Counting cycles rather than hours
Nobody sets an alarm for seven and a half hours' time. The reason these tables land on odd numbers is that they are not counting hours at all.
Eight hours is a round number in base ten and nothing else. Ninety minutes divides into it five and a third times, so a bedtime that gives you whole cycles will almost never give you a whole number of hours, and the times on every page here are odd-looking for exactly that reason.
That is the trade the site is making. It counts in a unit that is meant to match how sleep is structured rather than in a unit that matches how clocks are labelled, and the price is a table full of quarter-past and quarter-to.
How much the assumption moves the answer
Five cycles before a six o'clock alarm, with a quarter of an hour to drop off, at three different cycle lengths:
| Cycle taken as | Bedtime | Time asleep |
|---|---|---|
| 80 minutes | 11:05 pm | 6h 40m |
| 90 minutes | 10:15 pm | 7h 30m |
| 100 minutes | 9:25 pm | 8h 20m |
Ten minutes on the assumption is fifty minutes on the bedtime, because the error is multiplied by however many cycles you are counting. That multiplication is the argument for showing the assumption rather than burying it: a site that quietly uses ninety-five is not a little different from one that uses ninety, it is nearly half an hour different by the fifth cycle.
Which is why it is a slider
There is no correct value to hard-code. There is a conventional one, a range it plausibly sits in, and a reader who may know more about their own than any website does. The slider is on every page that counts cycles, and what you set follows you between them.
The conventional value has a source, set out in the page on where the ninety minutes came from.