the evidence is for work
For sleep, the cycle is easy to observe, because sleep stages repeat on about that schedule. For waking hours the picture is weaker. In 1977 Kripke and colleagues published a paper called There's no basic rest-activity cycle, arguing there was no evidence for one outside sleep. A 1992 review by Lavie, titled Ultradian cycles in sleep propensity: or, Kleitman's BRAC revisited, did report daytime cycles in how easily people fall asleep, but suggested they may run on different mechanisms than the REM and non-REM cycle at night.
The work version usually adds a second source, often credited to Tony Schwartz: that the best violinists in Anders Ericsson's 1993 study practised in sessions of no more than 90 minutes. The paper says something narrower. During the diary week, practice sessions averaged 80 minutes with no reliable difference between groups, the music teachers included, and the best players practised more because they had more sessions, about 19.5 a week against 7.1 for the teachers. That is an average session length, not a ceiling that the best players kept to.
That leaves a plausible idea with little direct support for the claim most people care about, that a writer or programmer should stop at minute 90. A person who works well in 50-minute blocks is not wrong, and neither is one who needs two hours. Hina's own break reminder uses 90 minutes of continuous presence, and the code comment that sets it says the number was chosen so the card shows up more rarely than Hina's idle-review question, not copied from health literature.
What Hina records that bears on session length
Hina does not measure alertness. It records which app was in front, for how long, and counts of key presses and clicks, plus gaps when you were away. From that it builds deep work, defined in the code as a stretch of work-category activity of at least 10 minutes. A detour shorter than 5 minutes into something that is not work does not end the stretch and is counted as one interruption, and a gap of up to 15 minutes with almost no keyboard or mouse input, such as waiting for an AI to finish or a short walk, is bridged.
Breaks are counted from the same data. Inside your working span, an away stretch of at least 5 minutes and under 6 hours counts as a break, and when there were at least two, the daily report's Rhythm of the day section lists how many there were, their total, the median, and the longest one if it ran 30 minutes or more. Shorter gaps are added up separately as gaps shorter than 5 minutes each. These are rules about the keyboard and the front window, so reading on paper, thinking at a whiteboard or a meeting away from the Mac will break a stretch once it runs past those tolerances, not extend it.
Read your own session length
Open the daily report for a day and look at the Attention section. It names when the longest stretch of deep work began, how long it lasted, and the app or project it was in. Hina can join work activity across short gaps, so this block is not necessarily uninterrupted. On the Week tab, the Golden hours card names the start hour that holds the most deep work in the week, and the Interruptions card shows what you switched to during work.
On the Month tab, Median deep work gives the typical daily total, Deep work calendar shows which days were heavy, and Deep work by weekday shows whether Tuesdays differ from Fridays. Taken together, these answer a more useful question than whether you fit the 90-minute pattern. They show how long you actually stay in one thing, at what time of day, and what ends it. Daily reports, Week and Month need a licence after the 7-day trial, while recording continues without one.
A two-week check
For the first week, change nothing. At the end of each working day, write down the start and length of the longest deep stretch from the daily report, along with what ended it. After five days you will have five numbers, and you can see whether they cluster. If they sit well below 90 minutes, a 90-minute plan would be fighting your own pattern.
For the second week, set one session length near the middle of those numbers and work in blocks of that size with a short break between them. At the end, compare the week with the first: Median deep work on the Month tab, the number and length of breaks in Rhythm of the day, and your own sense of how the afternoons went. One fortnight on one person is a weak experiment, so treat the result as a preference you adopted, not a finding.
Break reminders, and what they do not do
Hina's Sit-too-long reminder is on by default, in Settings, General. It watches for continuous presence, meaning no 3-minute gap in input, and after about 90 minutes a small card appears at the top-right corner of the screen. It shows at most three times a day, and Later hides it for 25 minutes while Take a break starts a timer of 20 seconds or two minutes depending on the prompt, and between 22:00 and 06:00 it becomes a bedtime reminder with only a Got it button.
There is no interval slider, so it cannot be set to your own session length. If you settle on 50 minutes, use a timer or calendar for that and keep the card as a backstop, or switch it off. It stays quiet while you are paused, away, locked, full screen or on a call, and it does not log whether you followed it. The separate article on gentle break reminders covers the rules in full.
What this cannot tell you
A long deep stretch is not a good one. It means you stayed in work apps with input for that long, and it could be a long stretch of rewriting the same paragraph. The records also start the day Hina starts recording, and a day where it was paused shows a gap, not a short session.
If your work is mostly meetings, reading on paper or physical, the numbers will mostly show the gaps between them. Use them for the part of your day that happens at the Mac, and keep the decision about how long to work in your own hands.