How Sahha Scores Are Built — and What "Accuracy" Actually Means

Why this question comes up

When someone first sees a Sahha score — a single number like “Sleep: 72” or “Readiness: 55” — the natural reaction is to ask: is that number right?

It’s a fair question, but it usually carries a hidden assumption: that a health score is a measurement with a true value, like a thermometer reading, that can be either accurate or inaccurate. That framing doesn’t fit what a score actually is, and answering “how accurate is it?” without first clearing that up tends to leave everyone talking past each other.

So this guide does two things. First, it explains exactly how Sahha builds a score, step by step, so there’s no black box left to be suspicious of. Second, it answers the accuracy question honestly — separating the parts where “accuracy” is a meaningful, measurable thing from the parts where the more useful questions are “is this grounded in evidence?” and “is it consistent?”

The short version: a Sahha score is not a prediction that can be right or wrong. It’s a transparent, research-backed synthesis of measurable behavioral and physiological signals, expressed on a consistent 0–100 scale. Every number can be opened up and traced back to the exact factors that produced it. That transparency is the accuracy story.

What a score actually is

A single raw metric is easy to collect but often misleading on its own. A step count tells you how many steps someone took — not whether they sat for eleven hours between two short walks. Eight hours of sleep sounds healthy until you learn it started three hours later than usual, after a week of mounting sleep debt.

A Sahha score solves this by synthesizing multiple related signals into one measure of a specific health dimension. There are five scores:

• Activity — daily movement, across 6 factors
• Sleep — sleep health, across 7 factors
• Readiness — recovery and preparedness, across 8 factors
• Wellbeing — a broad daily wellness measure, across 13 factors spanning activity and sleep
• Mental Wellbeing — behavioral patterns linked to mental health, across 6 factors

Each score is a value from 0 to 100. But the number itself is the least interesting part. What makes a Sahha score different from a generic “health number” is that it is built from independently measured factors, and every one of those factors is visible.

It’s important to be precise about what a score is not. Scores are not clinical instruments. They do not diagnose conditions and they do not replace medical assessment. They measure daily wellness patterns — the everyday behaviors that compound over time and shape how a person feels, performs, and recovers. Keeping that boundary clear is part of being accurate about what the scores claim to do.

How a score is built, step by step

Every score is constructed the same way. Understanding this four-layer structure is what dissolves the “black box” objection — because there is no black box.

Layer 1 — Biomarkers (the raw signals)

At the bottom are raw data points captured passively from the phone and, when available, a wearable: step count, sleep duration, heart rate, time spent moving, and so on. These come from Apple HealthKit and Google Health Connect. Before anything is scored, this incoming data is reconciled — when a user has multiple devices reporting the same thing, Sahha deduplicates and normalizes it into a single clean stream per user, automatically. Inaccuracy at this stage (double-counted steps from a phone and a watch, for example) is exactly the kind of error this step is designed to remove.

Layer 2 — Factors (each one scored on its own)

Each factor takes the relevant raw signals and turns them into four things:

• a value — what was actually measured (e.g. 8,200 steps)
• a goal — a static, evidence-based reference target (e.g. 10,000 steps)
• a sub-score — that factor rated 0 to 100
• a state — a plain-language label: minimal, low, medium, or high

Crucially, each factor is scored independently. The “steps” factor doesn’t know or care what the “sleep debt” factor is doing. This is what lets the system tell you not just that a score is low but which specific dimension is dragging it down.

Layer 3 — The score (the synthesis)

The overall score combines the factor sub-scores into a single 0–100 value with its own state label. The state bands are consistent across every score:

There are no hidden weights doing something mysterious here. The factor breakdown is the explanation. Anyone — a product team or an end user — can look at the factors and see precisely why the score landed where it did.

Layer 4 — The explanation (always attached)

Because every factor is exposed with its own value, goal, sub-score, and state, a score never arrives as a bare number. A worked example makes this concrete.

Here is a real Activity Score of 65 (medium) — despite a healthy-looking 8,200 steps:

Steps look great. So why only 65? The breakdown answers immediately: extended inactivity is minimal (9 hours of sitting against a 4-hour goal) and active hours is low (movement in only 4 of a target 12 hours). This is someone who walks to and from work but sits all day in between. The “right answer” here isn’t move more — it’s move differently, distributed across the day. No single metric would have surfaced that. The factor model does.

That traceability is the heart of the accuracy answer: you never have to trust the number on faith, because you can always inspect what produced it.

Where the factors come from: the evidence base

The most common worry beneath “how accurate is this?” is really “did you just make up these thresholds?” The answer is no. Every factor in every score — and the goal attached to it — is grounded in peer-reviewed research, not chosen arbitrarily.

Take the Sleep Score’s seven factors. Each one maps to an established body of evidence:

• Sleep duration — consistently sleeping 7–9 hours is associated with the lowest risk of mortality, heart disease, and metabolic disease; sleeping under 6 hours raises all-cause mortality risk by roughly 20–30%. The 8-hour goal reflects this.
• Sleep regularity — irregular bed and wake times are independently linked to higher mortality risk, even after controlling for how long someone sleeps. Someone sleeping 7 hours on a consistent schedule can have better outcomes than someone sleeping 8 on an erratic one.
• Sleep debt — accumulated shortfall over multiple nights measurably impairs reaction time, mood, and recovery, and a single long night doesn’t clear it.
• Circadian alignment — sleeping out of phase with the body’s clock is tied to elevated metabolic and cardiovascular risk.
• Sleep continuity, physical recovery (deep sleep), and mental recovery (REM) — each tied to specific, documented health outcomes.

The same is true across every score. The thresholds, goals, and the choice of which factors to include all reflect published health science. Each score has a companion “science” guide that lays out the specific studies behind every factor. When a customer asks “why is the steps goal 10,000?” or “why does regularity matter so much?”, there’s a citable answer — not a product opinion.

This is what “evidence-based, not vibes” means in practice: the model is built on external research, verified against Sahha’s own large-scale dataset, and developed with input from researchers in the field rather than assembled from intuition.

Two kinds of factors, and why it matters for accuracy

Within each score, factors split into two natural groups, and the distinction matters when you’re reasoning about reliability.

Behavioral factors capture what a person did — when they slept, how consistently, how much they moved, whether they’re accumulating a deficit. These are derived from patterns that the phone observes directly and are generally robust because they’re based on timing and counts rather than delicate physiological inference.

Physiological factors capture how the body responded — deep sleep, REM, heart rate variability, resting heart rate. These are more sensitive and informative, but they depend on signals that are measured most precisely with a wearable.

This grouping is why Sahha is upfront that some factors need a wearable and some don’t, rather than pretending every score is equally complete from a phone alone. Being explicit about that is itself part of being accurate.

”Accuracy” in the cases that worry people most

Two scores tend to draw the sharpest accuracy questions. They’re worth addressing head-on, because the honest answer is more convincing than a defensive one.

Mental Wellbeing — “how can a phone know how I feel?”

It can’t, and the score doesn’t claim to. The Mental Wellbeing Score is explicit that it does not measure mood and does not diagnose anything. It tracks six observable behavioral patterns — step volume, active hours, extended inactivity, activity regularity, sleep regularity, and circadian alignment — that research consistently associates with mental health outcomes.

The logic is correlational and behavioral, not psychic. When someone’s routines become irregular, their sleep timing drifts, and their daily structure erodes, that pattern often precedes or accompanies a change in mental wellbeing. The score surfaces the behavioral signal, which is objective and modifiable, without asking the user to self-report. Its published limitations are stated plainly: it can’t diagnose conditions, it can’t see psychological context like grief or a life event, and it isn’t a substitute for professional assessment. Framing the score accurately — as a supportive signal, not a verdict — is the responsible way to use it, and it’s the framing Sahha builds into the guidance.

Readiness — “I slept nine hours and feel great, why is my score low?”

This looks like an error and is actually the score working as designed. Readiness doesn’t measure last night in isolation. It compares today’s data against a shifting 30-day personal baseline and factors in accumulated sleep debt, recent physical strain, and cardiovascular recovery signals.

So a user who slept nine hours but has been under-sleeping all week, raced two days ago, and still has an elevated resting heart rate can legitimately score low — the accumulated load hasn’t cleared, even though last night was good. Think of it less as a performance report and more as a weather forecast: it flags hidden fatigue before the user feels it. When someone says “that doesn’t match how I feel,” the score isn’t wrong — it’s surfacing something they can’t yet feel.

The 30-day baseline also explains why Readiness becomes more accurate over time. In the first few days the baseline is rough; after two weeks it’s usable; after a month it’s well-tuned to the individual. It also adapts as fitness changes — a workout that once suppressed readiness for two days may barely register after months of training.

How the system stays reliable in the real world

Real-world data is messy. A robust scoring system has to handle gaps, multiple devices, and new users gracefully — and how it does so is a big part of whether you can trust the output.

Multiple devices are reconciled automatically. When a user wears more than one device, Sahha deduplicates and normalizes everything into one stream, so the same activity isn’t counted twice.

Missing data is handled honestly, not papered over. Not every factor is available every day — a user without a wearable won’t have sleep-stage data, and some days have tracking gaps. When a factor can’t be calculated it returns null, and the score adjusts around the factors that remain. Each score also defines a minimum factor threshold: if too few factors are available, the score itself returns null rather than producing a misleading number. In other words, the system would rather give you nothing than give you something falsely precise.

There’s no cold-start gap. When a user first connects, scores are generated retroactively for the previous 30 days. This matters specifically for accuracy: baseline-dependent scores like Readiness would be unreliable for weeks without that history, so the backfill removes that early period of bad estimates entirely.

Scores update in real time. Recalculation happens within about a minute of new data arriving, so the score reflects the user’s current state rather than a stale snapshot.

It works without a wearable — and says so. Every score is meaningful from phone data alone. Two scores, Activity and Mental Wellbeing, have 100% factor coverage from the phone. The others gain additional physiological factors from a wearable but never require one. Where a wearable improves things, it generally improves precision (better calorie estimates, better intensity classification) rather than adding or removing the score’s existence.

Consistency: the accuracy property that’s easy to overlook

There’s a kind of accuracy that matters enormously for a product but rarely gets named: consistency. Because every Sahha score uses the same 0–100 scale, the same state bands, the same factor model, and the same evidence-based goals across every user and every device, a score means the same thing everywhere.

A “Sleep: 72” for one user is directly comparable to a “Sleep: 72” for another, regardless of which phone or wearable produced it. That standardization is something a single device’s native score can’t offer, and it’s often more valuable to a product than chasing the last few percent of precision on any individual measurement. For dashboards, segmentation, trend detection, and triggering the right nudge at the right moment, consistency is the property that makes the scores trustworthy at scale.

How to answer the accuracy question

When a prospect or customer asks “how accurate are your scores?”, the strongest answer reframes rather than deflects:

1. A score isn’t a single measurement that’s right or wrong — it’s a transparent synthesis of multiple validated factors, each independently scored and fully visible.
2. Every factor is grounded in peer-reviewed research, with citable evidence behind every threshold and goal, verified against a large real-world dataset.
3. Nothing is hidden — the factor breakdown is the explanation, so any number can be traced to exactly what produced it.
4. The system is honest about its limits — it returns null rather than guessing, distinguishes phone-only from wearable factors, and states plainly that scores measure wellness patterns, not clinical conditions.
5. It’s consistent and personalized — the same scale means the same thing across all users and devices, and baseline-driven scores get more tuned to the individual over time.

That combination — transparency, evidence, honesty about limits, and consistency — is a more durable answer than any single accuracy percentage could be. The point isn’t that the number is infallible. The point is that you can always see exactly why it is what it is.

Further reading

For teams that want to go deeper, the source material behind this guide:

• What are Health Scores and Score Factors — the foundational model
• Score guides — Activity, Sleep, Readiness, Wellbeing, and Mental Wellbeing, each with full factor tables
• Science guides — the peer-reviewed evidence behind every factor in each score: Activity, Sleep, Readiness, Wellbeing, and Mental Wellbeing
• Scores documentation — API schemas, the 0–1 decimal format, null handling, and integration details
• Science & Research — Sahha’s research program, dataset, and published studies
• Scores platform overview — the product-level summary of all five scores