A sleep stage is an estimate, not a measurement.

A stage number begins as a signal.

A consumer wearable does not watch you sleep. It records signals such as movement and heart rate, and a model turns those signals into intervals marked REM, Core, or Deep. The label is the model’s reading of a proxy—not a direct observation of sleep itself.

Different devices carry different sensors and different models. Two of them can describe the same night and disagree, and both descriptions can be defensible. The watch is a sensor, not a judge.

No scandal. No defect. Only the distance between a signal and a stage.

Broad shape across many nights is what this kind of estimate is built to show. A single night’s stage percentages are the thinnest thing it produces, and the easiest to overread.

Provenance is the first honest question.

Distrust is not the useful response to an estimate. Provenance is. A stage percentage means very little until three plainer facts sit beside it.

Which source wrote the night → how much of it was actually covered → how many nights stand behind the pattern.

This is not a technicality. The same figure can come from a well-covered night inside a long, consistent record, or from a partly covered night on a source you started using last week. The figure looks identical. What it can carry is not.

Mama Bloom keeps those three facts next to the insight rather than behind it. The Living Almanac chapter carries the eyebrow “EVIDENCE · PROVENANCE”, and the reason is stated outright: “Coverage and provenance are shown so confidence comes from visible boundaries, not an invented score.”

Coverage says how much was really there.

Coverage is the share of a recorded interval that actually carries data. Mama Bloom shows it on the night detail as a percentage of that recorded interval, and again as a coverage figure beside each night in the record.

A gap does not disqualify a night. It changes how much weight the night can hold. When average coverage runs low, the Living Almanac says so instead of smoothing it over: the chapter reads “This chapter contains meaningful gaps,” and explains that the interpretation stays deliberately modest because the recorded intervals are incomplete.

A manual chapter is honest in a different way. You supply a start and an end, and Mama Bloom stops there: “Sleep stages remain unavailable because this interval was entered manually.” A manual entry never invents REM, Core, or Deep stages.

The free foundation Night Note inside the app puts the reading order plainly. Name the morning in your own words first. Then look only for provenance, coverage, and a pattern that repeats, and leave one-night stage percentages alone.

Every statement carries its night count.

A pattern needs nights, and Mama Bloom will not pretend to have them. A Night Accord requires at least three prior recorded nights. Below that it returns a Still learning reading and says why: “One night is a moment, not a pattern.”

Above that threshold it still shows its working. Each Accord ends with an evidence line naming how many nights the comparison used, and longer readings carry the hedge in full: “These are descriptive associations, not causes or medical findings.”

The thresholds are not decorative. A night is described as longer or shorter only when it differs from the median of the previous nights by 25 minutes or more. Anything closer gets a flatter sentence: “Your sleep record shows a night close to your recent duration.”

The 14-night practice is stricter still. A result is computed only when there are at least three prior nights, three nights you marked as tried that were also recorded, and three check-ins. Below any one of those it returns an honest non-result: “There is not enough tried-and-recorded data yet.” The evidence footer still prints, counting tried nights, recorded nights, prior nights, and check-ins.

A night count is not a formality. It is the difference between a coincidence and something worth watching a second time.

Two sources are not one record.

Sources change. A band replaces a watch. A watch misses a night and a manual chapter fills the gap.

Mama Bloom keeps that difference legible rather than blending it away. Entered rest and device-sensed sleep remain distinguishable throughout the Living Almanac, and the source ledger exists for exactly the moment when two nights should not be compared as if they were measured alike.

A seam is not damage in your record. It is the place where two kinds of evidence meet, and where a comparison quietly stops meaning what it appears to mean. Where a source changes, the app gives one plain instruction: let a few nights accumulate before comparing across the seam.

The boundary does not move.

Reading provenance does not make a stage estimate more true. It makes the estimate readable—a smaller and more useful claim.

Mama Bloom describes associations inside your own record. It says “associated with,” never “caused by.” It does not treat a sleep disorder, detect disease, or replace professional care.

Professional sleep-medicine guidance has been explicit on this point: consumer sleep technology is not a substitute for medical evaluation. Persistent or worrying sleep concerns belong with a qualified clinician, whatever a wearable reports.

Becoming a better reader of your own data is not learning to doubt every number. It is being able to say what a number is made of before deciding what it means. Source, coverage, night count—three questions small enough to ask in a moment, and enough to change what a screen is allowed to say to you. The estimate stays an estimate. The morning stays yours.