SYNTH-38 · Guide
Heart-rate anchors
Take the threshold from a field effort, never from a formula, and know the three conditions under which heart rate stops describing the session
The decision this framework makes
Which heart-rate number an athlete's zones are built on, how to obtain it without a laboratory, and when to stop reading heart rate and read power, pace or perceived effort instead.
Key Findings — the three facts this framework is built on
An age formula is a population statistic, not a person. Across 5069 endurance athletes (the abstract's count; the Results report 5311, corrected 2026-10-12), 9 of 13 formula-and-mode combinations differed significantly from the measured maximum, and every formula carried RMSE 9.1-10.5 bpm (REF-3494).The group-level bias can be near zero while the individual is a full zone out.The best-known replacement, 208 − 0.7 × age, is itself a population line: it was fitted to group mean HRmax values from 351 studies (18,712 subjects) and confirmed in 514 healthy adults with measured HRmax (REF-1082, Tanaka 2001, B+, healthy adults, not athletes), and its abstract reports no individual error — it estimates the average for an age, not this athlete's maximum.
A percentage of a maximum is a weak description of what a session did. In 22 well-trained cyclists (19 men, 3 women, VO2max 67.1) over 9 weeks of 5 × 8 min sessions with oxygen uptake measured in every session, the fraction of VO2max sustained during intervals predicted the adaptation (R²=0.54 for VO2max, 0.44 for max power at the VO2max test, 0.36 for a performance index, 0.25 for power at 4 mmol); %HRmax related to fewer outcome variables and repeated worse from session to session (REF-041, and SYNTH-16 for why).And a fixed percentage of a maximum does not even land people in the same intensity domain: in 100 adults the maximal lactate steady state sat anywhere from 75% to 97% of HRmax (lactate threshold 60-90%), women higher than men (REF-038).
Heart rate drifts upward at a fixed workload, and in the one controlled comparison the base holds, most of it came from heat. With sweat losses fully replaced, between minutes 15 and 45 of a 45-min ride at 59.2% VO2max heart rate rose 11% and VO2max fell 15% at 35 °C, against +2% and a non-significant 5% fall at 22 °C (REF-3496, 10 trained men); the rise begins after roughly 10 min (REF-3495).
1. Rank the anchors
| Anchor | What it is | Use it for |
|---|---|---|
| Measured HRmax (maximal ramp, or the highest value in a hard race or hill effort — practice; the cited study measured it by maximal CPET) | A real ceiling for this athlete in this sport | Percentage-of-max zones, if they are used at all |
| Threshold heart rate (LTHR) from a sustained field effort | The heart rate the athlete actually holds at the boundary they train against | The base's preferred field anchor (practice — no validation study in the base); zone boundaries |
| Age-estimated HRmax | A population mean with roughly ±10 bpm of individual error (RMSE 9.1-10.5 bpm, REF-3494) | Talking about heart rate at all when nothing better exists — and always with the error stated |
The ranking matters more than the arithmetic. Two estimates stacked — a percentage of an estimated maximum — is not a signal.
Measuring HRmax in practice (Olympiatoppen guideline, level C — expert practice, not a study): 3 × 3 min with 2 min of active rest, each interval harder than the last and the final one to exhaustion; the highest heart rate recorded is HRmax.
Test each sport separately — the same athlete can have ~5% higher HRmax running than cycling (the guideline cites Millet 2009 for this figure; REF-537, that review, states only that heart rate differs between running and cycling at maximal and submaximal intensities, not the size of the gap) — and retest at least every three years, since HRmax normally falls with age (REF-1077, C).
2. Getting LTHR from a field effort
A widely used coaching convention, and it should be presented as one: hold the hardest effort that can be sustained evenly for 20-30 minutes, on terrain that allows a steady effort, and take the average heart rate of the final 20 minutes as the threshold anchor. Run it fresh, warmed up, in cool conditions, on a route without long descents or stops.
This base has no validation study for that protocol. It is standard practice, not a finding — say so when prescribing it.
What the base does support is the shape of the reasoning: an anchor taken from the athlete's own maximal-sustainable effort avoids the ±10 bpm formula error (REF-3494 measured that error for HRmax formulas and did not test threshold anchors — inference), and a boundary is better measured than assumed (SYNTH-27 makes the same argument for power, where the anchor should be critical power rather than an FTP estimated from a 20-minute test with a correction factor).
Re-take it when the training block changes, after illness, and at least each season. Not from a study; the base's own practice — heart rate at threshold is not fixed.
3. The three conditions under which heart rate stops describing the session
Reps shorter than about two to three minutes (the base's own practice, not a measured constant). The response lags the effort, so the rep is over before the heart rate has arrived. Judge these by power, pace or the effort itself — and by whether the last rep matched the first, which is what actually tells you the session worked (SYNTH-33).
Heat, and anything past the first ten minutes of a long effort. The drift is real and reflects a genuinely higher relative intensity, not noise (REF-3495, C+); in the one controlled comparison it was far larger at 35 °C than at 22 °C (REF-3496), with dehydration and glycogen depletion not separated; a hot ride's heart rate is not comparable to a cool one's, and holding a heart-rate target in the heat means the power must fall.
Any day the athlete is not themselves (base practice; no cited entry measures these) — illness, heavy sleep debt, caffeine, altitude, dehydration on top of heat. The number moves for reasons that have nothing to do with the training intent.Day-to-day heart rate varies by roughly 6-12 beats, and the guideline lists altitude, climate, hydration, glycogen and mental stress among what moves it (Olympiatoppen guideline, level C — expert practice, not a study); the guideline cites two older studies (REF-1077, C). Inference: a difference of that size between two days is not a signal on its own.
Conversely (base practice), heart rate is at its most useful exactly where it is most often ignored: long steady aerobic work, where the effort is low enough that lag does not matter and long enough that the drift itself is information.
4. What follows for prescription
Not from a study; the base's own practice: where power or pace exists, use it to SET the session and heart rate to READ it. Where it does not, SYNTH-40 carries the no-meter case.
Practical Application
Never gate a decision on a percentage of an estimated maximum. If HRmax was never measured, an "83% of max" reading is a percentage of a guess with ±10 bpm of error in it (REF-3494) — it cannot justify cutting a session or declaring one easy.
Get a threshold heart rate from a 20-30 minute maximal steady effort and use the average of the last 20 minutes, on cool, steady terrain, when fresh. Present the protocol as standard practice rather than as evidence, because that is what it is.
Record whether the anchor was measured or estimated, and carry that label everywhere it is used. Measured and estimated are different data and should not be spoken about in the same tone.
Do not use heart rate as the target in short intervals, and do not compare it across conditions in the heat or late in long rides (base practice). In short intervals use power, pace or perceived effort; in the heat and late in long rides the rising heart rate is a real rise in relative intensity (REF-3495) and easing power to hold it is correct.In 5 × 8 min sessions (not short reps, not heat) %HRmax predicted fewer of the interval adaptations than %VO2max and repeated worse session to session (REF-041). The two-to-three-minute lag figure is the base's own practice, not a measured constant.
Compare heart rate to the athlete's own sessions in similar conditions, never to a table. A 13 °C temperature difference moves the drift from about 2% to about 11% with everything else held constant (REF-3496).
Expect a different maximum in each sport — measured HRmax was about 2 bpm higher in the 4043 runners than in the 1026 cyclists of the same cohort (abstract count; the Results give 1268 cycling tests, corrected 2026-10-12) (184.6 vs 182.7 bpm, REF-3494) — different people, so the within-athlete gap is inference, but a cyclist's zones should not be assumed to transfer to their running.
What NOT to do
Do not switch age formulas hoping for accuracy: the bias between the best and the worst is about 5 bpm, and the individual error inside every one of them is roughly twice that (REF-3494).
Do not tell an athlete that "75% of HRmax" is an easy ride for everyone — in REF-038 the lactate threshold sat anywhere between 60% and 90% of HRmax, so the same percentage is moderate for one athlete and heavy for another.
Do not treat a high drift on a hot day as dehydration. Sweat losses were fully replaced in REF-3496 and the heat effect appeared anyway.
Do not build a threshold anchor from a session with descents, traffic lights or a pacing collapse in it — an uneven effort produces an average that describes nothing.
Limitations
The field-test protocol in section 2 is a coaching convention. This base holds no validation study for it, and the 20-30 minute window and last-20-minutes rule should never be cited as a finding.
REF-3494 is a cross-sectional measurement study (B+); its cohort is 83-90% male, mean age in the mid-30s (racial homogeneity is a full-text statement) — the formula error in masters athletes, in whom age formulas matter most, is not well characterised there.
REF-041 is 22 well-trained cyclists over 9 weeks in a single-arm regression design; the weaker showing of %HRmax as a predictor is one study's result on one outcome set, not a proof that heart rate is uninformative.
The 5% and 10% decoupling lines used in this app are house conventions and appear in none of the cited papers. Do not attribute them.
How this entry was checked and changed
Why this level
regraded 2026-09-27 from B+: its primaries REF-3494, REF-3496 (B+) and REF-041 (B) were cited as A-level; the field-test protocols themselves are labelled conventions, not findings
Content correction 2026-09-27
"REF-041: 30 trained cyclists; %HRmax did not reach significance" -> 22 well-trained cyclists (19 M, 3 F, VO2max 67.1); %HRmax related to fewer outcome variables and had poorer session-to-session repeatability; R² for VO2max (0.54) and performance index (0.36) added.
Level parenthetical "A-level primaries REF-3494, REF-041, REF-3496" -> REF-3494 and REF-3496 are B+ (cross-sectional measurement study; crossover n=10); only REF-041 is A-rated. Synthesis level B+ unchanged.
REF-3496 drift stated over the measured window (minutes 15-45 of 45 min at 59.2% VO2max) with the non-significant 5% cool-condition VO2max fall; the 2-3 min interval-lag figure labelled base practice.
Added REF-038 (corrected entry, B+): MLSS 75-97% and LT 60-90% of HRmax across 100 adults — direct support for "a fixed % of max is not a zone".
Level B+ -> B: REF-3494, REF-041 and REF-3496 were cited as A-level primaries; they are B+/B.
Resync 2026-09-28: "running ran about 2 bpm higher than cycling in the same 5069 athletes" -> REF-3494 compared two different groups (4043 runners vs 1026 cyclists; 184.6 vs 182.7 bpm), not the same athletes in two sports; the within-athlete transfer is now labelled inference.
2026-10-01: cross-reference repointed from a merged framework to the one that replaced it (SYNTH-01→31, 05→34, 09→70, 12/15→32, 19→26).
Verification 2026-10-01: all numbers check — 5069 / 9 of 13 / RMSE 9.1-10.5 / 184.6 vs 182.7 / ~5 bpm bias spread (REF-3494); 22 cyclists, R² 0.54/0.44/0.36/0.25 (REF-041); MLSS 75-97% and LT 60-90% HRmax, n=100 (REF-038); +11%/−15% vs +2%/ns 5%, 59.2% VO2max, minutes 15-45, n=10 (REF-3496); ~10 min onset (REF-3495). Level parentheticals match head blocks.
Verification 2026-10-01: Level B is honest and the Level line already discloses that the field-test protocol is convention; no LEVEL-OVERALL flag. The 'never from a formula' half is evidence (B+); the 'take it from a field effort' half is unvalidated practice, correctly labelled in §2.
Verification 2026-10-01: REF-3495 is a C+ narrative review written from the abstract only (has DOI/PMID, so not a SOURCE flag); it is the only support for the 10-minute onset.
Verification 2026-10-01: the main defect is direction, not numbers — the cited drift entries say rising HR in heat is real relative intensity ('not noise'), while the framework's lead says HR 'stops tracking' and hands the session to power.
Revision 2026-10-04: added Olympiatoppen's Intensity Scale v2 (2024) REF-1077 as practice lines labelled "(Olympiatoppen guideline, level C — expert practice, not a study)"; no default, decision row or level changed.
Revision 2026-10-04: Key Findings, first bullet — added Tanaka 2001 (REF-1082, B+: meta-analysis of group means from 351 studies / 18,712 subjects + lab cross-validation in 514 healthy adults) to show that the 208 − 0.7 × age formula is itself a population estimate with no individual error in its abstract; cross-reference added. No rule, default or level changed.
Verification 2026-10-04: the ~5% running-vs-cycling HRmax gap is the guideline's figure; the base's Millet 2009 entry REF-537 contains no such number (only 'heart rate differs between the two activities at both maximal and submaximal intensities') — citation narrowed.
Verification 2026-10-04: 'varies by roughly 6-12 beats even without these' overstated REF-1077/olt.txt, which says the 6-12-beat daily variation is itself influenced by altitude, climate, hydration, glycogen and mental stress; the 'not a signal on its own' conclusion was unlabelled and is now marked inference.
Verification 2026-10-04: HRmax test protocol, retest every three years, Tanaka 2001 sentence (REF-1082: 351 studies, 18,712 subjects, 514 healthy subjects, no individual error) and cross-reference checked OK. No default, decision row or level changed.