Most wellness apps show you what happened. Steps taken, calories burned, hours slept. Whole has always gone further, building a weekly rhythm around your actual habits rather than an idealised schedule. But until now, the patterns we surfaced were simple: which pillar you favour, what time of day you show up, how much sleep you average.
Today that changes.
From observation to intelligence
Your rhythm card now pulls from the Wellbeing Model, a 30-day rolling analysis that runs across every domain Whole tracks: sleep, mood, movement, nutrition, completion, energy, and cycle phase. Instead of isolated statistics, it finds the connections between them.
That means insights like:
These are not generic tips. They are derived from your data, recalculated as new signals arrive, and ranked by how surprising and actionable they are. The three most important insights surface. Everything else stays available in your full Insights view.
Cross-domain correlations
The most valuable patterns are the ones that cross domains. You might not notice that your Soul completion rate rises on days when you sleep more than seven hours. Or that high workout intensity on Monday consistently predicts lower mood on Tuesday. These are the kinds of relationships that take months to spot manually, if you ever spot them at all.
Whole uses Spearman rank correlation across same-day and next-day signal pairs. When a correlation is strong enough (above 35%) and backed by enough data (at least 10 days), it surfaces as a pattern insight. The stronger the signal, the higher it ranks.
Trend momentum
Static averages hide what matters most: direction. A 6.5-hour sleep average means something very different if it has been improving for two weeks versus declining for two weeks.
The Wellbeing Model tracks 14-day trend directions for sleep, mood, energy, and completion. When a trend crosses a meaningful threshold, it surfaces immediately. Positive momentum gets reinforced. Negative trends trigger protective adjustments, lighter moments, recovery-focused scheduling, earlier wind-down prompts.
Day risk scores
Not every day is equal. Most people have one or two days each week where completion consistently drops. It might be a demanding work day, a social evening, or simply the day where motivation dips.
Whole tracks your day-of-week completion history and assigns a difficulty score from 0 to 100. When a day scores above 60, your plan proactively lightens the load. You do not have to remember that Thursdays are hard. Your plan already knows.
Cycle-aware adaptation
For users who track their menstrual cycle, the Wellbeing Model learns phase-specific patterns. If your completion tends to dip during your luteal phase, your plan adapts intensity automatically. No manual adjustment needed, no guilt about missing a session that was never realistic in the first place.
How it works under the hood
Every time you sync health data, log a meal, complete a moment, or check in with your mood, a normalised signal is written to your Wellbeing Model. A background computation step recalculates baselines, trends, correlations, and day risk scores. The rhythm card reads this model and ranks insight candidates by a priority score: cross-domain correlations rank highest, followed by declining trends, sleep baselines, cycle patterns, and finally the basic rhythm observations that were there before.
When no model exists yet (new users, or before enough data has accumulated), the card falls back gracefully to the original rhythm insights. Intelligence grows with your data.
A fix for HealthKit sleep fragmentation
While building this, we discovered a bug affecting Apple Health users. HealthKit reports sleep as multiple intervals per night (deep, light, REM, awake). Our previous logic grouped these intervals by their start date, which meant a sleep session spanning midnight got split into two records. An eight-hour night became a 1.5-hour record and a 6.5-hour record. Over time, this accumulated hundreds of fragmented entries, dragging the median sleep average down to unrealistic numbers.
We have fixed the bucketing logic to attribute pre-6AM intervals to the previous calendar day, so a full night of sleep merges into a single record. We also added a four-hour minimum filter to exclude any remaining fragments from pattern calculations. If your sleep insights have seemed off, they should correct on your next sync.
What comes next
This is the foundation for a genuinely adaptive planning system. The same model that powers rhythm insights will soon drive automated plan adjustments: lighter days when your trend is declining, intensity boosts when momentum is strong, and real-time schedule modifications when your calendar or energy state changes.
Your plan is no longer a template. It is a living system that learns from you.