Every sleep and recovery metric you're used to seeing shares the same fundamental limitation: it's retrospective. Your Apple Watch tells you this morning's HRV, last night's sleep stages, today's recovery score — all of it a report on a decision you've already made and can no longer change. Useful for understanding the past, useless for the one question that actually matters at 10:47pm: should I go to bed now, or does another hour not really cost me much?
Answering that question properly requires something categorically different from a score — a forecast. Not "how did I sleep," but "given everything I know about how you personally sleep, what would tonight's outcome look like under different bedtimes?" That's what a sleep forecast, or in Metrya's case the Tonight Simulator, actually does.
Why This Is Harder Than It Sounds
The naive version of a bedtime forecast is a straight-line rule: more sleep hours, higher recovery, done. That's not how the real scoring works, and a forecast that ignores the nuance would just be wrong. A trustworthy forecast has to run the actual recovery-scoring logic — the same algorithm that scores your real mornings — against a plausible synthetic night, not a separate, simplified guess.
That means building a realistic synthetic night first: how much of it would likely be deep sleep versus REM, how efficiently you'd actually fall and stay asleep given a specific bedtime and wake time, and how anything already in your system — caffeine, most obviously — would degrade that night's quality. Only once that synthetic night exists does it make sense to feed it through the real scorer and see what comes out.
What Personalizes the Forecast
A generic sleep calculator assumes everyone's a textbook sleeper. A forecast that's actually useful for you specifically has to be built from your own recent nights — not population averages:
The bedtime itself is the one variable you control directly — a slider spanning roughly an hour before to two hours after your personal-median bedtime, snapped to 15-minute steps. Move it, and every downstream number recomputes: the synthetic night's duration and quality, then the real recovery scorer's projected output.
What the Forecast Actually Shows You
| Output | What it tells you |
|---|---|
| Projected recovery score | Tomorrow's likely recovery number if you go to bed at the slider's current position — run through the same scoring logic as your real mornings, not a simplified stand-in. |
| Projected Reserve seed | The starting energy charge that tomorrow's Reserve gauge would open at — since recovery is the dominant input to the morning seed. |
| Confidence band | A ± range around the projection, shown only once the app has accumulated at least 7 reconciled forecasts against real outcomes — an honest "not confident yet" until the model has proven itself against your actual mornings. |
| Sleep hours | How many hours of actual sleep the chosen bedtime and your wake anchor translate to, once your personal efficiency ratio is applied. |
| Projected sleep debt | Where your sleep debt ledger would stand tomorrow morning if tonight goes as projected — the one debt figure that actually responds to the slider, since the standard ledger only looks backward. |
| Drivers | Plain-language reasons behind the projection — e.g. what's pulling the number up or down given tonight's specific inputs. |
A Worked Example
Say your personal median bedtime is 11:15pm and you've had two coffees today, the last one at 4pm. It's 10:30pm and you're deciding whether to wind down now or finish one more episode.
Why the Confidence Band Waits for Proof
Most forecasting tools display an uncertainty range from the first use, whether or not it means anything yet. The Tonight Simulator's ± band is deliberately withheld until the app has at least 7 reconciled nights — real forecasts compared against what actually happened the next morning. Before that point, showing a confidence interval would be manufacturing false precision from a model that hasn't been checked against your reality even once.
Once reconciliation history builds up, the band tightens or widens based on how well past projections actually matched outcomes for you specifically — a self-correcting model rather than a fixed formula applied blindly to everyone.
What the Forecast Can't Do
- It can't see stress, alcohol, or an unusual day — the model works from your sleep-stage medians, efficiency, wake pattern, and caffeine. A stressful evening or an unlogged drink will still affect real sleep in ways the forecast has no way to anticipate.
- It needs history to personalize — the stage-mix and efficiency medians need a minimum run of recent main nights before they're stable; early on, the forecast leans more on defaults.
- The confidence band starts absent — by design, not as a limitation to apologize for. It's the honest state before enough reconciled forecasts exist.
- It's a projection, not a promise — the actual number tomorrow depends on things that happen between now and wake-up. Treat the forecast as a decision aid for tonight, not a guarantee.
Turning a Score Into a Decision
The quiet shift the Tonight Simulator represents is moving sleep tracking from a rear-view mirror into something you can actually act on in the moment that matters. Every other number in the app — recovery, Reserve, sleep debt — tells you about a decision already made. This is the one that runs before the decision, using your own sleep patterns rather than a generic rule of thumb, computed entirely on-device from your Apple Health data.
See tomorrow before you sleep tonight.
The Tonight Simulator projects your recovery score, Reserve, and sleep debt from tonight's bedtime — built from your own sleep history, computed on-device.
Download Metrya Free