How Scoring Works

Why generic predictions fall short

Most race-time calculators and fitness trackers predict your times with a single, fixed formula applied identically to every athlete — the same population-average exponent whether you're a sprinter-built 5K runner or an ultra-endurance specialist, and the same "easy pace" credit whether your heart rate that day was textbook-controlled or drifting toward your max. That's a reasonable default with zero data on you — but it stops being the best available estimate the moment real evidence about your own physiology exists.

Split Index starts from the same published, standard formulas everyone else does — Riegel for race projections, DOTS/IPF GL for strength, ACWR for training load — but personalizes every one of them to your own logged history as soon as there's enough evidence to trust it, and is explicit about exactly when it's still falling back to a population default versus using your own data. The rest of this page is that methodology in full, not marketing copy — every constant and threshold named below is the real one the engine runs.

Every session gets two scores

One number cannot answer both of the questions you have about a session. "Was that any good?" and "was that any good for me?" have different answers, and forcing them into one number is how a scoring engine ends up telling you nothing: your standing against everyone else barely moves between a good easy run and a bad one, so a single score that tracks it barely moves either.

• vs everyone — this session measured against calibrated population standards for your sex and age. It is what the Engine Index, the Lab Index and the leaderboards are built from, and it moves slowly on purpose.
• vs you — the same session measured against your own recent sessions in that sport, at a comparable heart rate. 50.0 is your normal. Above it is a better day than usual, below it a worse one. This is the number that actually moves when a session goes well or badly.

The second one needs three comparable sessions in the last 90 days before it says anything. Until then it reads "calibrating" rather than inventing a middle.

What neither score reads is the session type you tagged it with. Easy, tempo, race — the tag says what you intended, your heart rate says what happened, and only the second of those is evidence. An identical session scores identically under every tag.

The one number both scores read

Both scores come from a single intermediate figure: the time this session implies you could post at your sport's benchmark distance — 5 km running, 2 km rowing, 400 m swimming, 20 km cycling, 2 km SkiErg, or per-kilometre pace for a walk — had it been a maximal effort, on flat ground, in comfortable conditions.

Building it takes five steps, in this order:

1. Project the session to the benchmark distance (Riegel's formula, with your own exponent when there is enough evidence for one).
2. Take out the time the hills cost you.
3. Take out the time the heat or the cold cost you.
4. On an erg, re-reference the time to the bodyweight the standards assume.
5. Scale for how hard you were working.

The same figure anchors your race predictions, which is why a well-executed easy run predicts sensible race times instead of stretching its deliberately slow pace out to a marathon.

How hard you were working

Your average heart rate is read as a fraction of your own heart-rate reserve — the range between your resting and maximum heart rates — not as an absolute number. 150 bpm is an easy jog for one athlete and near-threshold for another, and the engine should not confuse the two.

What your heart rate is compared against is the intensity a maximal effort of that session's own length would be held at — not a fixed number. A maximal 20-minute effort sits near 92% of your reserve; a maximal 90-minute one nearer 84%. Judging a 95-minute run against a 20-minute bar reads it as far easier than it was, and used to hand long steady runs a much larger advantage over shorter, faster ones than they had earned.

A session run below that bar implies you could have gone faster, so it earns credit. A session at or above it earns none and is never penalised: the average heart rate of an all-out 5 km always sits well below your true maximum, because heart rate ramps over minutes, and that gap is the effort rather than unused reserve.

How much pace a gap in effort buys is not constant either. Easing off buys proportionally more the further below race intensity you already are, so the conversion eases from about 2.2 at race-pace efforts to about 1.3 at easy-pace ones. That is Daniels' own pace table's shape, and it matters because easy pace is not one pace: people run easy anywhere between 4:00 and 7:00 per kilometre, and a model that reads every second of that spread as fitness over-reads it.

How much pace a given gap in effort buys depends on the sport, and the difference is large. On an erg, in the pool and on a bike, power rises with the cubeof speed, so 20% more effort is only about 6% more speed. On the road the relationship is much more direct. Scoring a rowing piece with running's conversion is what used to let a steady 6 km row read as a near-elite 2 km.

There is no ceiling on what an easy run can be worth. Through the whole range that has actually been measured, the credit is simply what the physiology says, so an outstandingly well-executed easy run scores like one instead of stopping at an arbitrary limit. Past that range the credit tapers, because beyond it the estimate is a guess rather than a reading, and the session's confidence falls away over the same stretch to say so. It never stops rising, so two sessions at different heart rates always score differently however easy both of them were.

Rowing, the SkiErg, swimming and cycling are damped sooner, and the reason is their scoring tables rather than their physiology. Erg pace compresses a wide range of fitness into a narrow band of time, so a few percent is worth hundreds of points there where the same few percent is worth tens on the road.

Walking is deliberately left unscaled: its economy changes sharply with speed and it is rarely a maximal effort, so a low heart rate on a stroll is not evidence of a fast benchmark walk.

Hills, weather and bodyweight

These adjust the score itself, not a separate secondary metric — the same pace on a hilly, hot day is a better performance than on flat ground in comfortable air, and the score should say so.

• Hills — only total ascent is recorded, so a route is treated as a loop: what goes up comes back down. The net cost is roughly 2% of your time per 10 m of climb per kilometre for running, less for walking and cycling, nothing on an erg or in a pool. It saturates, so each further metre of climb is worth slightly less than the last.
• Weather — credited above about 15°C and below about 5°C, in both directions and for outdoor sports only. Cycling carries a slightly larger cold adjustment than running, because wind chill scales with speed.
• Bodyweight— applied to rowing and SkiErg only, where the machine carries your mass and the standards assume a reference weight. It uses the same relationship Concept2's own weight adjustment does. Running, cycling and swimming have no accepted weight grading, so none is invented for them.

The score against yourself

Your baseline is the recency-weighted median of your own recent sessions in that sport, each read through the same five steps above. A median rather than an average, so one mis-logged session or one race does not redefine what counts as normal for you; weighted by recency, halving every 30 days, so this month's form counts for more than a block you finished six weeks ago.

Sessions are compared at a comparable heart rate. Your recovery jogs are judged against your recovery jogs and your tempos against your tempos, so a recovery jog does not permanently read as a terrible day and a tempo does not permanently read as a breakthrough. When you have not done a session at that intensity lately, the comparison stretches across intensities and says so.

Roughly 1% of performance is worth 2 points on the 0–100 scale, and the scale compresses smoothly at the extremes, so a genuine breakthrough still reads higher than a merely good day without either of them pinning to the top.

Your lifts work the same way: each one is scored against published strength standards for your sex, age and bodyweight, and separately against your own recent sessions of that lift, using your best set from each session rather than every set — warm-ups and back-off sets say nothing about what you could manage that day.

Without heart-rate data

If the session has no heart rate but you rated your effort, that rating is used instead. It is a rougher signal than a measurement, so the session is scored with visibly lower confidence, and it is never preferred over a real heart-rate reading when both exist.

With neither, the session is scored on pace alone. That is honest rather than generous: nothing is assumed about how hard you were working, so an easy run reads as the modest performance its pace shows. The fix is a heart-rate strap or an effort rating, and the session says which is missing.

Sessions recorded with and without heart rate are never compared against each other in your personal score, since one carries effort credit the other cannot. When there are too few of the same kind to form a baseline, the comparison widens to everything and tells you it has.

TRIMP

TRIMP (Training Impulse) is a single number blending how long a session lasted and how hard your heart was working throughout it — a rough measure of a workout's total training "cost," rather than a score of how well it went. A short, easy jog and a much longer one at the same intensity produce very different TRIMP values even though both might score similarly on effort quality.

As a rough guide: under 50 reads as light, 50–100 as moderate, 100–150 as hard, and above 150 as very hard. A 45-minute easy run might land around 60–80 (moderate); a hard interval session can push past 150. It uses the published Banister TRIMP formula (1991), weighted for your sex and heart-rate reserve.

Efficiency factor

Efficiency factor is how much speed (or power, for erg/bike sessions) you're getting per heartbeat — a rough proxy for aerobic fitness at a given effort. It has no fixed "good" or "bad" number: Split Index doesn't compare your efficiency factor against anyone else's, only against your own history. Trending upward over weeks at a similar effort means your aerobic fitness is genuinely improving; trending downward can be a sign of fatigue, illness, or detraining.

Erg/bike sessions show this as watts per beat-per-minute; pace-based sessions (running, rowing, swimming without a power meter) show it as meters-per-minute per beat-per-minute — the two aren't on the same scale, so don't compare a power-based reading against a pace-based one.

Decoupling

Decoupling compares your heart rate in the first half of a session against the second half, at a similar pace or power. A small amount of upward drift (your heart rate creeping higher for the same output) is normal, especially in heat or on longer efforts. A larger drift is a sign of fading aerobic durability or accumulated fatigue — even on a session that felt evenly paced throughout.

Lactate threshold and VO2max

Neither of these is a lab measurement — no consumer wearable measures actual blood lactate or gas exchange. Both are estimated from your own logged training.

Lactate thresholduses your own sessions explicitly tagged "Threshold" or "Tempo" — sustained efforts of 15 to 70 minutes. Races are deliberately excluded: a 5K sits near your VO2max effort, a marathon sits well below threshold, and mixing the two in would corrupt the estimate. The average heart rate and pace across your most recent (up to 3) qualifying sessions becomes the reading — more sessions means higher confidence, not a different number.

VO2maxuses Jack Daniels and Jimmy Gilbert's published VDOT formula, applied to your most recent logged race, or your predicted 5K time if you haven't logged a race recently. VDOT corrects for how long an effort was sustained, which is why it's calculated from a genuine hard effort rather than an easy run — feeding it an easy pace would read as a much lower fitness level than you actually have.

Your two 1RMs: all-time best and current

"My one-rep max" means two different things depending on when you ask, so you get both rather than one number pretending to be both.

All-time best is the heaviest single rep your logged sets have ever implied. It is a high-water mark: a bad session, a deload or six months off cannot lower it. Only beating it moves it.

Current 1RM is what your recent training says you could do today. Each session counts by how recently it happened, with a three-week half-life — a session from six weeks ago carries about a quarter of the weight of one logged today. So it rises when you train well and falls when you train worse, which is exactly what an all-time best must never do. Three weeks is our estimate rather than a measured constant: it sits inside the two-to-four weeks after which detraining studies start to see measurable strength loss, and roughly matches the length of a training block.

Each session is read by its best set, so warm-ups and back-off sets never make a good day look like a bad one. Neither number is stored — both are recalculated from your logged sets, so correcting a mis-typed session corrects them too.

Strength Index: DOTS and IPF GL

A raw total (squat + bench + deadlift) rewards bodyweight above almost everything else — a heavier lifter and a lighter lifter can move very different loads for the same underlying strength. DOTS and IPF GL are published, sport-standard formulas that adjust your total against your own bodyweight, so it can be compared fairly against lifters of any size, or tracked meaningfully over time as your bodyweight changes.

They're two independent formulas on two different scales — a DOTS score and a GL score for the same lift are not directly comparable to each other. Track each on its own axis over time rather than treating one as a conversion of the other.

Accessory and isolation lifts (anything outside squat/bench/deadlift) aren't part of either formula — they're scored instead against ExRx bodyweight-ratio tiers, a separate published standard for judging relative strength on lifts DOTS/GL don't cover.

Age grading

A 50-year-old and a 25-year-old lifting the same weight, or running the same time, have not done equally hard things. Age grading is how that gets accounted for, and the mechanism matters: your own numbers are never touched. Your lift, your bodyweight ratio, your finish time and your race predictions are exactly what you did. What moves is the standard you are judged against.

Endurance uses the established age-grading approach: your benchmark-equivalent time is compared against an age-adjusted standard rather than the open-class one, the same way published age-graded tables work. Under 36 the factor is exactly 1.0 — nothing is applied, and nothing is shown. Where it does apply, you are shown the exact figure: how much more lenient your standard is, and the factor behind it. Your predicted race times are never graded — those stay the real times you would have to run.

Strength works the same way — the anchor your ratio is measured against is eased by the age factor — and it covers both ends of the curve: juniors from 14 up, flat through the 23–35 peak, then a gentle masters climb. Where the factor is applied, the exact figure is shown to you on the lift.

The strength curve is marked beta, and that label is meant literally. The junior half comes from published Foster coefficients, but the masters half is our own estimate derived from a strength-by-age chart rather than a calibration against population data, and the two halves do not even agree on where the peak ends. We would rather show you the adjustment and tell you it is provisional than apply it silently or claim an accuracy it has not earned. Any lift carrying it is labelled beta.

Race predictions

Predictions across distances use Riegel's formula — T2 = T1 × (D2/D1)^k — the standard, published approach for translating a known performance at one distance into a projected time at another. The exponent k represents how much your pace naturally falls off as distance increases; a generic population value (roughly 1.06) is the reasonable default for an athlete with no history yet.

Once you've logged enough races and hard efforts across a couple of different distances, Split Index fits kto your own pace curve instead — somewhere in a realistic 1.03–1.10 range — so the ladder reflects how your own endurance actually degrades with distance, not a stranger's average. A real race result at or near a benchmark distance is trusted close to fully; an inferred projection from a different distance or session type is blended in more cautiously, so one unusual session can't swing your prediction on its own.

Predictions stay in a "calibrating" state until there's enough evidence logged to be worth showing — a single session isn't enough signal to project from confidently.

Injury risk (ACWR)

The Injury Risk Index is built on the Acute:Chronic Workload Ratio— your rolling 7-day training load divided by your rolling 28-day average weekly load. It's widely used in sports science as a proxy for whether recent training has spiked well above what your body has adapted to, which is when soft-tissue injury risk rises.

• Below 0.8 — Undertraining: recent load is well under your own baseline.
• 0.8–1.3 — Optimal: the well-supported sweet spot for building fitness without spiking risk.
• 1.3–1.5 — Caution: load is climbing meaningfully above your baseline.
• Above 1.5 — Danger: a spike large enough that the research literature links it to materially higher injury risk.

This needs a real baseline to mean anything — with less than ~2 weeks of consistent logging, the 28-day window is mostly empty and the ratio collapses toward a fixed, meaningless extreme rather than reflecting genuine overreaching, so it's hidden until there's enough history behind it. If you log a morning HRV reading, it nudges the index further — a suppressed HRV relative to your own baseline raises the index a little beyond what load alone would say, since incomplete recovery is itself a risk signal independent of training volume.

This is a training-load accountability tool, not a medical diagnosis — it reflects relative risk against your own history, not an absolute probability of injury.

The Recovery score

One number, 0–100, for what state your body is actually in today. It's a weighted blend of every recovery signal the app holds, and the weights are shown on the Recovery page next to each input rather than hidden:

• Training load (50%)— cross-domain readiness, built from your ACWR and fatigue across lifting and endurance together. A heavy squat session raises tomorrow's load exactly as a hard run does.
• HRV against your own baseline (30%) — a morning rMSSD reading compared with the mean of your recent readings, not against a population norm.
• Session density (20%) — how many sessions in the last 7 days, and whether a rest day happened at all. This is deliberately not a duplicate of ACWR: for someone whose norm is a lot of training, ACWR goes quiet while frequency still costs recovery.
• Alcohol — a deduction rather than a component, explained below.

Missing inputs are never counted as bad ones. If you've never logged an HRV reading, the remaining weights are renormalised over what is present and the breakdown says the component isn't contributing yet — an athlete without a chest strap shouldn't score the same as one whose HRV has collapsed.

Alcohol and recovery

Alcohol is modelled as a dose, not a drink count. What your body responds to is grams of ethanol per kilogram of bodyweight, so the same two pints are a materially different dose for a 95kg athlete and a 55kg one — and the distribution of alcohol through body water differs by sex, which is why the model uses your profile's weight and gender where you've given them.

Grams of ethanol come straight from volume × ABV × density — that part is arithmetic. Blood alcohol over time uses the Widmark model with a standard elimination rate, which is a well-established population estimate with wide individual variation: clearance rates differ by roughly a factor of two between people. Treat the curve as a guide to whether you're still processing a dose, never as a measurement. It is explicitly not a measure of fitness to drive and must never be used as one.

The cost to recovery scales with that dose and decays over about 48 hours, with an extra weighting for drinks taken inside the sleep window — the dominant mechanism is disrupted sleep architecture, and that only applies if there was sleep left to disrupt. Sustained weekly intake carries a smaller, separate cost, referenced against the UK Chief Medical Officers' low-risk guideline of 14 units a week.

The next-session forecast estimates a percentage decrement in endurance output and in peak force from the same dose-response shape, scaled by the hours between your last drink and the session. It follows the direction of the published literature — next-day decrements are real and dose-dependent, large around 1g/kg and small-to-absent at low doses, and a heavy dose after training suppresses muscle protein synthesis for around 24 hours, blunting the adaptation even when the session itself feels fine. These are estimates from a model, not measurements of you, and nothing here is medical advice or an assessment of anyone's drinking.