Born in January, picked in Doha? Relative age at the 48-team U-17 World Cup

On Wednesday the FIFA U-17 World Cup kicks off in Doha for the second year running, and once again the Aspire Zone becomes the centre of world youth football: 48 national teams, 104 matches between 19 November and 13 December, and up to eight games a day during the group stage. On 11 November FIFA published the official squad lists for all 48 teams, over 1,000 players in total, each with a date of birth. For anyone who studies talent identification, that is one of the largest, cleanest and fully public relative age datasets in youth sport, being played out a few kilometres from my office.

So I asked two questions. First, does doubling the tournament from 24 to 48 teams dilute the classic birth-quartile skew, or simply spread it to more federations? Second, which nations and positions break the pattern? And then a third question, the one I think matters most for practitioners: if these squads are the most skewed cohort in the pathway, how much does that skew actually tell us about who will become a senior international?

A quick refresher on relative age

Youth competitions group players by a fixed cut-off date. For this tournament that means players born on or after 1 January 2009. A player born in January is almost a full year older than a teammate born in December, and at 15–17 years of age that year can mean several centimetres of height, kilograms of muscle mass and a very different stage of maturation. Coaches selecting for today’s match tend to pick the bigger, faster, more mature players, and those are disproportionately the ones born early in the selection year. The result is the relative age effect (RAE): an over-representation of players born in the first quarter (Q1, January–March) and an under-representation of those born in the last quarter (Q4, October–December).

It is one of the most replicated findings in sport science, but the magnitude is not fixed. It varies by age group, sex, position, country and level of competition, and that variation is where the interesting questions are.

What 1,000-plus birthdates show

I extracted date of birth and playing position for every player in the official FIFA squad list (48 teams × 21 players = 1,008), assigned each to a birth quarter and compared the observed distribution with what we would expect if birth month were irrelevant to selection (an even spread, weighted for the number of days in each quarter). For comparison I did the same for the 1,008 players at the 2025 edition.

The result is about as clear as relative age data get. In 2026, 40.5% of players were born in January–March and only 14.2% in October–December (Q2 25.8%, Q3 19.5%). A Q1 player is almost three times as likely to be in a squad as a Q4 player: the Q1:Q4 odds ratio is 2.9 (95% CI 2.4–3.5; χ² = 163, p < 0.001; Cramér’s V = 0.23, a small-to-moderate effect).

Bar chart of birth quarters for all U-17 World Cup squad players in 2025 and 2026: about 40% born in January–March, 26–27% April–June, 18–20% July–September and 14–15% October–December, against an expected 25% per quarter.

There is a second, cruder layer of relative age hiding in these squads. The tournament is open to anyone born in 2009 or later, so in principle 15- and 16-year-olds could compete for places. In practice 80% of the 2026 players were born in 2009, the oldest eligible year (84% of the 2025 players were born in 2008). So the selection funnel favours the oldest year of the window, and then the oldest months within that year.

By confederation and by position

Previous work at this level suggests we should not expect a uniform picture. When Romann and Fuchslocher analysed 672 players at the FIFA U-17 Women’s World Cup, they found significant RAEs in the European and North/Central American zones, none in Asia, Oceania and South America, and significant inverse RAEs among African players. The usual suspects for these differences are the structure of national pathways, how early formal selection begins, school-year versus calendar-year grouping, and the reliability of birth registration in some countries.

Pooling both editions to give each confederation a reasonable sample, the gradient is striking. UEFA squads show the strongest effect: 47% of European players were born in Q1 and fewer than 9% in Q4, an odds ratio of 5.6 (4.0–7.8). Oceania (4.2), South America (4.0), CONCACAF (3.0) and Asia (2.7) follow. African squads are the exception: their distribution is close to flat (Q1 30.7%, Q4 25.0%; odds ratio 1.2, 95% CI 1.0–1.6), and in neither edition on its own does the African skew reach statistical significance.

Six small bar charts of birth-quarter shares by confederation, 2025 and 2026 pooled. Q1:Q4 odds ratios: UEFA 5.6, OFC 4.2, CONMEBOL 4.0, CONCACAF 3.0, AFC 2.7, CAF 1.2.

I would be careful about reading the African result as “no relative age bias”. It is consistent with Romann and Fuchslocher’s findings, but birth registration is part of the story: 13 of the 420 African players (3.1%) have a registered birthday of 1 January or 31 December, roughly six times what chance would predict, a possible marker of estimated or administratively assigned birthdates. When birthdates are partly administrative, the birth-quarter signal is diluted. Maturation-related selection may still be operating; we just can’t see it through the calendar.

At team level, the most skewed 2026 squads are the USA (15 of 21 players born in Q1, none in Q4), Italy (14 Q1, one Q4), China, France and Denmark (13, 12 and 12 Q1 players respectively, none in Q4). At the other end, Côte d’Ivoire, Senegal, Uganda and Tanzania have as many or more Q4 than Q1 players. Japan is a neat case study: its squad has more April–June than January–March births, which is what you would expect in a system where the school year, and much youth football, starts on 1 April.

Position matters less than I expected. In Italian Serie A rosters, the Q1 over-representation was strongest among goalkeepers (odds ratio 4.4), and goalkeeping is the position where height counts most and where selection decisions are often made earliest. Pooled across both tournaments, goalkeepers do have the highest odds ratio (3.4, 95% CI 2.3–4.9), but defenders (2.9), midfielders (2.8) and forwards (2.6) are not far behind and the confidence intervals overlap substantially. The goalkeeper effect was large in 2025 (4.9) and unremarkable in 2026 (2.6), which is a good reminder that 144 goalkeepers per edition is not a big sample. At this level, relative age is a squad-wide filter, not a positional quirk.

Forest plot of Q1:Q4 odds ratios by playing position for 2025 and 2026. Pooled: goalkeepers 3.4, defenders 2.9, midfielders 2.8, forwards 2.6; all confidence intervals well above 1.

2025 vs 2026: is the skew stable?

Qatar hosts this tournament every year until 2029, which is a rare opportunity: five consecutive editions of the same 48-team format, at the same venues, with public squad lists. Comparing the 2025 and 2026 cohorts tells us whether the RAE at this level is a stable structural feature or something that fluctuates with the particular group of federations that qualify.

The answer, so far, is that it is remarkably stable. The 2025 squads had 40.3% Q1 and 14.9% Q4 births (odds ratio 2.8, 95% CI 2.3–3.3); the 2026 squads 40.5% and 14.2% (2.9, 2.4–3.5). That is despite roughly a third of the qualified nations changing between editions. Individual confederations move around more (UEFA 6.3 → 5.0, CONCACAF 2.3 → 3.8, Oceania 6.7 → 2.8, the last on only 63 players per year), but the overall picture, and the European-high, African-low ordering, does not. That is what a structural effect looks like: it comes from how youth football everywhere is organised, not from which countries happened to qualify.

So, to the opening question: doubling the field has not diluted the skew. The expanded format simply exposes the same filter in more federations.

Why the U-17 skew overstates its own importance

Here is the part I want coaches and federations to take away. A strong RAE in a U-17 squad tells us a lot about how players are selected at 16. It tells us much less about who becomes a senior international.

In our analysis of 885 players selected for Italian youth and senior national teams between 2000 and 2021, the Q1:Q4 odds ratio was around 9 at U16 and fell to about 1.7 in the senior team. Looked at retrospectively, only a small minority of youth internationals (less than 10–20%) went on to be selected for the senior team, and most senior players had never been selected for youth national teams. The RAE was also smaller among players who successfully transitioned. In other words, the birthdate filter is strongest exactly where it is least predictive.

A recent longitudinal study by Nisbet and colleagues in Biology of Sport, covering 1,518 male and 487 female players in a national programme, adds nuance. Senior teams still showed an RAE (odds ratios of 2.27 for men and 1.50 for women), and youth-to-senior transition rates were 20% for males and 30% for females. Strikingly, males born in Q1 had the lowest transition rate of any quartile (17%). What predicted transition was opportunity: each additional youth-team appearance increased the likelihood of progressing to the senior team by 19.4%.

The Serie A data point the same way from the other end. Q1 players are over-represented in professional rosters, yet once players are inside the professional environment there are no differences between birth quartiles in appearances, playing time or consistency of selection. The bias operates at the selection gate, not in the coach’s weekly team sheet.

We have seen similar patterns outside football. In world-class middle- and long-distance runners, only around one in five top U18 athletes reached the senior top 50, and in sprint swimming the early-success athletes tended to plateau around 20 years of age while later developers kept improving. Success at 16 is a weak signal, and youth selection systems that treat it as a strong one will keep discarding late-born and late-maturing players who would have been competitive at 23.

Practical takeaways for federations selecting U-17 squads

  • Audit your own birth distribution every cycle. It takes an afternoon with a spreadsheet, and the numbers start better conversations with selectors than any lecture on RAE.
  • Measure maturation, not just age. Simple non-invasive estimates of maturity status let selectors see whether a player’s dominance is talent or timing, and bio-banded training and games can reveal what late developers can do when they are not physically outmatched.
  • Protect playing opportunities. If youth appearances predict transition, then rotating minutes and keeping a wider pool of late-born players involved is not charity; it is better talent management.
  • Don’t treat a youth cap as a filter for the senior team. Most senior internationals were never youth internationals. Keep the door open and track players who were not selected.
  • Changing the cut-off date is not a fix. Moving from calendar year to school year simply relocates the advantaged birth months, as Japan’s April-heavy squad illustrates. The effect is structural, so the solution has to be in how selection decisions are made.

What’s next

With five consecutive editions in Doha, the obvious follow-up is longitudinal: track the 2025 squads into senior national-team caps over the next few years and test directly whether birth quarter at U-17 predicts senior selection. I’ll come back to that as the data accumulate. In the meantime, enjoy the football at Aspire, and when you watch the squads line up, have a look at who was born in January and who was born in December.

Methods note

2026 data: FIFA’s official squad list PDF (all 48 teams, 1,008 players; a sample of teams was re-extracted independently and matched exactly). 2025 data: the squad tables on Wikipedia’s 2025 FIFA U-17 World Cup squads page (48 teams, 1,008 players), a secondary source. Expected quarter frequencies are weighted by days per quarter; I did not adjust for national birth-seasonality, which varies modestly between countries. Odds ratios compare observed Q1:Q4 counts with the expected ratio, with Wald 95% confidence intervals. Positions are as listed by FIFA (GK, DF, MF, FW).

References

  • Boccia G, Brustio PR, Rinaldi R, Romagnoli R, Cardinale M, Piacentini MF (2023). Junior to senior transition pathway in Italian football: the rocky road to the top is not determined by youth national team’s selections. PLoS One 18(7): e0288594. doi:10.1371/journal.pone.0288594
  • Brustio PR, Stival M, Cardinale M, Mulasso A, Rainoldi A, Boccia G (2024). Performance pathways in elite middle- and long-distance track and field athletes: the influence of a successful youth. Journal of Science and Medicine in Sport 27(9): 654–659. doi:10.1016/j.jsams.2024.05.007
  • Brustio PR, Cardinale M, Lupo C, Boccia G (2022). Don’t throw the baby out with the bathwater: talent in swimming sprinting events might be hidden at early age. International Journal of Sports Physiology and Performance 17(11): 1550–1557. doi:10.1123/ijspp.2021-0530
  • Brustio PR, Ungureanu AN, Beratto L, Li Volsi D, Lupo C (2026). Relative age selection bias does not translate into a playing-time advantage: evidence from Italian Serie A football. Frontiers in Psychology 16: 1733277. doi:10.3389/fpsyg.2025.1733277
  • Nisbet D, Mandorino M, Zmijewski P, Modric T, Teixeira J, Moreira A, Morgans R (2026). The relative age effect and transition rates across a national soccer program in male and female youth to senior players: a longitudinal analysis. Biology of Sport 43(1): 439–448. doi:10.5114/biolsport.2026.154941
  • Romann M, Fuchslocher J (2013). Influences of player nationality, playing position, and height on relative age effects at women’s under-17 FIFA World Cup. Journal of Sports Sciences 31(1): 32–40. doi:10.1080/02640414.2012.718442
  • FIFA (11 November 2026). The FIFA U-17 World Cup Qatar 2026 official squad lists confirmed. inside.fifa.com

Data sources: FIFA official squad lists, U-17 World Cup Qatar 2026; 2025 FIFA U-17 World Cup squads (Wikipedia).

Heat Acclimation Won’t Fix Your Fuelling

Heat acclimation makes you a better thermoregulator, but it does not make you better at burning the carbohydrate you drink. A study published in The Journal of Physiology this week shows exogenous glucose oxidation drops by about 27% at 40°C, and stays down after nine days of heat acclimation.

The timing matters. Elite triathlon is in the Gulf right now: WTCS Abu Dhabi closed the Championship Series era on 13 November, the Dubai T100 follows today, and the Qatar T100 World Championship Final comes to Doha on 11 December.

The standard playbook for hot races is: acclimate, then fuel as you would anywhere else. This paper says the second half of that sentence needs rewriting.

What the study did and found

Tim Podlogar, Gareth Wallis and colleagues at Birmingham, Exeter and USARIEM tested 17 endurance-trained adults (10 men, 7 women). Each cycled for 90 minutes at a moderate 45% of maximal power while drinking glucose at 90 g/h.

They did this three times: at 20°C, at 40°C, and at 40°C again after nine days of isothermic heat acclimation spread over ten days. Relative humidity was 20% throughout. Dual stable-isotope tracers, indirect calorimetry and muscle biopsies tracked where the glucose went.

Dot plot of exogenous glucose oxidation: 0.51 g/min at 20°C, 0.37 g/min at 40°C, and 0.38 g/min at 40°C after nine days of heat acclimation, with 95% confidence intervals.
Data: Podlogar et al., J Physiol, 2026 (doi:10.1113/JP289304). 17 endurance-trained adults, 90 min cycling, 90 g/h glucose.

The acclimation worked on the usual markers: the authors report clear thermoregulatory adaptation. Yet the oxidation of ingested glucose did not move back towards the temperate value.

The mechanism is the interesting part. Glucose still entered the blood from the gut at the same rate in every condition. What fell in the heat was glucose clearance and the oxidation of plasma glucose. The bottleneck is not absorption; it is uptake and use by the working tissues.

It is not a dehydration story

The obvious objection is that heat studies often confound temperature with fluid loss. Two strands of evidence argue against that explanation.

First, a Loughborough study last year kept runners euhydrated. Ten trained runners ran 100 minutes at about 65% of VO₂max, at 34°C or 19°C, replacing about 90% of sweat losses and drinking 60 g/h of glucose. Average exogenous oxidation was still about 20% lower in the heat, and the body leaned harder on its own carbohydrate stores to make up the difference.

Second, the new study points to a peripheral limit. Gut-to-blood delivery was unchanged, so the effect sits downstream, in how muscle takes up and oxidises circulating glucose. Two labs, two exercise modes, two temperatures: the direction and rough size of the effect (20–27%) agree.

Why it matters: more grams in is not more grams used

The practical consequence is a larger draw on glycogen for the same intake. In the new study the deficit was about 0.14 g/min, roughly 8 g of ingested glucose per hour that is not oxidised.

That sounds small, but it compounds. As a rough extrapolation, not a measured result, a T100 lasting around 3.5 hours at that deficit means about 30 g less fuel from the bottle. In the Loughborough study data, endogenous carbohydrate use rose by about 0.24 g/min in the heat, which over a long race is a meaningful extra withdrawal from muscle and liver glycogen.

This collides with the current push towards 100–120 g/h intakes. If oxidation is capped lower in the heat, pushing intake harder may simply load the gut, exactly when heat is already making gastrointestinal problems more likely.

Three caveats keep this honest:

  • Glucose only. The new study used glucose alone. An older Birmingham study found a glucose–fructose mix oxidised about 36% faster than glucose in the last hour of cycling at about 32°C. Whether multiple transportable carbohydrates escape the heat penalty is the obvious next experiment.
  • Moderate intensity. 45% of maximal power is well below race pace. At race intensity, carbohydrate demand and heat production are both higher.
  • Performance is not oxidation. A 2026 systematic review found carbohydrate’s benefit in the heat equivocal: five of nine trials showed gains, four did not, and seven of nine never measured gut symptoms.

What the T100 race data shows

If heat erodes fuel availability late in a race, run splits should suffer more in hot races than in cool ones. I tested that on every T100 result from 2024 to 2026 on PTO Stats: 37 race-days and 555 finishes by 91 athletes who raced at least three times.

Each split is expressed against that athlete’s own median, so +5% means five per cent slower than their usual T100 run. Heat is the simplified wet-bulb globe temperature (sWBGT) at the day’s peak, calculated from hourly temperature and dew point in the Open-Meteo historical weather archive.

Scatter plot of 37 T100 race-days: median run split relative to each athlete's usual against peak race-day simplified WBGT. Most races sit near zero; the five Singapore races, the hottest, are 5 to 10 percent slower. Doha 2025 is near zero.
Data: PTO Stats, T100 2024–2026 (37 race-days, 555 finishes, 91 athletes with 3+ races). Weather: Open-Meteo archive; sWBGT = 0.567T + 0.393e + 3.94, daily peak 07:00–18:00 local. Dubai 2025 men excluded (shortened run).

Most race-days sit within about 5% of zero whatever the weather. The exception is Singapore: all five editions, men and women, ran 5–10% slower than usual, roughly three to six minutes on an hour-long run.

Across all races, each extra degree of sWBGT was associated with a 0.48% slower run (95% CI 0.19 to 0.76; athlete fixed effects, errors clustered by race). The bike showed no such link (−0.29% per degree, CI −0.70 to 0.12).

That slope is really a Singapore effect. Remove the five Singapore races and the run slope falls to zero (−0.06% per degree, CI −0.30 to 0.17). Between roughly 18 and 29°C sWBGT, race-day heat explains very little of how elite athletes run.

The bike works as an internal control. If Singapore were simply a slow venue or a bad day, the bike should slow too.

Dot plot with 95% confidence intervals of run and bike splits relative to each athlete's usual, by heat band. Cool: run 0.0%, bike +1.0%. Warm: run -0.6%, bike -1.7%. Hot: run +4.2%, bike 0.0%.
Same dataset. Race-day bands by peak sWBGT; medians of athlete-relative splits, 95% CIs from 2,000 bootstrap resamples of races.

It didn’t. In the hot band the median bike split matched athletes’ usual, while the run was about 4% slower. The interval for the run crosses zero because Miami, also hot, showed no penalty. A run-specific slowdown confined to the hottest, most humid races fits a heat-and-fuel story, but it cannot prove one.

What this means for Doha. Last year’s final sat in the warm band (sWBGT 26.8°C), and its run splits were normal: +0.7% for women and −0.2% for men against each athlete’s usual. On that evidence, December in Doha does not trigger the run penalty seen in Singapore.

One limitation to state up front: race data cannot separate fuel from thermoregulation, pacing or tactics. A steeper fade in the heat is consistent with this mechanism; it does not prove it. December in Doha is also typically warm rather than extreme, well short of the 40°C used in the lab (hopefully as I will be in the race too, injuries permitting!), so the effect at the final may be smaller.

Three further caveats on the data:

  • Singapore is one venue. Its course, its start time and its five editions all travel together, so a slow run course could mimic a heat effect. The normal bike splits there argue against a general slow-venue explanation, but not against a slow run course.
  • Weather is approximate. sWBGT uses the day’s peak temperature and humidity, not conditions at race time, and ignores sun and wind.
  • One race excluded. Dubai 2025 men’s run splits were around 50–55 minutes, far quicker than any other edition, which suggests a shortened course; it is left out.

What practitioners should do before Doha

Keep heat acclimating; just stop expecting it to solve fuelling. A practical checklist:

  1. Acclimate anyway. Lower core temperature, better sweating and a lower heart rate are real gains for performance and safety. The new study confirms they happen; it shows they are not the whole story.
  2. Start with full glycogen stores. If more of the race is fuelled from the inside, the tank you start with matters more. Carbohydrate loading deserves attention before hot races, not only long ones.
  3. Prefer multiple transportable carbohydrates. Glucose–fructose blends have the best evidence for higher oxidation in the heat, though they have not yet been tested against heat acclimation.
  4. Gut-train in the heat, not just at home. Rehearse race intake in hot sessions and record gut symptoms. Tolerance in a cool training block does not transfer automatically.
  5. Don’t chase the highest number on the label. If oxidation is capped lower, extra grams mostly add gut risk. Find the highest intake an athlete tolerates in the heat, then hold it.
  6. Use cooling to protect pace. Pre-cooling, ice and cold fluids reduce heat strain directly; they may also buy back some of the metabolic penalty, though that is untested.
  7. Pace the first half conservatively. A heat-driven increase in glycogen use argues for restraint early, so there is something left for the run.

The bottom line

Heat acclimation fixes how hot you get, not how much of your drink you burn. In the best mechanistic data we have, ingested glucose oxidation runs 20–27% lower in the heat, regardless of hydration and regardless of nine days of acclimation. For athletes racing in the Gulf this winter, that shifts the emphasis towards starting glycogen, carbohydrate type, gut training and pacing, rather than simply drinking more.

The open question I most want answered: does a glucose–fructose blend, in heat-acclimated athletes at race intensity, close the gap?

References

  • Podlogar T, Gifford RE, Thomas J, et al. Heat acclimation does not alleviate the reduction in exogenous glucose oxidation that occurs when exercising in hot conditions. J Physiol. 2026. doi:10.1113/JP289304
  • Mougin L, Horner M, Edwards D, et al. Heat stress impairs exogenous carbohydrate oxidation during prolonged running when maintaining euhydration. J Appl Physiol. 2025;139(6):1436–1446. doi:10.1152/japplphysiol.00873.2025
  • Jentjens RLPG, Underwood K, Achten J, et al. Exogenous carbohydrate oxidation rates are elevated after combined ingestion of glucose and fructose during exercise in the heat. J Appl Physiol. 2006;100(3):807–816. doi:10.1152/japplphysiol.00322.2005
  • Salame A, Brown D, Oueijan K, McCullough D. Carbohydrate supplementation for endurance exercise in the heat: a systematic review with practical recommendations. J Int Soc Sports Nutr. 2026;23(1):2669307. doi:10.1080/15502783.2026.2669307

Race context: TRI247, WTCS Abu Dhabi 2026 results · PTO Stats, Qatar T100 World Championship Final

Seven nineteen: how much of Ditlev’s Ironman world best is the athlete, and how much is Florida?

On 7 November 2026, Magnus Ditlev won IRONMAN Florida in 7:19:17: a 47:52 swim, a 3:51:16 bike and a 2:33:31 run, more than 14 minutes clear of Rudy Von Berg. It is the fastest full-distance triathlon ever recorded. It beats Sam Laidlow’s 7:21:04 at Challenge Roth in July and Kristian Blummenfelt’s 7:21:11 in Cozumel (2021) and 7:21:24 in Texas (April 2026). The bike split is also the fastest ever ridden at the distance, ahead of Cameron Wurf’s 3:53:32 in Texas in 2025.

Triathlon has no official world records. Courses, currents, wind, temperature and course measurement all differ too much, which is why careful outlets call it a “world best”. That leaves an obvious question for anyone who works with performance data: how much of 7:19 is Ditlev, and how much is Panama City Beach on the day? Results data can answer that more precisely than the headlines do.

Stacked bars of swim, bike, transition and run splits for the four fastest full-distance triathlons: Ditlev Florida 2026 7:19:17, Laidlow Roth 2026 7:21:04, Blummenfelt Cozumel 2021 7:21:11, Blummenfelt Texas 2026 7:21:24.
Four ways to go about 7:20. Blummenfelt’s Cozumel time rests on a 39:40 swim with the current; Ditlev’s rests on the bike.

The problem: a time is the athlete multiplied by the day

A finish time is the product of two things we can’t observe directly: how good the athlete was that day, and how fast the course and conditions were. Comparing raw times across venues is like comparing sprint times without knowing the wind reading. The figure above shows the point. Four performances finish within about two minutes of each other, yet their make-up is completely different. Cozumel’s sea swim is current-assisted and comes in about eight minutes faster than the others. Ditlev and Laidlow built their times on the bike. Blummenfelt’s Texas run (2:30:47) carries his 2026 time.

The way through is to use the athletes themselves as the measuring stick. Most elite triathletes race several of these courses across several years. If the same people are consistently quicker on one day than on another, the difference belongs to the day, not to them.

How I did it, step by step

  1. Data. I took the men’s professional top-10 results (swim, bike, run and total) from the PTO stats database for the four courses that hold the fastest times: Florida (2019, 2021, 2023, 2026), Challenge Roth (2019, 2022–2026), Texas (2019, 2022–2026) and Cozumel (2019, 2021, 2022, 2024, 2025). That gives 21 races and 210 results. Cozumel 2023 is excluded because the swim was cancelled. I checked that every row’s splits plus transitions add up to the total.
  2. Model. I fitted a two-way fixed-effects regression on log finish time, with one term for each athlete and one for each race edition. The race terms estimate how fast each race day was for the same people. The model explains 94% of the variance, and the residual is about 0.9% of finish time, roughly four minutes. The network of athletes links all 21 races into a single connected set, which is what allows any course to be compared with any other.
  3. Out-of-sample test. I refitted the model without Ditlev’s Florida 2026 result, then asked what it expected him to do there. I did this for total time and for each discipline.
  4. Course adjustment. I re-expressed the fastest performances as if they had been raced on the Florida 2026 day, with 95% confidence intervals.
  5. Draft check. I compared the spread of top-10 bike splits before and after the 20 m draft zone came in (1 March 2026), alongside RaceRanger’s published drafting data from Roth.

Finding 1: Florida 2026 was a very fast day, but not an outlier

Dot plot of race-day speed for 21 full-distance races from 2019 to 2026. Texas 2026 and Florida 2026 are the fastest, about 14 to 15 minutes quicker than average; Florida 2021 is the slowest.
How fast each race day was for the same athletes, relative to the average race day in the sample, at a 7:30 finishing level.

For the same athletes, Florida 2026 was about 14 minutes faster than the average race day in the sample. Only Texas 2026 (about 15 minutes) was quicker. Roth 2026, where Laidlow set the previous best, comes out about 6 minutes slower than Florida 2026. The 95% interval on that gap runs from 3 minutes faster to 16 minutes slower, so treat it as likely, not certain. An independent rating from TriRating, using a different method, also puts Florida 2026 roughly five minutes quicker than this year’s Roth.

Two things stand out. First, conditions on one course swing a lot from year to year. Florida spans almost 40 minutes, from 2021, when a brutal, current-driven swim left the leaders around 58 minutes, to 2026. Second, the general drift is downwards: seven of the eight fastest race days are from 2023–2026. Part of that is the sport getting faster (equipment, fuelling, depth of field), and a fixed-effects model can’t fully separate that from conditions. I come back to this under limitations.

Finding 2: same course, same men, three years apart

Slope chart of IRONMAN Florida finish times in 2023 and 2026: Ditlev 7:48:52 to 7:19:17, Von Berg 7:34:41 to 7:33:25, Hanson 7:45:26 to 7:35:48.
The three men who were in the top 10 at Florida in both 2023 and 2026.

The simplest check needs no model. Three athletes finished in the top 10 at Florida in both 2023 and 2026. Von Berg, who set the 2023 course record, was only 1:16 faster. Hanson was 9:38 faster. Ditlev was 29:35 faster. In 2023 he rode 3:59:02 and then lost the race on the run (2:52:40). This year he took 7:46 off his own bike course record and ran 19 minutes faster. If the 2026 day had been 30 minutes quicker than 2023, Von Berg would have seen it too. The Florida day did help, but most of Ditlev’s improvement on this course is Ditlev.

Finding 3: he beat his expected time by about seven minutes

Actual minus model-expected times for Ditlev at Florida 2026: swim 26 seconds faster, bike 3:25 faster, run 4:22 faster, total 6:57 faster, all inside the 95% prediction intervals.
Model trained without his Florida 2026 result. Grey bands are 95% prediction intervals.

Given his seven other results in the data and how fast the rest of the field went in Florida, the model expected Ditlev to finish in 7:26:14 (95% prediction interval 7:10 to 7:43). He went 7:19:17, which is 6:57 better than expected. The gain came almost entirely from the bike (−3:25) and the run (−4:22), not the swim (−0:26). That is a very good day, but not an implausible one: it sits inside the prediction interval. The run gain is the more interesting one for coaches. His 2023 Florida run fell apart after a sub-four-hour ride. This time a much faster ride was followed by his best marathon in the data.

Finding 4: where the 7:19 came from

Waterfall chart: Ditlev's expected time on a typical 2025 to 2026 race day of 7:30:51, minus 4:37 for Florida 2026 conditions, minus 6:57 for his outperformance, equals 7:19:17.
Decomposition relative to Ditlev’s expected time on an average modern (2025–26) race day.

Put the pieces together. On an average 2025–26 race day in this sample, the model puts Ditlev at about 7:30:51. The Florida 2026 course and conditions take off about 4:37. His outperformance on the day takes off another 6:57. Roughly, 40% of the gap to his “typical” modern time is the venue, and 60% is the athlete. Measured against the full 2019–26 average day instead, the venue-and-era share grows to about 13 minutes, a reminder that the reference point drives the answer.

Finding 5: adjust for the day and the “best” is a statistical tie

Fastest full-distance performances adjusted to Florida 2026 conditions, with 95% confidence intervals: Laidlow Roth 2026 7:15:13, Blummenfelt Cozumel 2021 7:19:06, Ditlev Florida 2026 7:19:17, Ditlev Roth 2024 7:19:27, Blummenfelt Texas 2025 7:20:22, Blummenfelt Texas 2026 7:22:38. The intervals overlap heavily.
Hollow circles are actual times; filled circles re-express each performance on the Florida 2026 day.

Now move the other fast performances onto the Florida 2026 day. Laidlow’s Roth 7:21:04 becomes about 7:15. Blummenfelt’s Cozumel 7:21:11 becomes 7:19:06. Ditlev’s own Roth 2024 win (7:23:24) becomes 7:19:27, within ten seconds of what he did on Saturday. Blummenfelt’s Texas 2026 time actually gets slower (7:22:38), because Texas 2026 was an even faster day. The confidence intervals run to about ±9 minutes and overlap almost completely. On this evidence, Ditlev’s Florida race is one of five or six performances that are statistically indistinguishable. If anything, Laidlow’s Roth ride on a slower day may be the strongest of them.

Finding 6: did the 20 m draft rule matter?

Line chart of the interquartile range of top-10 men's bike splits at Texas, Roth and Florida from 2019 to 2026. Texas fell to 53 seconds in 2026 even after the 20 metre rule; Florida is 4:44 and Roth 7:02.
A small spread of bike splits among the top 10 is the signature of a group riding together.

IRONMAN and Challenge Roth increased the professional draft zone from 12 m to 20 m from March 2026, citing aerodynamic testing that showed a much smaller benefit at 20 m (Triathlon Magazine Canada). At Roth, RaceRanger reported that the top-10 men’s average illegal drafting time fell from 1:47 to 0:23 (Endurance Sportswire).

Bike-split spread tells a more mixed story. In Texas the interquartile range of top-10 bike splits shrank from 1:51 in 2025 to 53 seconds in 2026. All ten rode between 3:55:35 and 3:58:16. Legal or not, that is a group, and it helps explain why Texas 2026 comes out as the fastest race day in the model. Florida 2026 looks completely different. Ditlev rode alone at the front and was 8:43 faster than the next-best bike split (Sam Long, 3:59:59). Whatever helped him on Saturday, it wasn’t a wheel. For me that makes his 3:51 a more convincing benchmark than any bike split ridden in a pack.

What practitioners should take from this

  • Benchmark athletes against themselves and the field, not against the clock. A simple athlete-plus-race model turns noisy finish times into expected-versus-actual performance. That is far more useful for evaluating a training block than a PB.
  • Read the splits, not just the total. Cozumel’s swim and the Texas group ride both “buy” time that has nothing to do with the athlete’s physiology. Normalise each discipline separately before drawing conclusions about fitness.
  • The bike–run trade-off isn’t fixed. Ditlev’s two Florida races are a natural experiment: a faster ride followed by a much faster run. Pacing, fuelling and durability (the ability to hold power and economy late in the race), not just threshold power, decide whether a big ride turns into a big finish.
  • Report uncertainty. A ±9-minute interval on a course adjustment is the honest answer. “World best” headlines that rank times seconds apart across different venues are reporting noise.

Limitations

  • Top-10 men only, from four courses: 210 results in total. The women’s races and full fields would sharpen the estimates.
  • The athlete terms are fixed over 2019–26, so athletes improving over time partly shows up as faster recent race days. A model with athlete trajectories (random slopes) is the natural next step.
  • No weather covariates. Adding wind, air and water temperature, and whether wetsuits were allowed, would let the model say why a day was fast, not just that it was.
  • Courses change: swim layouts, bike routes, and run measurement. The race-day term absorbs all of this, which is useful for comparing days but not for diagnosing causes.

Bottom line

Florida gave Ditlev a fast day, worth around four to five minutes against a typical modern race day. The rest was the athlete: about seven minutes better than even a model that knew how fast the field went, delivered by riding alone and then running his best marathon. Is it the “best ever”? Adjust for the day and it is part of a cluster of five or six performances within a few minutes of each other, and the statistics can’t separate them. Ditlev’s 7:19 tells you who was in the best shape of his career on Saturday. It does not, on its own, settle who is the fastest long-distance triathlete ever.


Data and sources

Method note: OLS on log(finish time) with athlete and race-edition fixed effects (statsmodels), n = 210, R² = 0.94. Expected times come from refitting with the target result held out; intervals are 95% prediction or confidence intervals as labelled. The dataset and code are available on request.