Protocols and research
How the Index is measured
Why each part of your Index is worked out the way it is, every published study it rests on, and where each one stops. How to do each test is written beside the test itself.
The scale
Your score is your percentile.
If you score 68, you performed better than 68 per cent of people your age and sex. Fifty is average. It is that simple, and it is deliberate.
Fixed thresholds — the kind that say everyone should manage a certain number whatever their age — mark too many older adults as failing and let too many younger ones off. Mayhew and colleagues showed this in 25,470 Canadian adults and recommend scoring against your own age and sex instead (Age and Ageing, 2023). This is that recommendation, implemented.
But average is not where we draw the line. Scoring 50 means declining at the same rate as everyone else your age. That is ageing on schedule rather than holding your ground, so Holding begins at 55: you have to beat typical to reach it.
The consequence follows, and it is better said here than discovered on your own result: scoring 50 puts you in Slipping, and roughly six in ten people land in Slipping or Reversing. That is not a fault in the scale. It is what the population data shows, and it is the whole point of measuring rather than guessing.
Why a percentile can rise while your reading stays the same
Your percentile compares you with people your age and sex, and on most of these tests the people you are compared with get weaker, slower or less fit as they get older. So if your grip holds steady while you age, your percentile climbs: you stayed where you were, and the comparison moved. That is a real result — most people lose ground as they age, and you did not. It is not an improvement in the reading itself.
Some of the published tables group ages in bands of five or ten years, so the change can arrive in one step when you move into the next band rather than a little each year. The chart for each test can show your readings either way, in their own units or as percentiles, so you can see which of the two moved.
Any one domain below 40 holds the Index at 54. A domain in Reversing means you cannot be better than Slipping overall, however strong the arithmetic elsewhere — a good heart does not buy back a floor you cannot get up from.
Failing the ten-second balance stand holds it at 39, but only between the ages of 51 and 75. That is the range the study validated, and outside it a failure is reported without a cap. Where either cap applies, the report states the uncapped figure and which domain was responsible.
The range
Why there is a range next to your number.
Every measurement carries error. A tape measure has less than an estimate of body composition; a laboratory VO₂ max test has less than a calculation from your resting heart rate.
Rather than hide that, we add it up and print it. 68, give or take 6 means the honest answer is somewhere between 62 and 74. The range says how much of your Index has been measured, and how well.
What narrows it
Each measurement adds certainty to its domain, and one taken on equipment — a treadmill test, a hand dynamometer, a scan — adds more than one you time yourself. A domain resting on a single self-timed test carries a wide range; a domain where everything has been measured on equipment carries a narrow one. Your Index’s range is built from the three. It never goes below 2 points, however much you measure: nine measurements do not make a number good to half a point.
Your resting heart rate and heart rate variability narrow it too, and never move your Index. How far depends on how much of your life the reading covers. A 90-day average counts in full, a 30-day average for 70 per cent of that, a 7-day average for 40 per cent, and one morning’s reading for 10 per cent. With both at full credit, your range is a third narrower than it would otherwise be. That is the same rule the assessment team uses. Like a measurement, a reading counts for less once it is more than 90 days old.
Heart rate variability comes as one of two measures, and one cannot be turned into the other. The Health app on an iPhone shows SDNN; Whoop and Oura show rMSSD. Either one counts, and each is checked against the range that measure can take. They are never added together.
What widens it
Time. A measurement counts in full for 90 days. After that the certainty it adds falls steadily, to about a third at a year. Your score does not move as a reading ages — only the range around it widens, because we know less about you than we did, not something different. An old measurement is never dropped: dropping one could empty a domain and take away an Index built on real data.
Which reading counts
You can update any one test every 14 days. Updating one does not hold up the others.
Your newest reading of each test counts, better or worse. Keeping your best would describe your best day, not what you can do now.
A reading taken at an assessment counts ahead of one you take at home, even when the home one is newer, until your next assessment replaces it. The home reading is still kept and shown, and because it does not count, it does not start the 14-day wait. A measured VO₂ max counts ahead of an estimate for the same reason: it is a better measurement of the same thing.
An assessed Index is a dated record. It stays in your history exactly as it was issued, on the day it was issued, whatever you measure afterwards.
Change
What counts as a change.
A change smaller than the error is not a change. So your history reports a move in your Index of 4 points or more as a change, and anything smaller as no change.
Four points is an estimate, carried over from the assessment protocol. Our own retest study will replace it with a measured figure.
For five single tests there is a figure: the smallest change the test can pick up, from published studies, the same figures the assessment team works to. The chart for each of these tests shades that much either side of your previous reading, and a new reading inside the shading is not a real change.
| Test | Smallest real change |
|---|---|
| VO₂ max, measured | 2 ml/kg/min |
| Heart rate recovery | 6 beats |
| Grip strength | 4 kg |
| Fat-free mass index, by DEXA or BodPod | 0.5 kg/m² |
| Sit-rise | 1 point |
These describe each test done carefully, under good conditions. Our own retest study will measure the same thing for these tests as members actually do them, and will replace them.
For the other tests there is no figure yet, and we do not guess one: their charts show your readings and say nothing about which differences are real, because guessing small would mean calling measurement error progress. That includes the fitness estimate, which moves when its inputs move rather than when a test is repeated. Rough guides, not measured figures: a dead hang varies by about 10 to 15 per cent from one attempt to the next, a pulse counted by hand by about 8 beats, and the one-minute sit-to-stand by about 2 repetitions. Treat a difference smaller than that as noise.
Capacity Age
The age your measurements are keeping up with.
Your Capacity Age is the youngest age at which your measurements would count as Holding — a score of 55 — against people of that age and your sex.
It is not the age at which you would score 50, because 50 is average, and average is declining on schedule. It describes what you can do and nothing more: it is not a biological age, a metabolic age or a heart age.
Why it is a range
It is worked out from your Index, so it carries your Index’s range with it. A two-point change in the Index can move it by a year or two, so it is shown with its range — 47 (44–51) — rather than as a single age that would claim more than the measurements can support.
Why the floor is 30
Capacity peaks between the late twenties and the thirties, and below 30 the reference values stop moving in one direction: on some tests, grip among them, people in their twenties score lower than people in their thirties, and on others they score higher. An age below 30 would come from that arithmetic rather than from you, so a result that already counts as Holding against 30-year-olds is reported as 30 or younger. Your Index goes on separating performance above that point.
Where it stops
It is worked out only over the ages at which the same tests can all be scored, so at every age it compares the same measurements. Where it reaches the end of those ages it says so — “or older”, “or younger” — rather than inventing a number. A test that is scored over only a narrow band of ages is left out of it and still counts towards your Index. Today that is eyes-closed balance, which is scored only between 50 and 69.
The tests
What each one predicts.
Cardiorespiratory
The strongest single predictor of mortality we measure. Scored against the FRIEND registry, which holds maximal treadmill tests verified by gas analysis. Where you have not had a laboratory test, it is estimated from age, sex, body mass index, resting heart rate and training habit, using the model of Jurca and colleagues. Both sides are maximal VO₂ max, so the estimate is compared like for like.
The estimate carries a standard error of about 5 ml/kg/min — roughly a sixth of the whole range from the 5th to the 95th percentile, and the largest single source of error in a self-measured Index. Scoring an estimate rather than a measurement widens the error around your whole Index, and the report says by how much rather than quietly absorbing it. A measured test narrows your Index more than any other single number you can add.
How fast your heart rate falls in the minute after exercise. Recovery below twelve beats carried a relative risk of 4 for death in the original study. A threshold of thirteen was confirmed in 19,551 people with no cardiovascular disease followed for twelve years: below it, 1.56 times the risk of death from any cause, and 1.95 times the risk of death from heart disease or stroke.
It discriminates hardest in people who are unfit to begin with. If your cardiorespiratory fitness is already normal it tells you less, and it is weighted accordingly — 30 per cent of the domain against the estimate’s 70.
There is no published table of recovery by age and sex anywhere, so the score beside the flag is read against a single older cohort — people referred for exercise testing, at an average age of 57 — so it reads high for most people here and lifts the Index with it. It has no age or sex bands at all, so a 30-year-old and an 80-year-old are scored against the same numbers. It counts, it is marked provisional wherever it appears, and it will be replaced by our own figures, taken under the same walking protocol.
Strength and tissue
The best-evidenced measurement in the battery. Across 139,691 people in seventeen countries, every 5 kg of lost grip carried a 16 per cent higher risk of death — a stronger predictor than systolic blood pressure. Scored against British centiles drawn from 49,964 people aged 4 to 90.
Stand and sit as many times as you can in sixty seconds. Reference values come from a Swiss population study of 6,926 adults.
No values exist above age 79. More than a fifth of over-80s could not complete the test, so the authors declined to publish percentiles for them, and neither do we — over-79s return unscored on this test rather than being given a number.
Upper-body pushing strength, and lateral trunk endurance. The side plank is scored on the difference between your two sides rather than how long you hold: an asymmetry needs no reference population, because you are compared against yourself. Two equal sides score 100.
There is no published percentile table for press-ups covering both sexes across the adult range. The band table this test used to carry was extrapolated from a study of 1,104 men and has been withdrawn, so your count is recorded, shown, and contributes nothing to your Index until real percentiles exist.
Sides differing by more than a tenth are flagged. That threshold is ours and provisional, from our own pilot rather than from a paper. The hold times themselves are not scored: the tables in circulation disagree with each other by a factor of two or three.
Fat-free mass index: your lean tissue for your height, from a DEXA scan, a BodPod or a multi-frequency scale. It needs no scan, so far more people can have it measured, and it is what the assessment at Grosvenor Square measures too. A single-frequency scale, calipers or an estimate is recorded and not scored. Scored against the US national health survey of 1999 to 2004.
The reference values come from DEXA scans on one make of scanner, Hologic’s. Every other method is scored against them too, and none agrees with that scanner exactly: the same person reads differently on a GE scanner, a BodPod or a scale. How far apart they are has not been measured for this product, so keep to one method, and treat a change between methods as no change at all. The values cover ages 20 to 85 and no others.
The survey publishes separate values for three ethnic groups. Rather than ask for your ethnicity, everyone is scored against one of them, the non-Hispanic white group; the differences between the groups are small but real. Two of the paper’s three authors worked for Hologic, the scanner’s maker, and said so in the paper.
Relative muscle power is a notably strong predictor: comparing the lowest quarter with the highest gave hazard ratios of 5.88 in men and 6.90 in women, where the same analysis for strength was not statistically significant. Power matters more than most people training for strength assume.
It is recorded and not scored. The jump has no published table by age and sex anywhere. The five-repetition chair rise has published tables, but no agreed weight in the Index yet, and setting one is a decision for the reference data rather than for a page.
Movement
Sitting down to the floor and standing back up, scored on how much support you needed. Followed in 4,282 adults for a median of twelve years: the lowest scoring group had a natural-cause mortality hazard ratio of 3.84 against the highest. Death rates ran from 3.7 per cent among those scoring a perfect 10 to 42.1 per cent among those scoring four or below.
The reference values are published as a chart of four bands rather than a table, so this test is coarser than the others by the nature of its source: your score is the middle of the band you fall in.
Standing on one leg. In 1,702 adults followed for a median of seven years, those who could not hold ten seconds had 1.84 times the risk of dying of those who could, independent of age, weight and existing illness. One in five could not.
That study covered ages 51 to 75 and nobody younger, so below 51 the ten-second test says very little either way. It is also observational: poor balance and earlier death travel together, which is not the same as one causing the other.
Balance is two measures doing two jobs. The ten-second eyes-open stand is the gate: pass or fail, and the only part of this domain with survival evidence behind it. Eyes open, the test ceilings badly in midlife — around seven in ten 53-year-olds hold it for the full thirty seconds, so it separates almost nobody. The eyes-closed version is the one that discriminates under sixty, and it is scored against 7,658 English adults aged 50 to 69 and nobody outside that range. That score says where you stand against people your age and sex; it does not predict anything. The same study’s eyes-open percentiles exist and are deliberately not used: their authors found no clear change with age, which is the reason the eyes-closed version was added.
How each test is done
The instructions are with the test.
What you need, the steps, what counts and the usual mistakes are written once, and shown beside each test in your Index and on each test’s own page — where you need them, which is while you are doing it.
The sources
Every study, and how many people.
Fifteen studies. Every reference value your Index is scored against comes from one of these.
| Study | What it gave us | People |
|---|---|---|
| PURELeong et al, The Lancet, 2015 | Grip and mortality | 139,691 |
| Twelve British studiesDodds et al, PLOS ONE, 2014 | Grip centiles, ages 4–90 | 49,964 |
| Non-exercise fitness modelJurca et al, Am J Prev Med, 2005 | VO₂ max estimate | 38,137 |
| Canadian Longitudinal Study on AgeingMayhew et al, Age & Ageing, 2023 | The case for scoring by age and sex | 25,470 |
| Mayo primary prevention cohortSydó et al, JAHA, 2018 | Heart rate recovery | 19,551 |
| Muscle power reference valuesAlcazar et al, 2021 | Lower-body power | 9,906 |
| FRIEND registryKaminsky et al, Mayo Clin Proc, 2015 | Measured VO₂ max percentiles | 7,783 |
| English Longitudinal Study of AgeingGrgic et al, GeroScience, 2026 | Balance and chair rise, English | 7,658 |
| Swiss population studyStrassmann et al, Int J Public Health, 2013 | One-minute sit-to-stand | 6,926 |
| Sit-rise reference scoresAraújo et al, Eur J Prev Cardiol, 2020 | Sit-rise norms, ages 16–98 | 6,141 |
| Sit-rise mortality follow-upAraújo et al, 2025 | Twelve-year outcomes | 4,282 |
| Cleveland Clinic cohortCole et al, NEJM, 1999 | Heart rate recovery threshold | 2,428 |
| Balance and survivalAraújo et al, BJSM, 2022 | Ten-second stand | 1,702 |
| Firefighter cohortYang et al, JAMA Netw Open, 2019 | Press-ups and cardiovascular events | 1,104 |
| US national health survey, 1999–2004Kelly et al, PLOS ONE, 2009 | Lean mass by DEXA scan, ages 20–85 | — |
| Total, without the US survey | 320,743 | |
The US survey’s count is left out because we have not yet checked the number of people its published values were drawn from against the paper itself. None of the values your Index is scored against comes from our own testing yet. When one does — the press-up tables will be the first — your report will say so, and mark it provisional until our group is large enough to publish.
The record
What we changed, and what we took out.
Every change to how a test is done or scored, and every table that has been withdrawn, with the reason. Nothing is removed from this list.
Heart rate recovery changed protocol in this version, v1.1, and is now measured during a slow walk rather than after a complete stop. Recovery thresholds are specific to what follows the effort — twelve beats upright with an easy walk, eighteen lying down, twenty-two sitting at two minutes — and the page previously asked for a stop while quoting the walking figure, which was a threshold borrowed from a protocol it did not belong to. The threshold is now native to the test. Below twelve beats is reported as a flag: Cole and colleagues (N Engl J Med 1999) found a relative risk of 4 for death, and Sydó and colleagues (J Am Heart Assoc 2018) found the same pattern in 19,551 people without cardiovascular disease.
The balance cap moved from 45 seconds to 30 on 3 September 2026, to match ELSA. The longer cap gave better discrimination in theory and had no published figures behind it; the shorter one is scored against 7,658 English adults with published percentiles. Both balance tests are a single attempt, not the best of three.
No printed tables any more. Every test here is either scored against published reference data or shown unscored with the reason, and the page reads that data directly: a printed copy is a second instrument waiting to disagree with the first, and this test has had two of those already.
Withdrawn: the Evolution band ladder, in full. Its last rung was the press-up table, extrapolated from a male cohort rather than measured.
Before that went the graded heart rate recovery table, the sit-rise ladder, the graded stair table, the single-leg sit-to-stand, and the seat-height and press-up-height conversion factors. The CLSA single-leg balance table is not used either: it is censored at its 60-second cap in 304 of 738 cells and cannot discriminate above the 20th percentile.
Recorded and shown, never scored, each for a stated reason. The dead hang has no reliable published reference data for adults. The stair climb’s two thresholds come from a conference abstract on 165 heart patients that was never published in a journal, so it is a signal and nothing more. Waist-to-height is reported as a flag by design. Press-up percentiles are being built; until they arrive, press-ups are recorded and shown unscored. Lower-body power and the deep squat hold are recorded without a score, for the reasons given above. Where a number is missing we leave the test unscored and say so, rather than filling the gap with one of our own.
The limits
What we cannot tell you.
The Index measures physical capability against people your age. It is not a medical test, it does not diagnose anything, and it cannot tell you how long you will live.
What the studies above show is that people who perform poorly on these measurements die earlier, on average, across large populations. That is a statement about populations and not about you. A high score is not a guarantee and a low one is not a sentence — and every one of these measurements is trainable, which is rather the point of measuring them.
If a measurement has no reference data for someone your age, we do not score it and we do not guess. You will see that stated on your own report rather than buried here. If a whole domain is left with nothing scored, there is no Index until it has something: a domain’s share is never quietly handed to the other two.
Where to start
Measure it yourself, or have it measured.
The Provisional Index is free. You enter your own measurements and get the score, the Capacity Age and the report — create your account on that page and it is kept, along with everything you add later.
A certified Index is measured by the Evolution team at Grosvenor Square on calibrated equipment. A self-measured result says so on its face: every estimate in it widens the error around your Index, and the report says by how much.