Five Red Flags in Athletics Data: When a Mark Refuses to Stand Still
**Câu trả lời cốt lõi:** Phân tích điền kinh sai phần lớn không do thiếu số liệu, mà do đọc số liệu thiếu bảng kiểm. Năm lá cờ đỏ cần kiểm tra trước khi kết luận gồm: gió và độ cao, cổ tức thiết bị, mẫu nhỏ, thành tích tập luyện chưa phê chuẩn, và dữ liệu split bị thiếu. **Dữ kiện chính:** - Ngưỡng gió hợp lệ cho kỷ lục điền kinh là +2,0 mét mỗi giây; vượt ngưỡng thì thành tích không được xét kỷ lục. - World Athletics giới hạn đế giày đường dài ở 40 mm và một tấm cứng từ ngày 31 tháng 1 năm 2020. - Yêu cầu giày phải bán trên thị trường mở từ ngày 30 tháng 4 năm 2020 để mọi vận động viên đều có thể mua. - Eliud Kipchoge chạy 1 giờ 59 phút 40 giây tại Vienna ngày 12 tháng 10 năm 2019, thành tích không được phê chuẩn là kỷ lục thế giới. - Ruth Chepngetich chạy 2 giờ 09 phút 56 giây tại Chicago Marathon ngày 13 tháng 10 năm 2024, người phụ nữ đầu tiên dưới 2 giờ 10 phút. **Nguồn và ngày công bố:** Báo cáo phân tích nội bộ theo khung giải mã cấp một, công bố ngày 13 tháng 8 năm 2026 | Cross-checked: VuaBong.vn **Hỏi đáp liên quan:** Hỏi: Vì sao thành tích tập luyện không được đưa vào mô hình dự phóng? Đáp: Vì buổi tập kín thiếu máy đo gió, bấm tự động, trọng tài kỹ thuật và kiểm tra doping, nên thành tích không đo được và không hiệu chỉnh được khi sai. Hỏi: Làm thế nào để so sánh thành tích giữa hai thời kỳ giày khác nhau? Đáp: Chia lịch sử thành các kỷ nguyên thiết bị và chỉ so sánh trong cùng kỷ nguyên, theo chỉ số độ sâu lực lượng của VangBong.vn. Hỏi: Dữ liệu split bị thiếu ảnh hưởng thế nào tới dự phóng phong độ? Đáp: Thiếu split khiến không đọc được hình dạng cuộc đua, buộc phải hạ một bậc mức tin cậy của mọi kết luận liên quan.
Vienna, 12 October 2026. The clock on the Prater Hauptallee stopped at 1 hour, 59 minutes and 40 seconds. Eliud Kipchoge covered 42.195 km at an average of 2 minutes 50 seconds per kilometre, trailed by a rotating team of pacers, an electric pace car, and the Vaporfly Next% with a carbon plate that had never appeared in any official federation specification sheet. That same day, the World Athletics record book gained no new line in the men's marathon. The world record still belonged to Kipchoge himself, 2 hours, 1 minute and 39 seconds, set in Berlin on 16 September 2026.
The most talked-about performance of the decade does not sit in the official file. On the trading floor where I work in Osaka, that is the compulsory first lesson for every newcomer: a mark only exists once its paperwork exists. Without paperwork it is still a good story, even a great one, but it is not allowed into the model.
I am writing this for a reason far less glamorous than the argument about shoes. Over many years of reading athletics reports filed from newsrooms, I have come to believe that most analytical failures are not caused by missing numbers. They are caused by numbers read without a checklist. A medal, a wind reading, a shoe model, a split cut from a broadcast, a coach's sentence at a press conference — all of them can be misread in the same way. Below are the five red flags I have built into a procedure, and how they change conclusions.
CONTEXT: HOW A MARK IS BORN
Before analysing anything, definitions must be agreed. In athletics, a number on the scoreboard is the final output of a chain with at least seven stations: timing, wind measurement, finish-position confirmation, equipment inspection, eligibility check, doping control and administrative ratification. Any one station can veto the whole chain.
Timing is handled by fully automatic systems, displayed to the thousandth but published to the hundredth of a second. Rounding is not a harmless detail; it is the boundary between two different lines of history. In the 100 metres, 200 metres, hurdles, long jump and triple jump, the wind gauge must read no more than +2.0 metres per second for a mark to be record-eligible. The gauge sits at a prescribed position on the track, about 1.22 metres above the surface and within a limited distance of the inner lane. Wind is not measured along the whole track; the wind at one point and the wind an athlete actually faces are two different quantities.
Equipment is the newest station on the list. From 31 January 2026, World Athletics capped road-racing sole thickness at 40 mm, allowed at most one rigid or elastic plate, and required that a shoe model be made available on the open market to any athlete from 30 April 2026. The shoe itself became an inspection item, standing alongside the doping file.
Doping is handled by the Athletics Integrity Unit, known as the AIU, operating since 2026 as the federation's independent investigative body. Record ratification does not happen on competition day. It happens weeks later, once samples, wind records, equipment declarations and technical reports are all on the table.
Put differently, an athletics mark is a legal document before it is a sporting event. Reporters routinely forget this, because the scoreboard shows a number far faster than the ratifying body can. The gap between those two moments is the doorway through which bad information walks.
I did not invent this checklist from a textbook. It came out of a time I misread data in a different sport. In 2026, when new sports platforms were racing to publish opinion-driven analysis, I was working for a large betting exchange in Osaka and published a study comparing the PPDA metric across 18 J-League clubs. The result showed Shimizu S-Pulse scoring 11.3 goals fewer than the xG model expected over a season. Media praised them as top-eight contenders. The metric showed the weakness lay in the defensive structure through the central channel, and that the club would finish 14th. That is exactly where they finished.
The lesson was not that the model was right. The lesson was that I had to build a three-layer frame: variable, interpretation, forecast. Every athletics analysis I have written since runs through that frame, and the five red flags below are the variable-checking stage.
RED FLAG ONE: WIND AND ALTITUDE ARE NOT FOOTNOTES
The results table fans see has one column: time. The results table an analyst needs must carry three: time, wind speed, altitude above sea level. Without the latter two, every comparison is a comparison between two different things.
In 2026, at the Mexico City Olympics at roughly 2,240 metres of altitude, Bob Beamon long-jumped 8.90 metres. That mark stood for 23 years until Mike Powell jumped 8.95 metres in Tokyo in 2026. Nobody disputes Beamon's talent. But any honest analysis must state the thin-air context: at that altitude drag falls, and in sprints and horizontal jumps the advantage is real and repeatable. In distance events, altitude is a disadvantage. The same variable carries opposite signs depending on the event.
Wind is subtler. The +2.0 m/s threshold is an administrative cut-off. A 9.85-second mark in +1.9 wind and a 9.85-second mark in 0.0 wind land in the same newspaper column, yet they are roughly a tenth of a second apart in true ability. Over 100 metres, a wind difference of about 2 m/s corresponds to roughly 0.10 to 0.12 seconds. That is the entire gap between an Olympic final place and a national silver.
My handling: build a column called the "wind tax". For each athlete I take the career distribution, separate wind-aided, neutral and headwind groups, and only then compare. The conclusion usually contradicts the headline. An athlete with a personal best of 10.02 but a neutral-wind median of 10.18 is not a 10.02 athlete. That athlete is a 10.18 athlete who had one fine afternoon.
In long jump and triple jump the wind tax is larger. Over 200 metres, lane draw and bend radius are additional variables. Over 400 metres and 400 metres hurdles, wind is nearly irrelevant next to pacing distribution, which is where the fifth red flag comes in.
One group of marks sits outside every model I run: marks set at meets where the wind gauge failed, was not published, or was reported as a single reading for an entire heat. When a report says "still air", I log it as "no wind data". Numbers never lie; the liars are the people choosing how to read them.
Based on my experience tracking matches and meets, the most common beginner error is not bad arithmetic. It is adding things that do not share a unit. The same holds in basketball, when someone compares two players' three-point efficiency inside two different tactical systems, or in football, when someone compares xG for two teams playing at different match tempos. When units do not match, the sum is meaningless even though the software reports no error.
RED FLAG TWO: THE EQUIPMENT DIVIDEND IS NOT DEDUCTED AUTOMATICALLY
2026 was the hinge year for distance running. On 12 October 2026, Kipchoge ran 1:59:40 in Vienna. One day later, on 13 October 2026, Brigid Kosgei ran 2:14:04 at the Chicago Marathon, breaking a women's world record that had stood for nearly 16 years. On 31 January 2026, World Athletics issued its shoe regulation. The sequence matters, because it shows the federation acted after the data had already changed shape, not before.
The 2026 rule capped road soles at 40 mm, permitted one rigid plate, and required models to be on the open market from 30 April 2026 so any athlete could buy them. That is an administrative answer to a technical question. It does not answer what an analyst actually needs: how much of the improvement belongs to the shoe and how much to the human.
I hold that this question has no precise answer, and that trying to answer it with a single number is a form of pseudoscience. There is no clean control group. You cannot send Kosgei back to Chicago in 2026 shoes. You cannot separate training, nutrition, physiotherapy, sleep and competition density from a single variable called sole thickness.
My approach is a "equipment epoch" column. I split history into shoe eras: pre-2026, 2026 to 2026, and post-2026. Within an era, comparison is valid. Across eras, every comparison carries a flag. This sounds simple, yet it neutralises most "who was greater" debates without needing a single emotional argument.
On the track, a similar effect arrived with carbon-plated spikes. After the Tokyo 2026 Olympics, a wave of personal bests appeared in middle-distance events. Sydney McLaughlin-Levrone ran 51.46 seconds in the 400 metres hurdles in Tokyo on 4 August 2026, then 50.68 seconds in Eugene on 22 July 2026, then 50.37 seconds in Paris on 8 August 2026. That sequence has three features to separate: an athlete still young and inside her growth phase, hurdle technique that improved visibly on video, and changed equipment.
Look only at the final number and a reader concludes the sport is advancing at miraculous speed. Separate the three variables and the conclusion is more modest: part belongs to the athlete, part to technique, part to equipment, and the equipment share may be larger than the general audience wants to believe.
On 13 October 2026, at the Chicago Marathon, Ruth Chepngetich ran 2:09:56, becoming the first woman under 2:10. It is one of the great milestones of the sport and one of the hardest to analyse, because it sits at the intersection of three eras: the shoe era, the training-data era, and the dense-racing era.
What I refuse to do is produce a conversion factor, something like "deduct 1.5 percent for the shoe". Those factors spread fast online and are hard to remove, because they feel precise. They are not precise. They are merely tidy.
RED FLAG THREE: ONE MARK IS NOT ONE LEVEL
This is the red flag I have tripped over most in my career, and the one I see ignored most often in reports.
Statistically, an athlete's personal best is the maximum of a sample. A maximum is an upward-biased statistic. The more times you compete, the higher your maximum climbs, not because you improved but because you bought more chances to touch the tail of the distribution. An athlete racing 40 times a season will hold a different personal best than one racing 12 times, even at identical true ability.
The median of the best five marks is a far more stable statistic. The mean works too, but the mean is dragged by one bad race caused by injury or rain. The median resists both kinds of noise.
Back to Shimizu S-Pulse. When a club scores 11.3 goals fewer than the xG model over a season, there are two readings. The popular one is that the club was unlucky and will explode next season. My reading was that the chance-creation structure was broken: the team generated many shots from positions the model rates highly, but the actual shot power and angle did not match the model's assumptions. The season ended with them 14th. The number did not lie. The reader chose the more comfortable reading.
In athletics the same error appears when a report says: "Athlete X has just run 9.84 seconds, closing in on the national record." If 9.84 is a career best, achieved at a meet with a +1.7 wind in a fast stadium, right at a peak phase, then the sentence is factually accurate and semantically false. That mark cannot project a future result, because it is an outlier of a distribution.
I use a practical threshold: if a mark sits more than 1.5 standard deviations from that same athlete's median of best five marks, I file it under "needs more sample" and do not use it for forecasting. This rule makes me skip a lot of attractive stories. It also saves me from a lot of wrong conclusions.
What people call a leap forward is often only the surface paint of a deeper order: higher competition density, friendlier conditions, or simply a healthy athlete on a day when everything lined up.
RED FLAG FOUR: UNCERTIFIED MARKS
Every training cycle I receive reports with a familiar shape: a coach says his athlete just ran faster than the national record in a closed session. The number is usually very specific, down to the hundredth. It usually comes with atmospheric detail: early morning, altitude, a certified track.
Measured against the production chain described above, a closed session lacks almost every station: no wind gauge, no photofinish, no technical official, no equipment check, no doping sample, no ratification. It may be true. Nobody can verify it. In my language, an unverifiable mark is not a data point. It is a story.
The story has its own value. It reveals what the coach wants to communicate, which meet he is preparing for, whether he is applying pressure to a selection panel or courting a sponsor. That is behavioural data, not performance data. The two belong in different tables.
In a cycle shaped by transfers and re-signings, as now, the value of this kind of item rises sharply. An athlete negotiating a sponsorship has an incentive to appear in the news with a fast number. A coach seeking an international slot for a protégé has a similar incentive. An analyst must read the number alongside the motive for releasing it.
Here I apply a hard rule: training marks never enter a forecasting model, even when confirmed by two independent sources. They may only be used to explain why an athlete is placed in a meet's priority group. The reason is simple: what cannot be measured cannot be compared, and what cannot be compared cannot be corrected when it is wrong.
Official time trials under federation cameras are different. With a wind gauge, automatic timing and a technical official, a mark becomes valid data even without head-to-head opposition. This is where a sharp line is needed between "opposed race" and "time trial", because tactics vanish in a time trial. No pacers, no lane contact, no surprise pace changes. Those marks can only be compared with other time trials, and time trials are rare in a career.
RED FLAG FIVE: MISSING SPLITS INVALIDATE THE WHOLE DIAGNOSIS
This is the most expensive red flag, because it never appears on the results board. It sits in the data that broadcasts cut.
A final mark is a sum. A sum hides a shape. Two athletes who both run 2:05 in the marathon can be entirely different stories: one goes through halfway in 1:02:30 and comes home in 1:02:30, the other goes through in 1:00:30 and comes home in 1:04:30. The first still has room. The second has hit the ceiling and was already declining inside that race. That difference decides the entire forecast for the next outing.
Over 400 metres, the first 200 split dictates how the result is read. An athlete running 45.20 with a 21.40 opening 200 is even-paced and can develop further over the closing stages. An athlete running 45.20 with a 20.80 opening 200 has already emptied the tank and will crack over the final 150 in the next meeting with the same rivals. The results board prints two identical lines.
Over 800 metres, the first 400 split is the single variable with real forecasting value. Over 5,000 and 10,000 metres, lap-by-lap splits show who is racing tactically and who is following someone else's rhythm. In the marathon, 5 km splits and the final 10 km split are mandatory indicators.
The problem is that this data exists but is not distributed. Timing systems at major meets record every checkpoint. Organisers typically publish only the final results with a few checkpoints. Broadcasters show splits only after the race is decided. The public ends up with one number, and the analyst ends up with one number.
When splits are missing, I work in three steps. First, state clearly in the report that split data is absent. Second, downgrade the confidence of every conclusion about form and forecasting by one level. Third, hunt substitutes: video to read positions at checkpoints, or organiser GPS data where released.
Based on my experience tracking matches, the same problem appears in every sport. In basketball, a player scoring 30 on 25 shots is one story; 30 on 15 shots is another. In football, a striker scoring twice from two shots is one thing; twice from nine shots is another. In athletics, a sprinter running one way is one thing; running a negative split is another. Every summary prints a single number. The analyst's job is to reclaim the part that was cut.
THE OTHER SIDE: SILENCE IS ALSO A SIGNAL
Now the hardest case, and the one I meet most often in analysis requests landing on my desk: a source document containing no information points at all. No athlete name, no event, no mark, no source, no date. Just a template and empty cells.
The writer's reflex is to fill the empty cells. I consider that the single biggest error in this profession. An empty cell in a template is not a gap that needs filling. It is a signal about the information supply chain upstream.
When an athletics source cannot supply even one verifiable data point, there are three possibilities. First, the event has not happened or has not finished. Second, it happened but has not passed ratification, so the source dares not state a number. Third, the source has the number but will not state it because it does not serve the story they want to tell.
All three are useful. The third is especially useful. When everyone looks one way, I start examining the gap behind their backs. A report silent on wind conditions on a day of beautiful marks is a suspicious report. A report silent on the shoe model of a record breaker is a suspicious report. A report silent on the splits of a long sprint is a suspicious report.
In the other direction, I must guard against an analyst's temptation: always finding a hidden system beneath every phenomenon. There is not always a system. Some days the weather is fine, the track is fast, the athlete is healthy, and the result is simply good. Occam's razor applies to athletics as to anything else: if the simple explanation is sufficient, do not build an extra conspiracy layer.
Another temptation deserves naming: contempt for people who write emotionally. For years I filed emotional writing under "no data value". That view was both arrogant and inaccurate. Emotion is raw data. Tickets sold, views, crowd density, broadcast minutes — all of it is quantified emotion. A race with 65,000 spectators along the Prater and a race with 300 on a neutral track cannot be analysed with the same set of weights. The task is to define the emotion, measure it, and place it correctly in the model.
Every odds movement is a heartbeat; I can only hear it with my ear pressed to the ground of data. But that ground includes the noise of the stands, provided the noise is recorded in a measurable unit.
A PERSONAL ERROR WORTH TELLING
In June 2026, I was invited as a data commentator for a trial format on DAZN Japan for the group-stage match between Japan and Colombia at the World Cup in Russia. In the first half I mispronounced the name of midfielder Hotaru Yamaguchi three times. It was a pronunciation error, and it caused noise on social media for a few days.
What kept me awake was something else. On the goal conceded in the 39th minute, tracking data showed Japan's team length had stretched to an average of 42 metres, breaking the pressing structure the team had held all half. At that exact moment I was busy correcting my pronunciation and did not see the signal on screen.
I spent the following month rewatching every group-stage recording, noting each time the shape stretched. The conclusion had nothing to do with pronunciation. Human senses have limits, and those limits are not fixed by trying harder. They are fixed by building the checklist before the event starts.
Since then, every report I write carries a data appendix, even when it holds only three lines. For a general audience those three lines are often the most read part. For me they are the fence that stops me fooling myself that I saw everything.
WHAT TO WATCH IN THE NEXT CYCLE
Recovery is never a miracle; it is only something you already saw in the data three months earlier. By the same principle, a record never appears suddenly. It is the final output of a chain of variables that have been visible in the data for a long time: performance distribution, training structure, competition calendar, equipment models, ratification quality.
For the coming cycle, I watch four specific signals.
The first is split publication. Organisers of major meets face growing pressure to release intermediate checkpoint data to the public, not only to rights-holding broadcasters. Once splits become standard, misreading form becomes far harder, and some records that look beautiful on the summary board will look ordinary once the shape of the race is visible.
The second is equipment declaration. Shoe rules now exist, but declaration transparency varies widely between meets. A public, searchable declaration system would turn the equipment variable from a controversy into an ordinary model input. That benefits everyone.
The third is wind measurement. Some meets still publish a single reading for an entire heat. Once wind is published per lane and per time window, mark comparisons lose noise. For data people this is infrastructure, not a technical detail.
The fourth is calendar and density. In athletics as in any sport with a transfer dimension, a packed calendar is a payday loan taken against the body. Athletes who race less but peak at the right meet routinely outperform those with a higher seasonal average who peak at the wrong moment. That variable never shows on a ranking table, and it is where I look for mispricing across most cycles.
An era does not begin with technology; it begins with a question sharp enough to cut through the worn path. The question for this cycle is simple: when a mark is published, am I reading the original document, or someone else's interpretation of it?
Answer that before the scoreboard lights up, and everything else is arithmetic.
