Golf's Two Scorecards: The Data Provenance Crisis, from the Bangabandhu Cup to ShotLink

Golf's Two Scorecards: The Data Provenance Crisis, from the Bangabandhu Cup to ShotLink On the 14th green of the second round at the Bangabandhu Cup in Mirpur, I fell into an old habit and kept my eyes on three places at once — the walking scorer's paper card, the LED scoreboard hanging toward the clubhouse, and the live scoring app in my hand. One player, one shot, three different readings. Par on the card, bogey on the board, par again on the app. And yet I had watched that shot with my own eyes — the ball stopped four feet right of the pin, the player had slightly misread the backspin. The fact that is true is written in my eyes; the fact that is moving money is not fully present on any of those three screens. This is golf's biggest and least-discussed problem. We talk about golf through scorecards, trophies, and the world ranking. But where decisions are actually made — in the market, on the sponsor's desk, in the federation boardroom — a different number arrives, and nobody asks where it was born, who typed it, who verified it. For ten years I have sat at courses watching the shape of shots, rebuilding strokes-gained tables in spreadsheets, and hitting the same wall again and again: golf's data system is weakest not in measurement but in provenance — the source, path, and verification record of the information. Context: one ball, four truths Every shot in golf produces at least four separate truths, and they sometimes collide. The first truth is the walking scorer's card — handwritten, real-time, the eye of a person standing beside the green. The second is the broadcast graphic — a number typed at a TV producer's desk, arriving seconds late and occasionally landing in the wrong lane. The third is the tour's official stat system — ShotLink on the PGA Tour, a different pipeline on the DP World Tour, and on regional circuits like the Asian Tour a thinner one, more dependent on human hands. The fourth is the market feed, which reaches the betting companies and moves the line within seconds. When I say 'one ball, four truths', it is not a metaphor — it is a system diagram. Four nodes, no direct link between one and the next, and at each node a different person with a different interest. The walking scorer wants the shot recorded correctly. The TV producer wants a smooth broadcast, and delay is tolerable. The tour stat team wants a clean database. The market feed wants speed, because speed is its business. Nobody wants to lie; but nobody takes responsibility for reconciling the four numbers either. I chose an event like the Bangabandhu Cup deliberately. Its purse is around US$400,000 — mid-sized by Asian Tour standards, but enormous against Bangladesh's domestic reality. One step below sit the Bangladesh Professional Golfers' Association weeks, where the winner's cheque stops in the low thousands of dollars. The gap between those two tiers is my real research site, because that is where the data pipeline is thinnest, and in the thinnest pipeline the provenance deficit becomes clearest. I do not claim golf is growing or shrinking in Bangladesh — I do not publish that kind of unverified claim. I only say that what can be measured should be measured, and what is measured should have its source written down. Golf's problem is not too little measurement; it is that measurement has lost its birth certificate. Core analysis: the chain of evidence A. Walking card versus broadcast graphic I once bought tickets to two Wembley semi-finals within twenty-four hours, sat there not watching the ball but hand-charting build-up sequences. I carried that habit straight into golf: instead of watching ball flight, I watch the people around the green, when the scorer's pen goes down, when the broadcast desk types the number. Strokes gained rests on the start and end location of every shot. The start is often estimated — who counted, who wrote it down? The end is even more suspect, because it is decided by a tape measure and human eyes in combination. Take one par-five from the third round. A player's tee shot found the rough. The walking scorer wrote: 'tee shot rough, second from rough to fairway, third to left of green.' On the tour stat sheet, that second shot later landed in the 'approach' category, distance eyeballed at roughly 160 yards. But the broadcast graphic showed the same shot as a 'recovery', because the producer decided a shot from rough is a recovery. One shot, two categories, and two categories mean two strokes-gained values. Where does the gap come from? From the boundaries of the categories. Strokes gained is divided into four buckets — off the tee, approach, around the green, putting. But there are no buckets on the course. There is grass, wind, wet bunkers, and a human being whose legs are shaking. A shot that starts in rough, passes through fairway, and stops on the green — is it approach, recovery, or around the green? The answer depends on who is tagging. And because the tagger differs, live strokes gained and broadcast-derived strokes gained can differ by more than ten per cent. I ran a simple test. For the same round of the same event I built two strokes-gained tables — one from the walking scorer's card, one from screenshots of the broadcast graphics. The biggest divergence was in approach, and that is no surprise, because approach has the blurriest definition. Off the tee, the difference was near zero, because everyone hits from the same place. Putting differed moderately, because both sides estimate putt length. But approach diverged most, and that is precisely where the most money is staked. So here is a plain statement: live scoring and broadcast scoring are two different sports wearing the same scoreline. The scoreline matches because the scoreline is an integer — par, bogey, birdie. But the fractions hidden beneath the integer belong to different sports. And the fractions move money, not the integers. B. Rebuilding strokes gained: my method I believe reading a reported number is not enough; you should rebuild it yourself. I learned this in 2026, when at eighteen I hand-charted all 1,690 shots of the 64 Russia World Cup matches, logging body part, angle and defensive pressure. One uncomfortable result emerged: France won the trophy with 14 goals from 10.9 xG, including Benjamin Pavard's 25-yard volley. The model was not wrong; it had found an edge case, and I learned that failure and variance are not the same thing. I brought the same principle to golf. I do not just read the tour's strokes-gained table; I take the raw shot-by-shot data, rebuild it myself, then compare it with the tour's number. Where it matches, confidence rises. Where it does not, a question appears. This rebuilding is easier in golf than football, because golf's discrete events are more numerous — seventy to eighty shots in a round, each measurable on its own. What happens in a continuous flow in football happens in separate dots in golf. That is why golf is an ideal lab for data analysis — if, and only if, provenance is clean. My method runs in four steps. First I gather the raw shot log — start location, end location, club used, and ball flight where possible. Second I drop each shot into a bucket, but if the boundary is doubtful I write the doubt down. Third I compute each shot's expected strokes from distance and line. Fourth I subtract actual strokes to get strokes gained. Every step involves human hands, and every fingerprint should be documented. Where there is no document, there is a number but no evidence. Rebuilding this way, I keep seeing one thing: for the same player and same round, the difference between the tour's strokes gained and mine is usually small, in the low decimals. But occasionally the difference grows large, and those large differences are not random — they cluster in specific categories, specific courses, specific weather. The error is not random noise; it is a pattern. And a pattern means the error can be corrected, if anyone wants to correct it. I keep a table logging, for each event, the data source, the sample size, the course setup, and my own level of doubt. That table is not a shield; it is a flashlight I point at my own bias. Because sample size gives no safety — it only shows where your ignorance is hiding. C. The caddie-to-pro pipeline: golf's cheapest edge Talking about golf data, I keep returning to a forgotten pipeline — the journey from ball-boy at Kurmitola Golf Club to professional. Bangladesh's greatest golf success story is Siddikur Rahman, who rose from ball-boy and caddie work to the Asian Tour and won the 2026 Brunei Open. This path is golf's cheapest edge — the apprenticeship happens on the course, not in a school; shot-reading is learned carrying a bag, not on a driving range. But when I go looking for data on this pipeline, the problem is clear: the story is known, the measurement is not. How many ball-boys get how much coaching per year, how many get tournament starts, how many earn a first tour card — none of it is routinely recorded. So we know one name, Siddikur, but not a rate. And without a rate, the failure point of the pipeline is invisible. I keep asking myself: with the same structural conditions, why did no second Siddikur emerge? There are many possible answers — lack of opportunity, lack of financial security, inconsistent coaching, travel-visa friction, family pressure. But none can be verified, because there is no evidence record. Here golf's provenance crisis becomes most costly — it is not just a wrong number, it is a lost generation. I deliberately invoke blockchain here, because it is a workable analogy, not a fashion. Blockchain's value lies not in proof but in provenance — who wrote it, when, and whether it matches the prior record. Golf's caddie-to-pro pipeline needs exactly this kind of auditable record: if a ball-boy's journey were logged at every step, we could see which joint in the pipeline snaps. This is no revolution; it is just keeping accounts. D. The 51-week circuit: where the model actually lives I take a contrarian position: golf's real model is not one week, it is fifty-one. Majors, the Ryder Cup, big-purse events — those are golf's windows, but golf's house is built in the domestic weeks. The distance between the Bangabandhu Cup's US$400,000 and the BPGA's small winner's cheques is the true scale for measuring golf's depth. The reason is statistical. Winning a big event depends on one week's form, one hot putting week, one lucky draw. Surviving a circuit depends on something else — consistency, injury management, travel tolerance, and the ability to stay financially afloat. Winning a major is an event outcome; staying on a circuit is a system outcome. And a system outcome needs fifty-one weeks of data, not one. Looking at regional circuit numbers, one pattern recurs — in the first two or three years many players rise, then the curve suddenly drops. The cause is usually not a lack of talent but a lack of resources. The cost of holding a tour card never recovers on a small-purse circuit. This is where the gap between the betting market and reality is clearest, because the market watches big events, but players are made in small ones. I do not claim the betting market is golf's biggest problem. I say that selling live data to betting companies is the darkest side of golf's datafication, because in that process a player's shot becomes a commodity while the human behind the shot stays outside the ledger. And the pace of that commerce is so fast there is no time for verification. I do not chase winners; I chase the moment the market forgets to update — but my fear is that those forgotten moments are now so frequent they can no longer be called exceptions. Contrarian angle: correlation is not causation There is a trap in this whole analysis, and I own it in my own name. The trap is thinking that because two datasets differ, one is false and one is true. Reality is messier. The walking card can be wrong, because the scorer is tired, standing in the sun, watching five shots at once. The broadcast graphic can be wrong, because the producer is getting information late. The tour stat system can be wrong, because category boundaries are blurry. None is singly reliable; the truth is probably not fully in any one of them, but at their intersection. I borrow one lesson from football here, carefully, because cross-domain conversion needs an explicit exchange rate or it becomes lazy analogy. In 2026, in crowdless football, I built a 92-match dataset and saw the home-win rate fall from 45.2 per cent to 38.0 per cent. That number showed a correlation, not a cause. Whether home advantage fell because crowds were absent or for some other reason cannot be proven from that dataset. Golf needs exactly the same caution. Two sets of scores differing is a correlation. Why they differ is a cause, and finding a cause needs a different kind of evidence. Another trap is thinking a big sample means a safe decision. In a season a golfer plays twenty-five to thirty events, four rounds each, roughly 100 to 120 rounds a year. That looks big at first. But split by strokes-gained category and each holds a few hundred shots, and split by course type the number often falls into double digits. Firm decisions cannot be drawn from double-digit samples. Those who do are mistaking their own bias for evidence. I admit strokes gained is itself incomplete. It measures where the ball started and ended, but not how hard the wind blew, how fast the green ran, how much the player's knee hurt, or how well they slept. These invisible variables can change a round, yet no strokes-gained table has a column for them. Every model has a France — the match that turns your confidence into a case study. Mine arrives at least once a season, and I do not hide it, because hiding it makes the model a lie. Takeaway: the signal for the next round So when I look at the next event, I do not start with the scorecard. I start with the source — where did this number come from, who wrote it, who verified it, and if nobody verified it, what is the cost of that gap to my decision. Golf's next big signal is not any player's score; the signal is when tours and regional federations begin writing down data's birth certificate. The day the walking card, the broadcast graphic and the market feed reconcile in one auditable ledger, golf analysis truly begins. Until then, I read every number as a claim, not as proof. And the question stops right here: the last time you trusted a scorecard, how much did you actually know?