The evidence

Two runners finish the same marathon. One takes 2:45, the other 5:15. They cover identical ground and they need completely different fuelling plans, because the thing that limits fuel delivery is how fast your small intestine can move sugar into your blood — and that is a rate, in grams per hour. Distance tells you almost nothing about it. Duration and absolute metabolic rate tell you nearly everything.

Below about 45 minutes, carbohydrate during exercise is pointless. Muscle glycogen is not limiting, and anything you swallow cannot be absorbed and oxidised fast enough to matter 12. Between roughly 45 and 75 minutes the only defensible intervention is a carbohydrate mouth rinse or a token 15–30 g/h, and the mechanism is in your mouth and brain, not your muscles 28.

The ladder

The 2.5-hour split matters more than it looks. The 2016 joint position stand of the Academy of Nutrition and Dietetics, Dietitians of Canada and the American College of Sports Medicine recommends 30–60 g/h for endurance exercise lasting 1 to 2.5 hours, and reserves 90 g/h for exercise beyond 2.5 to 3 hours 1. A lot of apps quietly apply the 90 g/h number from the two-hour mark. That hands a 2:10 runner an upper bound from a band whose cited authority caps that athlete at 60 g/h. We split at 150 minutes.

Notice that the curve is humped. It peaks somewhere in the 3–5 hour band and comes back down. That is deliberate. As events get longer, absolute metabolic rate falls sharply, gut tolerance degrades, sweet things start to taste awful, and the field data for 12–24 hour races cluster at 30–66 g/h 3. The ISSN ultra-marathon position stand sits at the conservative pole with 30–50 g/h, or 150–400 kcal/h 6. The descending limb is supported by observational data (grade B); the exact point where it turns over is our judgement (grade C).

What people actually do

The gap between the guideline and the practice is enormous. In a field study of 221 competitive endurance athletes, the marathon cohort self-selected 35 ± 26 g/h — the lowest of any event measured 7. In single-stage ultras, finishers averaged 66 ± 27 g/h against 42 ± 23 g/h for non-finishers 3. Meanwhile the best dose-response study we have, n = 51, found returns flattening above about 78 g/h 5. For most runners the useful advice is not "push to 120", it is "you are probably taking half of what you could tolerate."

Does body mass change the number?

The mouth rinse, honestly

Swilling a carbohydrate solution and spitting it out works, in a narrow window. Two independent meta-analyses support it for high-intensity efforts of roughly 30–60 minutes 89. In the fasted state the effect is 2–3% 8. In the fed state — which is how most people race — the pooled effect is Hedges g = 0.18 (95% CI 0.09–0.28), rising to 0.26 for aerobic exercise and falling to a non-significant 0.02 for anaerobic work 9. Protocol: about 25 mL of a 6–6.5% solution, swill 5–10 seconds, spit. Beyond 10 seconds the effect disappears (g = −0.03) 9. There is no concentration–response relationship, so do not chase a stronger mix (β = 0.01, p = 0.54) 9.

Glucose, and the glucose that maltodextrin is made of, crosses the wall of your small intestine mainly on a transporter called SGLT1. SGLT1 saturates. Once you are swallowing more than about 1.0–1.2 g/min, exogenous carbohydrate oxidation plateaus at 0.5–1.1 g/min — that is 30–66 g/h, across individuals — and swallowing more glucose does not raise it 12.

Fructose does not use SGLT1. It crosses on GLUT5, a different, unsaturated route. Co-ingesting the two adds a second lane to a road that was already at capacity. Mixed glucose-plus-fructose feeds raise exogenous oxidation by 20–55% over an isoenergetic single-transporter feed 14, with measured peaks of 1.26–1.40 g/min at 108 g/h 24, 1.51–1.68 g/min at 120 g/h 789, and 1.70–1.75 g/min at 144 g/h 3.

The ratio question dissolves

People argue about 2:1 versus 1:0.8 versus 1:1 as if the ratio were an input. It is not. Stop asking "what is the best ratio?" and ask "how much glucose is enough to saturate SGLT1, and where does the remainder go?" Every credible position — Jeukendrup’s, Rowlands’, Podlogar and Wallis’, and the 2026 Morton review — collapses onto one rule: cap glucose-type carbohydrate at about 60 g/h and put every additional gram in as fructose 114.

The cleanest ratio evidence in the literature comes from two studies that fed equiosmotic, equicaloric drinks and varied only the ratio. At 108 g/h, a fructose:glucose ratio of 0.8 produced 1.14 g/min of exogenous oxidation against 1.04 at 0.5 and 1.05 at 1.25 5. At 90 g/h the same pattern held: 1.10 g/min at 0.8 against 0.92 and 1.04 6. Read the sub-numbers and the mechanism appears — at the fructose-poor ratio, glucose oxidation efficiency was only 65%, and it jumped to 85% once more fructose was present 6. The extra fructose is not just adding a channel; it appears to relieve the glucose channel.

Below 60 g/h, the argument inverts

At intakes up to about 1 g/min, single-source glucose or maltodextrin is what the 2026 review typically recommends 1. There is even a running study in which a glucose-plus-fructose gel at 60 g/h lowered whole-body carbohydrate oxidation by roughly 0.4 g/min during a 1:50 half marathon compared with glucose only 16. That study was designed to look at fluid intake, the conditions differed in self-selected drinking volume as well as in carbohydrate type, and finishing times were comparable across all conditions — so treat it as grade C support for a default that rests on firmer ground elsewhere. The practical upshot: do not blindly push a 2:1 blend onto a 90-minute hard race.

Table sugar is not a compromise

Sucrose is one glucose bonded to one fructose, so by mass it is exactly 50/50, and it is metabolically equivalent to feeding the two sugars separately — peak exogenous oxidation of 1.40 vs 1.29 g/min, p = 0.999 4. It is also the cheapest and lowest-osmolality way to buy fructose. At a 1:1 target, table sugar is the whole recipe. Maltodextrin does a different job: it delivers glucose at roughly one-fifth the osmotic load of the same mass of dextrose 12, which is how a 120 g/h drink stays something your stomach will empty.

Efficiency is not 100%, and the app never pretends it is

Exogenous oxidation efficiency for glucose-fructose blends is 69–75% 68. To oxidise 90 g/h you have to ingest 90–120 g/h 1. Any planner that treats the number you swallow as the number you burn is overstating the fuel by a quarter. One more caveat travels with all of it: the licence to point cycling oxidation data at runners rests on a single matched-intensity comparison, at 90 g/h and about 60% VO₂max, which found 1.25 g/min running against 1.19 cycling (p = 0.13) 11. It says nothing about tolerance at 120 g/h or at marathon race intensity.

The high-carbohydrate era is genuine. It is also being sold well past what it demonstrates. Here is the split between what is established and what is not.

What 120 g/h does

• Keeps carbohydrate the dominant fuel for three hours — it prevents the crossover point that would otherwise shift you toward fat 1.
• Raises exogenous oxidation to roughly 96–105 g/h in trained males 1.
• Lowers the oxygen cost of running in elite male marathoners: 8.1 mL O₂ per kg per km lower at 120 g/h than at 60 g/h (p = 0.039), about 3% 4. Carbohydrate supplied 65%, 51% and 43% of energy at 120, 90 and 60 g/h 4.
• Reduces markers of exercise-induced muscle damage and improves 24-hour neuromuscular recovery after a trail marathon, at 120 g/h versus 90 and 60 56.

What it does not do

There is no demonstrated performance advantage of 120 g/h over 90 g/h. The best dose-response study, n = 51, found diminishing returns above about 78 g/h 3. One trial found 30-minute time-trial power best at 90 g/h (228 W) against 80 (219 W) and 100 (212 W). Another found 74 g/h beat placebo while 111 g/h did not. Podlogar measured higher exogenous oxidation at 120 versus 90 g/h with no endogenous sparing 11. Wilson’s 2026 narrative review concludes there is not clear evidence for 100 g/h and above over 60–90 g/h in single-day events 2.

The ceiling that makes it impossible for slower runners

This is the part almost nobody says out loud. Carbohydrate intake cannot usefully exceed carbohydrate oxidation, and oxidation is bounded by how much energy you are burning per hour and what fraction of it is coming from carbohydrate. Running costs roughly 1.0 kcal per kg per km, and that is close to independent of pace 12. So the arithmetic writes itself.

For the elite, the gut is the limiter, and gut training is exactly the right intervention. For the five-hour runner, the metabolism is the limiter. A 120 g/h prescription would exceed their total carbohydrate oxidation, never mind the exogenous share of it. Training the gut does not raise the number of calories per hour a slower runner burns. Prescribing 120 g/h to that athlete is not aggressive fuelling; it is a plan to carry sugar that has nowhere to go, in a gut that will complain about it.

How the gate is set

The standing consensus ceiling is still 90 g/h, and it is reserved for exercise beyond 2.5 to 3 hours 10. The trials that justify going past it in running had an explicit prerequisite. Urdampilleta 2020 states it plainly: every athlete had undertaken gut training involving more than 90 g/h intakes on more than two days a week for the four weeks before the race, in elite male trail runners with over five years of ultra experience, across a 4.5–5 hour event 6. Any app that hands out 120 g/h on a self-declared dropdown is prescribing the trial’s dose to an athlete with a fraction of the trial’s exposure. Our level-3 gate is at least eight sessions across at least four consecutive weeks at 90 g/h or above, at least 90 minutes each, with GI scores at or below 3 out of 10, and at least two sessions over 2.5 hours.

The gut is trainable. It is not trainable in the way the marketing implies. The human evidence is narrow and very specific, and it is worth knowing exactly what it consists of.

The protocol that worked

Costa 2017 and Miall 2018, from the same lab, ran the same intervention: ten one-hour runs at 60% VO₂max over two weeks, each with 30 g of 2:1 glucose:fructose in a 10% solution every 20 minutes — 90 g/h, about 900 mL/h of fluid 12. That is it. An ordinary easy hour with fuel added, five times a week for two weeks. It adds nutritional load, not training load, which is the single most useful thing to know about it, because it removes the objection that gut training will wreck your training week.

Three things it does not do

1. It does not speed gastric emptying during exercise. In the only repeated-trials running study that measured it, emptying went from 12.1 to 12.3 mL/min between the second and sixth trial while discomfort fell 26% 5. The 2023 systematic review found no change in either study that measured it 3. The correct promise is comfort at volume, not faster transit.
2. It does not armour the gut barrier. Three studies measured I-FABP before and after; none found improvement. In one arm it rose 18% 3. Nobody should be told gut training heals a leaky gut.
3. It has not been shown to raise exogenous oxidation. The 2026 review is explicit: no evidence for increases in exogenous carbohydrate oxidation 13. We therefore treat gut training as a tolerance gate, not a capacity multiplier. It changes what you can carry, not what you can burn.

What does adapt, and on what clock

• Gastric emptying of a specific nutrient: 3–7 days, and nutrient-specific. Three days of 400 g/day of glucose cut the median emptying half-time of a glucose drink from 29.1 to 20.7 minutes and left a protein drink unchanged 9. Three days of fructose sped fructose emptying but not glucose 10. The training stimulus has to be the same sugar mix as race day.
• Symptom habituation: 5–10 exposures. This is the fastest and most reliable outcome, and it is partly perceptual 125.
• Transporter density: animals only. SGLT1 rises 1.9-fold in mice after two weeks on a high-carbohydrate diet 11, with similar findings in piglets and horses. There is no human intestinal biopsy study in athletes. The widely quoted "doubling of SGLT1 in two weeks" is an explicit extrapolation from animals, grade C 6. The best human proxy is 28 days at 8.5 versus 5.3 g/kg/day raising exogenous oxidation in cyclists 8.
• Splanchnic blood flow: not trainable. Perfusion collapses within the first ten minutes of exercise, tracks intensity and heat, and does not care about your training history 12. Intensity and heat are the two dials that decide how hard a given g/h feels on a given day.

The part we are guessing at

No study has ever compared progression rates. Costa and Miall put runners straight onto 90 g/h from session one with no ramp and no dropouts — but their sessions were 60 minutes, so the total session load was 90 g, not 270 g 12. King ramped elites from 60 to 90 g/h over 14 days 7. Practitioner guidance converges on +10–15 g/h every one to two weeks. Our weekly increment, our symptom-gated autoregulation rule, our detraining half-lives and our maintenance frequency are all grade C — constructed, and labelled as constructed in the program itself. Detraining in particular has zero human data. The one piece of that machinery with a published foundation is the symptom scale: our 0–5 logging scale is a compression of a validated 0–10 tool, where 1–4 means the sensation is present but does not interfere with normal activity and 5–9 means it does 16. The scale is grade A; our mapping onto it is not.

The 2025 Sports Dietitians Australia and Ultra Sports Science Foundation joint position statement grades gut training Grade I — good evidence for symptom management, with the caveat that the effect depends on the quantity, quality and duration of exposure and on individual responsiveness 4. That is the right level of confidence: it works, for the thing it works for.

Exercise-associated hyponatremia has an unambiguous consensus pathophysiology: sustained ingestion of hypotonic fluid in excess of sweat and urine losses, combined with non-osmotic vasopressin secretion. Every reported fatality traces to that mechanism, not to sodium deficiency 1. Across seven outdoor activities, 89.2% of documented cases came from running and hiking 10. So fluid intake is the safety-critical variable, and everything else in this section is downstream of that.

Your sweat rate is a measurement, not an estimate

Sweat rate in athletes runs roughly 0.5–2.0 L/h, with about 2% of people above 3.0 L/h 2. Pooled across 1,303 sweat-tested endurance athletes the mean is 1.28 ± 0.57 L/h 3. A validated running-specific prediction equation exists and is good — R² = 0.77 outdoors — but its 95% confidence band is +0.44 / −0.38 L/h 4, which is wider than any prescription difference we would make from it.

The 2% threshold, honestly

The upper bound is your gut, not your stomach

Gastric emptying above 1 L/h is documented — 1,134 mL/h emptied when 1,600 mL/h was ingested during two hours of cycling 9. So the folk figure of "about 900–1,000 mL/h" is a practical running ceiling, not a physiological wall. In running we cap lower than the physiology allows because GI tolerance, mechanical jostling and splanchnic hypoperfusion are the binding constraints, not the stomach. Our defaults: 800–1,000 mL/h, and 800 mL/h when hyponatremia risk factors are present. That cap is grade C — our synthesis.

The rules the app will not break

1. The fluid plan never exceeds your measured or estimated sweat rate. Body water and body mass should not rise during exercise.
2. No plan exceeds 1,000 mL/h in running, or 800 mL/h if hyponatremia risk factors are flagged.
3. Weight gain during a race is a medical red flag, not a hydration success. Any gain from baseline, plus any of nausea, headache, puffiness, confusion, vomiting or disorientation: stop drinking and seek medical attention.
4. In suspected hyponatremia, do not drink more. Oral hypotonic fluid is contraindicated pending medical assessment 1. The treatment is hypertonic saline given by medical personnel — we describe it so you recognise correct care, never so you attempt it.

Thirst is a legitimate guide, and the 2015 consensus says so explicitly, rejecting the older "drink before you are thirsty" advice as having fostered the misconception that thirst is a poor guide and having facilitated inadvertent overdrinking 1. What we ship is a planned band with thirst as the in-band guide and a hard ceiling on top. The honest caveat is that thirst can lag at very high sweat rates, which is exactly where the lower bound of the band earns its place.

What sodium does

• Palatability, and therefore voluntary drinking. The hedonic optimum for a drink sits around 18–40 mmol/L (roughly 414–920 mg Na/L). Above about 50 mmol/L (1,150 mg/L), intake falls 910.
• Plasma sodium maintenance when fluid roughly matches losses. In a cycling trial, a 60 mmol/L beverage held plasma sodium (+0.8 ± 2.4 mmol/L) while a 21 mmol/L beverage let it fall (−1.5 ± 2.2) 5. n = 11, cycling.
• Post-exercise retention. A rehydration drink at 40–60 mmol/L markedly reduces urine output compared with plain water, which is why the recovery target is 125–150% of the deficit with sodium 8.
• Co-transport. Glucose crosses the intestinal wall on a sodium-coupled transporter, so some sodium in the bottle is mechanistically sensible 2.

What sodium does not do

How much you actually lose is less predictable than you think

Whole-body sweat sodium runs about 10–70 mmol/L, which is 230–1,610 mg/L 2. Pooled across 1,303 sweat-tested athletes, the endurance-sport sodium loss rate is 51.7 ± 27.8 mmol/h — roughly 1,190 ± 640 mg/h 3. Look at that standard deviation before you trust any population default. Day-to-day variability in the same person is 11–17% for whole-body washdown, and the same person’s concentration varies 80–120% between body sites 2. Marathon-specific data show the spread plainly: sweat chloride in marathoners averaged 32.2 ± 15.6 mmol/L with a range of 7.3–90.6 12.

Doses, and a units trap worth knowing about

The trap: the NATA position stand recommends "modest amounts of salt, 0.3 to 0.7 g/L", which reads as sodium chloride — about 118–275 mg of sodium per litre 7. The ACSM stand says 0.5–0.7 g/L and explicitly equates it to 20–30 mEq/L, which is sodium the element 6. Those are roughly threefold apart. We follow the ACSM figure, because its units are internally verifiable through the milliequivalent cross-check, and every number this app prints states which species it means.

Kitchen chemistry

Solubility is a non-issue — table salt dissolves to about 359 g/L. Palatability is the real ceiling, at roughly 1,000–1,200 mg Na/L, which is about 300 times below saturation. If you genuinely need more sodium than that, it goes in as a salted food or a capsule, not into the bottle where it will stop you drinking.

Heat changes the sodium number more than it changes the fluid number

Sweat sodium concentration is positively related to sweat rate, and heat raises both. Since total loss is rate times concentration, sodium scales super-proportionally with heat — roughly double at a moderately hot WBGT, against about 1.9× for fluid. Heat acclimation flips the concentration term hard the other way: ten days of acclimation lowers the intercept of the sodium-to-sweat-rate relationship by about 15 mmol/L 11, with older work reporting up to 50% reductions, while sweat rate rises. An acclimatised athlete needs more water and less salt per litre than the same athlete unacclimatised.

Heat runs a cascade, and each leg of it has a number. Heat production exceeds heat loss, core temperature rises, skin blood flow rises to move heat outward, cardiac filling falls, the splanchnic and renal beds vasoconstrict to defend blood pressure, gastric emptying and exogenous carbohydrate oxidation fall while muscle glycogenolysis rises, and you slow down. Those legs point in opposite directions for fluid (up) and carbohydrate (flat or down). Getting that asymmetry right is most of what this part of the engine does.

Humidity does not make you sweat more. It makes your sweat useless.

This is counter-intuitive and it matters. Bright and colleagues held temperature at 33 °C and varied absolute humidity across 1.6, 2.5, 3.5 and 4.5 kPa 2. Whole-body sweat rate was essentially unchanged across all four conditions, at 1.91–2.03 L/h. What collapsed was sweating efficiency — from 0.50 down to 0.16 — as sweat coalesced and dripped instead of evaporating. Power fell 260 W to 222 W, a 16% loss.

So an app that multiplies sweat rate by a large humidity factor at a fixed pace is double-counting. In a self-paced race, pace falls, heat production falls, and sweat production is partly self-limiting. Apply the pace decrement first, then compute sweat.

Computing WBGT properly

Wet bulb globe temperature is what sports medicine has standardised on, and it turns out to be an excellent single predictor of required evaporative heat loss. Outdoors it is 0.7 × natural wet bulb + 0.2 × black globe + 0.1 × dry bulb, so it carries humidity, radiation and wind in one number. We use an analytic Liljegren-style implementation when the weather API gives temperature, humidity, wind and solar or cloud data 10, with the Stull psychrometric wet-bulb equation as a cross-check 9. We also compute the National Weather Service heat index 11, because the best marathon-pace-versus-weather dataset we have is indexed on it 8.

Running partitional calorimetry across a 0–40 °C by 30–90% relative humidity grid, modelled sweat requirement collapsed onto WBGT almost linearly at about 0.06 L/h per °C WBGT for a 70 kg runner at roughly 12 km/h, with little residual scatter from humidity. The method is standard heat-balance physics (grade B); the specific coefficient is our own derivation and is grade C until it is checked against field sweat data.

Carbohydrate in the heat: up or down?

The concentration consequence is the real intervention, and it is a home-brew advantage. Because fluid goes up roughly 1.6–2.2× while carbohydrate stays flat, the drink dilutes itself. Sixty grams per hour in 1.0 L is a 6% drink; the same 60 g in 1.6 L is 3.75% — well inside the comfortable zone, and far from the 12% and 16% solutions the review associated with GI distress 6. You cannot dilute a commercial gel. You can absolutely put the same 60 g of sugar into more water.

Sun and wind are large terms, not decorations

Solar radiation at 800 W/m² roughly halves time to exhaustion at 30 °C compared with no radiation, and airflow substantially rescues it 1. An app that ignores cloud cover and wind is discarding two of the biggest terms in the heat-balance equation. Clear-sky irradiance at the surface peaks near 1,000 W/m² 1; we attenuate it from cloud fraction, and that attenuation curve is grade C.

Altitude

• VO₂max falls about 6% per 1,000 m above roughly 1,500 m 14.
• Acute hypoxia suppresses exogenous glucose oxidation, and the suppression eases after acclimatisation 14. Same conclusion as heat, different mechanism: do not push a higher g/h at altitude on an unacclimatised athlete.
• Respiratory water loss rises — about 1,900 mL/day in men and 850 mL/day in women at 4,300 m — with roughly 500 mL/day extra urinary loss 14.
• Respiratory water loss carries no sodium. Altitude raises fluid need without raising sodium need, so the correct altitude adjustment *lowers* the drink’s sodium concentration. Scaling sodium with fluid at altitude is an error, and it is a common one.

Unlike during-race carbohydrate rates, where the ceiling is still moving and the evidence is dominated by cycling, the peri-workout envelope is covered by a formal joint position stand whose numbers have survived a decade essentially unchanged 1. It was re-stated by a 2022 review 11 and re-affirmed for the loading dose specifically by a 2026 randomised dose-response trial 3.

The load

10–12 g/kg/day for the final 36–48 hours, with training cut to near nothing, for events longer than 90 minutes 1. The classic Scandinavian deplete-then-binge protocol is obsolete for trained athletes — Bussau reached maximal muscle glycogen in 24 hours at 10 g/kg plus complete rest, with no depletion phase at all 2. Jones 2026 fed 6, 8 or 10 g/kg/day over 48 hours and found a clean dose-response (r = 0.71 between intake and muscle glycogen), with 10 g/kg significantly highest 3.

Below 90 minutes, loading buys nothing 4, and the position stand prescribes ordinary fuelling up at 7–12 g/kg per 24 hours instead 1. Our engine reproduces the 90-minute rule from a glycogen-stress calculation rather than hard-coding it, which also catches the awkward cases: a 4:30 marathoner is under-fuelled and running at a lower carbohydrate fraction; a 3:20 half-marathoner genuinely needs a load; a 1:20 half-marathoner does not.

The pre-race meal

• 1–4 g/kg of carbohydrate, 1–4 hours before 1. The field operationalisation everyone uses is "grams per kilogram roughly equals hours available" — that is a practitioner rule, grade C, and it is a good one.
• Fat, fibre and protein come out of the meal as the start line approaches. Under two hours, effectively no fat; under 90 minutes, no protein.
• Fluid: 5–10 mL/kg in the two to four hours before, then stop about 20 minutes out so you can void the excess 1.
• Low residue: under 10 g of fibre a day for one to three days, four at the outside. Longer offers no additional benefit and costs you microbiome diversity.
• Low FODMAP for 24–48 hours reduced exercise GI symptom severity by roughly 50% in runners with a history of symptoms 7. If you have no history, this is not your problem to solve.
• During the race, feed little and often. Frequent ingestion beat a single bolus for glycogen sparing and time to fatigue in runners 14 — though that study is n = 6 and it compared the two extremes. Whether 15, 20 or 25 minutes differs at a matched hourly rate has never been tested, so our feed interval is engineering judgement.

Pre-race sodium loading is off by default

For it: sodium hyperhydration reliably raises plasma volume 3.5–12.6% and lowers exercise heart rate by 3–11 bpm, and in the strongest running study it raised treadmill time to exhaustion by about 26% in the heat 89. Against it: time-to-exhaustion protocols systematically inflate effect sizes relative to time trials; the time-trial results are inconsistent; it adds 0.7–1.5 kg of fluid you then carry; and GI upset is common — 26 of the studies in the systematic review reported symptoms 8. The doses involved are far saltier than anything palatable. Ten millilitres per kilogram at 150 mmol/L for a 70 kg athlete is about 700 mL carrying 2.4 g of sodium, which is roughly 6 g of table salt in a large glass of water.

We enable it only when the event is over two hours, the heat flag is set or you report a high sweat rate, there is no contraindication, and you have rehearsed it at least twice. This is the lowest-confidence recommendation in the whole product, and it is labelled that way in the plan.

Recovery

• 1.0–1.2 g/kg/h of carbohydrate for the first 4 hours when the next hard session is under 8 hours away; 8–10 g/kg over 24 hours otherwise 1.
• 0.25–0.3 g/kg of protein per feeding. Protein co-ingestion only rescues glycogen synthesis when carbohydrate is below about 0.8 g/kg/h 10 — which is exactly the situation of a real athlete with a suppressed appetite after a hard race, so the rule earns its place.
• 125–150% of the fluid deficit, with sodium rather than plain water 1.

Caffeine, sodium bicarbonate, dietary nitrate, creatine monohydrate and beta-alanine all carry International Society of Sports Nutrition position stands, and none of them is on the 2026 WADA Prohibited List — caffeine sits on the Monitoring Program, which is not the same thing as banned 8. That is where the similarity ends.

Caffeine

3–6 mg/kg, about 60 minutes before the start 1. Peak plasma from anhydrous capsules comes at 84–120 minutes; chewing gum peaks at 44–80 minutes and so goes in 5–15 minutes before. Nine mg/kg gives no additional benefit and a high incidence of side effects. Half-life is around 5 hours, with a wide 2.5–10 hour spread, which is what governs re-dosing in an ultra. Our split-dose schedule for events over four hours is practitioner-derived — no trial establishes it, grade C, and it is the largest evidence gap in this module.

Sodium bicarbonate, and its sodium bill

The dose is 0.2–0.3 g/kg, 60–180 minutes before the start 2. The effect window is exercise lasting 30 seconds to about 12 minutes — so 1500 m through 5 km. Reported gains: 1.1–2.9% in 400–800 m running; about 2% in a 4 km cycling time trial; and nothing at all in a 30-second all-out effort (d = 0.02, 95% CI −0.07 to 0.11) 2. A prolonged running trial found no effect on time to exhaustion, with every dropout caused by bicarbonate GI symptoms 11.

The one argument for bicarbonate in a long race is a decisive sprint finish or a decisive climb. The supporting evidence is a roughly 3% gain in mean power during a 90-second sprint after three hours of cycling 2. It has never been tested in runners, where the GI cost across three hours is far higher. We route it as optional with a low-confidence label, never as a default. It also requires the most rehearsal of anything here — at least three sessions, because its GI response varies within the same athlete across exposures 9. Individual time-to-peak bicarbonate is similarly variable: at 0.3 g/kg the mean is 87–89 minutes with an individual range spanning roughly 10 to 180 minutes, which is why individualised timing outperformed a fixed schedule in elite runners 10.

Dietary nitrate

6–8 mmol of nitrate, 120–180 minutes before, or better, loaded over 3–6 days 36. The 2025 umbrella review of 20 systematic reviews is unusually clean about where it works and where it does not.

A road race is a closed-loop time trial. That is the one domain where nitrate is null. Two further constraints: above roughly 65 mL/kg/min of VO₂max the benefit approaches zero 36, and antibacterial mouthwash abolishes the effect entirely, because oral bacteria are a required step in converting nitrate to nitrite 1213. Home sources work — beetroot carries about 2.5 mg of nitrate per gram fresh (range 1.1–15), rocket 2.5–4.5 mg/g 6. Beetroot powder density is unknown and label-dependent, so we refuse to compute a dose from it rather than guessing.

The two that are usually wrong for marathon runners

Creatine monohydrate. It is a training-phase and recovery supplement, and runners keep miscasting it as a race aid. A 2023 meta-analysis and a 2025 network meta-analysis in trained endurance populations find trivial to slightly negative effects on endurance outcomes 1416. It costs 1–2 kg of body mass through intracellular water retention of 0.5–1.0 L during loading 4, and in a weight-bearing sport every extra kilogram costs about 1 kcal per km. It also corrupts body-mass-based hydration tracking, so you have to re-baseline after loading. Where it does earn a place: recovery from damage. After a 30 km race, creatine users showed creatine kinase 19% lower, prostaglandin E₂ 61% lower and TNF-α 34% lower, with 10% better isokinetic and 21% better isometric strength during recovery from muscle damage 4. That is a build-phase and post-race argument, not a start-line one.

Beta-alanine. 4–6 g/day over 4–12 weeks raises muscle carnosine by 20–30% at two weeks, 40–60% at four, and up to 80% at ten 5. The median performance gain is 2.85% at 179 g cumulative — in efforts lasting 1 to 10 minutes, best at 4 to 10 515. Above about 25 minutes there is no demonstrated benefit at all 5. It is a genuinely good 3 km and 5 km supplement. It is not a marathon supplement, and its presence in marathon stacks is a marketing artefact. Practical note: single non-sustained-release doses above 800 mg cause paresthesia, so split to 1.6 g or less.

Osmolality is about particle count, not sugar mass

A gram of dextrose contributes roughly eight times the osmoles of a gram of DE-12 maltodextrin, and about 1.7 times a gram of sucrose. That single fact is the whole reason maltodextrin exists in sports nutrition. Vist and Maughan measured it directly, with a 600 mL test drink 1.

Two lessons. Maltodextrin buys you a concentrated drink at a dilute drink’s osmolality — 188 g/L of polymer and 40 g/L of glucose sit at essentially the same osmolality. And Vist and Maughan’s own conclusion was that carbohydrate content influences emptying more than osmolality does, so lowering osmolality is a real but secondary lever. We do not promise that a low-osmolality 120 g/h drink empties like water.

Emptying is only the first gate. Once fluid reaches the intestine, what matters for net water uptake is the number of transportable solutes as much as the tonicity: multiple-substrate solutions produce greater water absorption than single-substrate solutions at the same osmolality 2, and hypotonic beverages absorb water faster than hypertonic ones during exercise 5. That is a second, independent argument for a glucose-plus-fructose bottle in the heat, separate from the oxidation argument in section 2. The maltodextrin side of the model is driven by average molecular weight, which is fixed by the dextrose equivalent on the bag — the relationship that lets us compute osmolality from a DE number is established by osmometry 11.

For calibration: a typical commercial 40 g gel delivering 22 g of maltodextrin-fructose in about 18 g of water computes to roughly 3,500 mOsm/kg. Concentrated is not in itself dangerous. Concentrated and unaccompanied by water is. The advantage of a syrup flask over a gel is only that the water pairing becomes explicit instead of implicit.

The 6–8% rule is a proxy, and it is out of date

The classic 4–8% w/v range came out of the glucose and sucrose era, where 8% sucrose lands right at isotonic. The percentage was standing in for osmolality. With a maltodextrin and fructose blend you can run 9–10% and still be isotonic; with dextrose and fructose you are hypertonic at 5% 36. Code the osmolality, not the percentage. This is also why a drink whose carbohydrate and fluid targets are set independently and then divided becomes most hypertonic exactly when you drink least — cold weather, low sweat rate — which is precisely backwards.

Weighing things

• Table sugar is 99.8 g sucrose per 100 g, and sucrose is exactly 50/50 glucose and fructose by mass 7.
• Dextrose monohydrate is 0.909 g of glucose per gram of powder (180.16 / 198.17) — the rest is water of crystallisation. A recipe calling for 60 g of carbohydrate needs 66 g of powder. Ignoring this under-doses you by 9%.
• Maltodextrin is about 0.95 g of carbohydrate per gram, the balance being moisture.
• Sodium and bicarbonate doses live in the 0.5–20 g range, where a level teaspoon is roughly ±8% and a heaped teaspoon roughly ±30%. At a bicarbonate dose of 0.2–0.3 g/kg 12 that error is measured in grams of sodium, not milligrams. A 0.1 g scale is a hard requirement, and a pre-mixed brine stock is the accuracy workaround for small salt doses.

Cost

The saving is 4–12× against premium drink mix and 10–100× against gels. What you give up is real and worth stating: packaging convenience, flavour engineering, hydrogel encapsulation (whose benefit is itself contested), third-party batch testing for banned substances, and the ability to grab something at an expo. If you are drug-tested, the batch-testing point is not a small one — though the counter-argument is that single-ingredient food-grade materials sidestep the 9–15% undeclared-contamination rate found across tested commercial supplements.

Teeth

Enamel demineralises below a critical pH of about 5.5, and frank erosion accelerates below about 4.0–4.5. Commercial sports drinks typically sit at pH 2.4–3.5 with high titratable acidity — citric acid is triprotic and chelates calcium, so it keeps dissolving enamel long after the initial pH hit. Surveys of elite athletes find roughly 49% with caries requiring intervention and 42% with erosion 9. Exercise makes it worse: it reduces salivary flow and buffering, and mouth-breathing dries the front teeth, which are the exact surfaces that erode.

Racing a number you have never rehearsed

This is the product’s name and it is also the single highest-yield rule in the literature. Gastrointestinal symptoms, not absorption, are the binding constraint for most people, and a prior history of GI distress is the strongest predictor of recurrence — upper and lower symptom scores correlated with reported history at r = 0.37 and 0.51 4. Our engine refuses to put a rate, an ingredient or a timing into a race plan that you have not logged in training.

Prescribing by distance instead of duration

A 3:00 marathon and a 5:30 marathon are different physiological events. See section 1.

Drinking to a schedule instead of to your sweat rate

A fixed hourly drinking schedule is, in the 2015 consensus framing, prescribing the exact behaviour that causes hyponatremia 1. Fluid intake must never exceed sweat loss, and body mass must not rise during exercise.

Treating salt as protection against overdrinking

It is not, and the consensus says so at grade 1C 1. Every drinkable sodium concentration is hypotonic against serum.

Letting the drink’s concentration be an accident

If you set a carbohydrate target and a fluid target independently and then divide, the concentration is an emergent property that nobody is watching. It becomes most hypertonic exactly when you drink least. We have measured this in a shipping competitor product: 120 g/h against 950 mL/h works out to 12.6% w/v, roughly 444 mOsm/L in the bottle, at which point the classic 4–8% guidance and the gastric-emptying literature are both being ignored silently.

Adding carbohydrate because it is hot

Heat raises how much carbohydrate you burn and lowers how much of what you swallow you can absorb. The extra comes from stored glycogen. Raising grams per hour into a gut with a quarter less emptying capacity is how a hot race becomes a DNF. See section 7.

Not counting bicarbonate’s sodium

Twenty-one grams of baking soda carries about 5.7 g of sodium 3. Either it goes in the ledger or the plan is wrong. The opposite failure exists too, and we have seen it shipped: a product that zeroes in-race sodium entirely when bicarbonate is enabled, then displays a sodium overshoot warning. Buffering sodium taken two hours before the gun and hydration sodium taken during the race are not interchangeable and are not even in the same time window.

Trialling two things in one session

Bicarbonate and a maximum-rate carbohydrate protocol are both osmotic loads on the same gut. Change one variable per session or you learn nothing from a bad one.

Under-dosing dextrose by ignoring its water

Dextrose monohydrate is about 9% water by mass. Weighing out "60 g of carbs" as 60 g of powder gives you 54.5 g. Small, systematic, and entirely avoidable.

Mixing bicarbonate with an acidic flavouring

One gram of citric acid liberates roughly 0.39 L of carbon dioxide from 1.31 g of bicarbonate. In a sealed flask that is a mess; in your stomach it is worse. Sodium citrate is the reaction product and does the same job without the fizz.

Fuelling every easy session at race rate

Fuel for the work required. Chronic over-fuelling of low-intensity sessions can blunt mitochondrial and fat-oxidation adaptations 2. Gut training should be targeted at key long and race-pace sessions, not smeared across the week.

Reading a lactate number wrong after fructose

Fructose is partly converted to lactate in the intestinal wall and liver before being oxidised, so blood lactate rises with fructose feeding. That is a substrate artefact, not a change in your intensity domain 2. If a device or a coach reads it as one, the session gets misprescribed.

Declaring your gut tolerance from a dropdown

Self-declaration is not evidence. Picking "high carb machine" from a menu and receiving a 120 g/h prescription is how an untested athlete ends up with a trial-grade dose. The honest version of that number is a demonstrated ceiling with a date, a duration, a pace, a concentration and a symptom score attached to it.

Every number on this site carries a grade. A grade of C means the number is a decision we made rather than a finding we found. Here are the places where the honest answer is that nobody knows yet, ordered roughly by how much they change what we tell you.

Does more than 90 g/h make a runner faster?

There is no performance trial of doses above 90 g/h in running, at any distance 1. Everything above 90 g/h in runners measures metabolism or recovery. We therefore label the 90–120 g/h tier as targeting durability and recovery, and refuse to sell it as a personal-best lever.

Any of this in women

There are no female data at 90 g/h or above in running. Every absolute oxidation rate cited on this page comes from male cohorts, and the 2026 review names female data as a major research priority 1. There is an important counterpoint that keeps this from becoming a reason to under-fuel women: when carbohydrate is ingested during exercise, sex differences in substrate metabolism disappear and both sexes oxidise ingested carbohydrate at the same rate. So there is no positive reason to lower the target — only an absence of confirmatory data at the top of the range. We show the caveat rather than quietly scaling the number.

What gut training actually changes

Breath hydrogen fell 45–54% in the two core trials, which is consistent with genuine absorptive improvement 2. But no human study has measured SGLT1 protein in an athlete’s intestine, so we cannot separate real transporter upregulation from symptom habituation plus a smaller residual osmotic load. This is the field’s central open question and this app’s most load-bearing extrapolation.

How fast gut adaptation is lost

Completely unstudied in humans. Our detraining half-lives, our 21-day regression threshold and our one-session-per-week maintenance rule are constructed values. We label them as guesses in the program itself, because they are.

Whether the high-carbohydrate case survives the heat

Heat shifts substrate use toward carbohydrate, which argues for more, while worsening splanchnic perfusion and gastric emptying, which argues for less. No trial has tested "more carbs in heat" against "same carbs in heat" for running performance with GI endpoints 3. That is the single highest-value gap for this product, and until it closes we hold grams per hour flat.

The rest of the list

• The true water-to-glycogen ratio in a trained runner. Published values span 1:1.6 to 1:3.8, and the most recent controlled trial measured no mass change at all 4. That spread is embarrassing for a number that drives our weight-gain warning.
• The exact slope of sweat sodium against sweat rate. We could not open the primary regression. Our heat coefficient for sodium concentration is a placeholder, and it directly sets salt dosing.
• Sweat sodium heritability. Commercial sweat-testing companies say it is largely genetic. We found no published heritability coefficient, and diet and heat acclimation demonstrably move it.
• Altitude between 1,000 and 3,500 m for runners. Almost all quantified altitude nutrition data comes from mountaineering above 4,000 m.
• Male thresholds for low energy availability. Coding 30 kcal per kg of fat-free mass for men is defensible as a conservative safety trigger and indefensible as physiology 5.
• Whether gut training changes bicarbonate tolerance. Plausible, untested, and directly relevant to a product that does both.
• Carbohydrate dose for sub-two-hour events at genuinely low intensities — a three-hour half marathon, for instance. That describes a large share of the people likely to use this app, and there are no data.
• Beetroot powder nitrate density. Unknown and label-dependent, which is why we refuse to compute a dose from it.
• Whether the humidity finding holds in running. The study that shows humidity does not raise sweat rate was cycling with a pedestal fan, and runners generate airflow differently. That finding is central to why we do not apply a separate humidity multiplier.