Health score
The current estimate of the health pattern represented by the information available.
Evidence & methodology
Evidence-informed health context,without pretending certainty.
Longevity Signal combines nine health domains into one connected view. The scoring system is deterministic and versioned: defined inputs go through defined health models. Evidence quality, missing information and the health estimate are deliberately kept separate.
Scoring engine v2.0.09 health domainsMissing data stays unknownAI does not set the score
Three different ideas
Score, representation and confidence are not the same thing.
Collecting more information can make the health picture more complete without automatically making the health score better. Longevity Signal keeps these concepts separate on purpose.
Representation
How much of the nine-domain model is currently supported well enough to contribute.
Confidence
How strongly the available evidence supports the estimate that is being shown.
Overall model
Nine signals. Defined weights.
Each assessable signal receives a domain score. Those scores are combined using the current model weights below.
Cardiometabolic14%
Activity14%
Tobacco13%
Nutrition12%
Bloodwork12%
Sleep11%
Alcohol9%
Social connection8%
Stress & resilience7%
These percentages are product-model weights. They reflect the structure of the current Longevity Signal model; they do not mean that, for example, exactly 14% of a person's lifespan is caused by cardiometabolic health or activity.
How the number is built
A weighted health estimate, not a lifespan prediction.
The overall score is produced by a fixed calculation after each contributing domain has been interpreted.
01
Score each assessable domain
Each domain uses its own defined model. A domain that does not have enough usable evidence is not silently assigned a poor score.
02
Combine represented domains
The observed score is the weighted average of the assessable domains. The calculation is normalized over the weight that is actually represented.
03
Keep missing information unknown
A missing signal does not become zero. Instead, representation shows how much of the complete nine-domain model is currently contributing.
04
Measure evidence confidence
Confidence is calculated separately from health state. It describes the reliability and representativeness of the available evidence, not whether the health result is good or poor.
05
Require enough evidence
The current engine does not publish an overall numerical score when fewer than four domains are assessable or when less than 40% of the model weight is represented.
06
Surface high-impact findings
Certain clearly high-impact patterns can be highlighted separately and can reduce the overall score. The combined high-impact adjustment is capped at 25 points in the current model.
Simplified scoring formula
Observed score
Σ(weight × domain score) ÷ Σ(represented weights)
Final numerical score
observed score − applicable high-impact adjustment
Estimate maturity
The app shows how established the picture is.
Representation determines whether enough of the model exists to show an overall number. Confidence then helps describe how mature that estimate is.
Insufficient data
<4 domains or <40% represented
The overall numerical score is withheld.
Early estimate
Confidence <40%
Enough of the model is present, but supporting evidence is still limited.
Developing estimate
Confidence 40–74%
The picture has useful evidence but is still developing.
Established
Confidence ≥75%
The contributing information provides stronger support for the current estimate.
Source quality also matters. In the current scoring engine, questionnaire-only Activity, Nutrition or Sleep evidence cannot produce a domain score above 90. This prevents a limited source from appearing more definitive than it is.
The nine signals
Why these health domains are included.
The domains are informed by established research connecting behaviors, exposures, biomarkers and psychosocial factors with long-term health. The scientific literature supports the importance of the domains; the exact Longevity Signal weights and point calculations remain product-model design choices.
Activity
14%Looks at movement and physical-activity evidence such as daily activity, aerobic exercise, strength activity and fitness information when available.
Evidence context
Regular physical activity is associated with lower risk of major noncommunicable disease and better physical and mental health. WHO recommends 150–300 minutes of moderate aerobic activity per week for adults, or an equivalent amount of vigorous activity, together with muscle strengthening.
Sleep
11%Uses sleep evidence to describe the recent sleep pattern rather than treating one unusual night as the whole signal.
Evidence context
Sleep is recognized as an important component of cardiovascular health. The American Heart Association includes sleep in Life's Essential 8 and describes 7–9 hours per night as the usual target range for adults.
Nutrition
12%Builds a dietary pattern from repeated meals over time. Meal-photo AI helps identify visible foods, but the scoring model uses the resulting structured pattern rather than asking AI to invent a health score.
Evidence context
The model emphasizes overall dietary pattern and recurring food categories. WHO describes adequacy, balance, moderation and diversity as core principles of a healthy diet, with minimally processed and unprocessed foods forming an important foundation.
Cardiometabolic
14%Brings together available body and cardiovascular measurements such as blood pressure, body measurements and other supported cardiometabolic observations.
Evidence context
Blood pressure, body weight, blood glucose and blood lipids sit within established cardiovascular-health frameworks. Longevity Signal keeps these measures in context rather than treating any one measurement as the entire health picture.
Bloodwork
12%Organizes supported laboratory biomarkers into a current view using confirmed results. Lipid, glucose, kidney and other supported markers are interpreted with their units, dates and clinical context preserved.
Evidence context
Clinical guidance evolves by biomarker and clinical situation. The model draws on established guidance rather than treating every laboratory value identically. Examples include contemporary kidney-disease and lipid guidance.
Tobacco
13%Distinguishes never, former and current smoking and can consider cessation history and reported second-hand exposure. Missing smoking status remains unknown rather than being assumed favorable.
Evidence context
Current cigarette smoking is a major modifiable health exposure. Health benefits begin after cessation and continue to accumulate over time, including reductions in cardiovascular, respiratory and cancer risk.
Alcohol
9%Looks at the reported pattern of alcohol exposure, including overall consumption and heavy drinking episodes, rather than awarding a health bonus for drinking alcohol.
Evidence context
Alcohol is associated with acute and chronic health risks. WHO notes that risk depends on the amount and pattern of drinking, with much of the harm arising from heavy episodic or heavy continuous consumption.
Stress & resilience
7%Uses short check-ins across stress burden, coping ability, sense of control and interference with daily life. The aim is to describe a pattern, not label a normal difficult day as disease.
Evidence context
Stress responses vary substantially between people. Persistent or excessive stress can affect physical and mental health, while coping and resilience influence how demands are experienced and managed.
Social connection
8%Focuses on perceived support, meaningful contact, belonging and loneliness. Living alone, introversion or enjoying solitude are not automatically treated as poor social health.
Evidence context
Social connection is associated with health and well-being across multiple outcomes. The U.S. Surgeon General's social-connection materials describe social isolation as a meaningful public-health risk.
High-impact context
Some findings deservemore attention than an average.
A weighted average can hide an important health exposure. The current model therefore has a separate high-impact layer for selected patterns such as current combustible smoking, markedly elevated blood pressure or glycaemia, kidney dysfunction and clearly harmful alcohol exposure.
A high-impact flag is not a diagnosis.
It is a product-level signal that a finding may deserve appropriate context or professional review. Individual clinical interpretation depends on history, repeat measurements, medications, symptoms and other information the app may not know.
AI boundaries
AI assists.Defined models score.
AI can help with bounded tasks such as observing foods in a meal photo, extracting structured information from a laboratory report and explaining structured findings in clearer language.
The central Longevity Signal score is not generated by asking an AI model to judge whether someone is healthy. The scoring path is deterministic: the same structured inputs and model version produce the same scoring result.
Users should review extracted or AI-assisted information where the app provides that opportunity, especially when the original source is a medical document.
Important limitations
A health picture is nota medical verdict.
Longevity Signal does not predict how long a person will live. The score is not a validated clinical risk equation, diagnosis, treatment recommendation or substitute for professional medical care.
Population research can show associations and inform health guidance, but it cannot determine an individual's outcome. Genetics, medical history, medications, socioeconomic conditions, environment, measurement error and many other factors can affect health and are not fully represented by this model.
The methodology is versioned so that the model can evolve as the evidence, product and supported data improve.