The nap planner scores naps based on the following:
- How much we expect your next sleep to be delayed after the nap.
- How much grogginess we expect you to feel after waking up.
- How difficult we expect it to be to wake up.
with more dimensions to come.
How do we estimate these things?
We run hundreds of simulations of how your body’s clock is shaped by your activity patterns in the normal course of running Arcashift. The nap scores are natural extensions of these simulations, taking on the “what if?” task of trying out each of the naps (virtually) and seeing what the effects are.
Your body’s transition from sleep to wake and back comes from the combination of two processes. These have to line up for your body to be able to sleep, and wake up feeling refreshed.
- Your “sleep hunger” which rises when you’re awake, and drops when you sleep
- Your body’s time-of-day, aka “phase” or circadian state.
When you take a nap, it satisfies some of that sleep hunger. That’s part of the goal for a nap, after all! But it could mean that you have less hunger later, when you’d like to settle in for a longer sleep.
Sleep delay
We decide how much we expect your next sleep to be delayed by simulating each nap in the time window you give us, and seeing how much longer it takes you to fall asleep the next time after that nap. Those with little-to-no effect (30 minutes or less delay) are pushed to the top and offered, assuming they aren’t terrible in some other way.
Difficulty waking
This is again related to the sleep hunger + phase causing sleep delays, but sort of the reverse: causing awakening delays. At first glance, this might seem to be the same, or really close to, grogginess. After all, if your body isn’t ready to wake up when your alarm goes off, one expects mental fog to be part of the effects. But unlike the grogginess score (see below), this score takes into account the combined forces of your sleep hunger and your circadian state on awakening.
Grogginess
People tend to be more groggy when they wake up near CBTmin. Over the summer, we built a machine learning model paradigm that reflects this, and can be tailored to an individual user quickly based on their responses to a simple 1-10 rating of how sleepy they are over the course of the day. Sleep inertia (grogginess) is a feature we’re working on personalizing even more to the individual, since not everyone feels equally groggy in the morning. One beta tester recorded their grogginess on this scale first 8-ish hours after waking for several weeks, and applied our approach. We found a clinically significant (p=0.025) trend between the total grogginess and that user’s simulated body time when waking up, as compared to CBTmin phase. While this level of prediction is still being refined, at time of release we can predict high/medium/low grogginess levels for a general user.
