Three Points in Five Hours and Twenty-Nine Minutes: The Analyst's Discipline of Writing 'Insufficient Data'

The 2026 Roland Garros final lasted five hours and twenty-nine minutes. Carlos Alcaraz saved three championship points in the tenth game of the fourth set, then beat Jannik Sinner in five sets, 4-6, 6-7, 6-4, 7-6, 7-6. It was the longest final in the tournament's history on Court Philippe-Chatrier. I stayed in my Liverpool office until nearly three in the morning, rebuilding the entire point-by-point dataset. When I finished the last column, the gap between champion and runner-up after almost five and a half hours was three points. Not thirty. Three.

The next morning my inbox filled with commissions. Alcaraz is mentally stronger. Sinner choked again on the big stage. The Alcaraz era has begun. I opened the spreadsheet, looked at the notes cell, and typed a single line: insufficient information to conclude. It is the hardest line in this job, and the most valuable one.

I work as a sports data analyst in Liverpool, covering tennis for the English market. I studied sociology, so I always ask in what circumstances the data was generated before I ask what it says. A metric stripped of surface, season phase and opponent quality is decoration.

Tennis is a small-sample sport. A match runs 150 to 250 points. A set can be 40 points. A tie-break is 12. Judging a player from one set means reading a table whose error bars are larger than the signal you are chasing.

The metrics I use daily: first-serve points won, second-serve points won, return points won, break-point conversion, winner-to-unforced-error ratio. In football I have xG. In tennis, the equivalent model lives at point level: every point carries a win probability that depends on the server, the surface and the score state. It sounds rigorous. But a model is only as rigorous as its inputs, and the inputs to a Grand Slam final are usually distorted by exactly the things the model leaves out.

In 2026 I was twenty-three, interning at a sports analytics firm. I charted the World Cup round of sixteen in Russia. Spain held 71.4 percent possession, completed 1,029 passes and generated 0.9 xG across 120 minutes. I predicted a Spain win. They lost on penalties. I spent a week rewatching the tape and learned my first lesson: possession describes territory, not chances. Since then every piece I write opens with actual chances.

Back to Paris. Three points across five hours and twenty-nine minutes. If you replayed that match a thousand times with the same two players, the same surface, the same conditions, you would get roughly five hundred Alcaraz wins and five hundred Sinner wins. Three points are not evidence of nerve; three points are evidence of a match sitting right on the edge of a coin.

Yet an entire media industry built a psychological narrative out of those three points. I have nothing against storytelling. I object to taking a three-point sample and loading it with the weight of a career.

This is why I open every analysis with a ten-match rolling window rather than a single match. For a player who competes in 60 matches a season, one match is 1.7 percent of the sample. Nobody values a stock on a single trading session. Yet tennis still values a player on a single afternoon.

The ATP ranking is a 52-week rolling data structure. Points do not vanish after a match; they vanish exactly 52 weeks after they were earned. Every player therefore carries a pre-scheduled drop calendar. A Masters 1000 semifinal is worth 360 points. Defending it does not mean playing better; it means reproducing a five-match winning streak in that precise week.

And here the data gets interesting. The variance of a five-match hard-court run is enormous. One draw that hands you a bad stylistic matchup, one rain delay, one evening when the serve misfires — any of these flips 360 points. When I model drop probability for the top ten in March, I always add a 15 to 20 percent error band for draw luck. No model predicts the draw.

Error is the most disagreeable friend I have, but it is the only one that never lies to me in a meeting room.

The 2026 season posed a different problem. Jannik Sinner was suspended for three months, from February 9 to May 4, under a settlement with the World Anti-Doping Agency. He missed Indian Wells, Miami, Monte Carlo and Madrid. Four Masters 1000 events disappeared from the sample.

What is the professional handling? No interpolation. I do not fill the gap with an estimate from the previous season, because the surfaces, the opponents and the physical state have all changed. I mark that region as insufficient information and widen the confidence interval on every subsequent forecast. A model projecting Sinner to win Roland Garros, built only on data from May onward, carries far larger error than one built on a complete season.

People assume more data is always better. It is not. Data missing in the dangerous place is worse than data missing at random.

An older example. Daniil Medvedev reached the 2026 Australian Open final after more than twenty hours on court in a fortnight, led Sinner two sets to love, and lost in five. The story told was a mental collapse. The data told something else: workload. Medvedev played four five-set matches in the same tournament, and before the final his minutes on court exceeded Sinner's by more than seven hours.

Alexander Zverev reached the 2026 Australian Open final and lost to Sinner in straight sets. The press called it his third Grand Slam final defeat. Three times, on three surfaces, against three different opponents. Three is an anecdote with good lighting, not a statistical sample.

Take one technical example. A player hits roughly 70 first serves in a three-set match. If his true first-serve percentage is 64, the standard deviation of the observed figure in a single match is close to six percentage points. A match at 58 percent and a match at 70 percent can come from the same player, the same serve, the same practice week. When a commentator says he served far better today, I usually check and find the difference sitting comfortably inside the noise.

Only when I pool ten matches does the error shrink enough for me to make a statement of any confidence.

There is another variable the spreadsheet never records. In 2026, with stadiums empty, I worked as an analyst for a tactical consultancy. At the Merseyside derby that June, Liverpool drew 0-0 with Everton. Liverpool's PPDA — passes allowed per defensive action — rose from 9.8 to 11.5. Their high-intensity running distance fell 4.3 percent. Same players, same system, one difference: nobody in the stands.

The empty stadium taught me something brutal: noise never appears in the spreadsheet, but it always appears in the heartbeat.

Since then every report I file carries a mandatory note: crowd or no crowd, home or away, days of rest between matches. Without that note, a number is just a well-formatted lie.

Money is a variable too. In October 2026 the Six Kings Slam exhibition in Riyadh paid champion Sinner around six million US dollars, many times a Grand Slam title cheque. I do not put those matches in the same column as ATP matches. An exhibition carries no ranking points, no ranking-defence pressure, and often no incentive to compete to the end. Placing it beside a Grand Slam fourth round is a methodological error, not a moral one.

Old data is not wrong; I was simply laying it on the operating table in the wrong season.

The irony of this profession is that the reward does not come from being right. It comes from being certain. Live data feeds sold to betting companies do not need you to be correct; they need you to be willing to stake. An analysis that concludes insufficient information generates no traffic, no click-through rate, and nobody pays for it.

That is precisely the most honest product an analyst can deliver.

I have learned to separate two kinds of question. The first: what happened. The second: what will happen. The first can always be answered, provided I put in the hours. The second can only be answered when the sample is large enough, and most of the time the sample is not large enough.

Correlation is not causation, and in tennis spurious correlation shows up more than in any other sport.

Last season a player changed coaches and his second-serve points won rose from 52 to 55 percent across eight matches. Many called it the new coach's fingerprint. I checked opponent quality: across the eight matches before the change, the average opponent ranking was 28; across the eight after, it was 74. A three-point swing sits inside the error band once the opposition drops two tiers. I am not denying it. I am saying there is no evidence yet.

The principle I hold tightest is assigning responsibility to structure before assigning it to individuals. In 2026 I analysed Leicester City's fifteen-match slump after their FA Cup triumph. Seven centre-backs injured. Jonny Evans out for twelve matches. Their expected goals conceded rose 24 percent. Nobody on the coaching staff wanted to hear the word unlucky, and I did not put it in the report either. I went into the centre-backs' running distances: 8.2 kilometres per match on average, falling 12 percent after every fixture spaced fewer than 72 hours apart. The cause lay in the schedule and the rotation policy, not in one man's legs.

An injury chain is not a curse; it is a map exposing the depth of a system being eroded.

In tennis the same principle applies. When a player repeatedly struggles with a shoulder or a lower back, my first question is not about his physiology. It is how many serves he has hit across how many consecutive weeks. Load is a variable, and load is measurable.

The next cycle of the season gives me three signals to check. First, the points-defence windows in the spring hard-court swing, where a few top players will lose positions to variance rather than form. Second, how the market prices Sinner's three-month gap as the calendar turns to clay. Third, whether the next major final is again decided by three points, and whether anyone is willing to write that three points prove nothing.

I do not trust a number, but I trust the story it tells after I have cross-examined it three times.