Strokes gained explained: how golf finally learned to count

A player misses eleven greens, chips it stone dead eleven times, holes out from inside three feet every time, and finishes the day leading the field in putts per round. Somebody else hits every green, faces nothing shorter than twenty feet, two-putts fourteen of them and finishes near the bottom of the same list. One of those two putted well. The statistic says it was the other one.

That is not a freak result. It is what putts per round is built to produce. Strokes gained is the golf statistic that fixed it, and the fix was drastic: throw out the old columns and rebuild the measurement from the ball up. Fairways hit and greens in regulation fail in the same way for the same reason, and almost nobody noticed until somebody took the defect seriously enough to start again.

The idea underneath it is not complicated, and the arithmetic fits in one line. What takes explaining is why the old numbers fail, what the baseline actually is, and what the whole apparatus still cannot see.

Why fairways hit, greens in regulation and putts per round all lie

The traditional golf box score has four columns: driving distance, driving accuracy, greens in regulation, putts per round. Scrambling and sand saves get added when there is space. Every one of them fails in one of two ways. Either it compresses a huge range of outcomes into a tick box, or it credits one shot with work done by a different shot.

Fairways hit is a tick box. A ball a yard into the light rough with a clean lie, 290 yards down the hole, is a miss. A ball buried in thick grass 190 yards down the hole is also a miss, and the two are recorded identically. A 210-yard iron laid up into the middle counts the same as a 320-yard drive that finishes in the same strip of grass. The column knows nothing about distance, which is why driving distance sits next to it, and the two together still fail to say whether the tee shot was any good, because the box score has no exchange rate between a yard of distance and a percentage point of accuracy. Par-3 tee shots are excluded by convention, so a chunk of the tee shots on any course never appear at all.

Worse, hitting fewer fairways is often correct. A player who takes driver on a hole where the fairway pinches at 300 yards is buying a shorter approach with a wider dispersion, and on many holes that trade is clearly the right one. The statistic penalises him for making it.

Greens in regulation is the same tick box one shot later. The ball is on the putting surface in two under par strokes, or it is not. A putt from ninety feet across two tiers counts as a green hit. A ball resting on the fringe eight feet from the flag does not. The range of outcomes inside the word "hit" is enormous, and the column throws all of it away.

It also inherits the tee shot wholesale. A player driving it forty yards past his playing partner hits his greens with a wedge while the other man hits a 5-iron. The greens-in-regulation column reports that as superior approach play. It is nothing of the kind.

Putts per round is the contamination case in its purest form, and it is the one that produces the opening paragraph of this article. The number of putts on a hole is set almost entirely by where the ball was when it arrived on the green, and where it arrived is set by the shot before. The most reliable way to lower your putts per round is to miss more greens, chip close, and hole the short one. The statistic rewards a worse round of golf.

Then there is the distance problem. Two putts from sixty feet and two putts from four feet are the same entry. One is close to the best outcome available and the other is a stroke thrown in the bin. Putts per green in regulation was the patch applied to this, and it does fix the missed-green contamination. It leaves the distance problem completely untouched.

Scrambling and sand saves repeat the pattern at smaller scale. A chip to two feet from a clean lie and a bunker shot short-sided to a downhill lie forty yards away sit in the same denominator, and whether you convert is largely decided by which of those two you are facing.

Proximity to the hole is the best of the old numbers, because it is continuous rather than binary. It still says nothing about where the shot was played from. Twelve feet from 180 yards out of the rough is a very good shot. Twelve feet from 95 yards in the fairway is a poor one.

Put the whole box score in one sentence and the flaw is obvious. Golf is a sequential game in which the difficulty of every shot is decided by the shot before it, and the traditional statistics record outcomes as though each shot appeared from nowhere. What was needed was a measurement that knew where the ball started.

The one number that fixes it: expected strokes from where you stand

The fix requires something the old statistics never had, which is a record of where the ball was, before and after every single shot. Professional golf now has that. Laser measurement and surveyed course models, operated by armies of volunteers, log the position of every shot played in a tournament to within a very small margin. Once every shot has a start point and an end point, the whole box score can be thrown away and rebuilt.

The rebuild starts with a question that sounds trivial and is not: from here, how many more shots is this going to take?

Take every recorded shot that started within a narrow band of distance from the hole, from the same kind of lie, and count how many strokes each of those players actually needed to finish the hole from there. Average them. That average is the expected strokes to hole out from that position, and it is the baseline that everything else is built on.

In the usual implementations the baseline is a function of two things and only two: the distance to the hole, and the lie the ball is sitting in, coded into a handful of categories such as tee, fairway, rough, sand, recovery and green. That is it. No wind, no slope, no pin position, no player identity. The simplicity is deliberate, and it is also the source of most of the limitations later in this article.

You can reason out the shape of the curve without seeing a single figure.

It rises with distance and never falls, because further from the hole is never easier. It cannot go below one, because from any position that is not in the hole at least one more shot is required. It is concave, which is the property that does most of the work: between one inch and one foot the expectation barely moves, since both are a tap-in, while between three feet and thirty it climbs steeply, because the make rate collapses across that range. Out beyond a hundred yards it flattens again, because the back half of a long approach is played much the same whether the ball started at 160 yards or 175.

At any given distance the curve is ordered by lie. Sand is worse than rough is worse than fairway. Close to the green that ordering starts to creak, because a greenside bunker with a clean lie and a decent stance is frequently an easier proposition than thick grass on a downslope, and the lie categories are too coarse to know it.

None of this is exotic, and none of it is unique to golf. It is the same construction as run expectancy in baseball, where each of the twenty-four combinations of bases occupied and outs recorded carries an average number of runs that follow, or expected points in American football, the number that sits underneath every fourth-down decision a modern coach makes. The state is the position. The value is what usually happens next. The credit for a play is the change in the value.

Strokes gained explained: the arithmetic of a single shot

Here is the whole thing.

The value of a shot is the baseline where it started, minus the baseline where it finished, minus the strokes it cost.

The last term is the part people skip and it is the part that makes the measure honest. Moving from an expectation of 4.0 to an expectation of 3.0 is not progress. It is exactly average. You spent one stroke and you bought precisely one stroke's worth of position, so the shot broke even. Anything better than that is gained, anything worse is dropped, and zero does not mean nothing happened. Zero means it went as expected.

Work one through. The numbers below are invented, chosen to be round so the arithmetic stays legible. They are not tour figures.

A tee shot on a 400-yard par 4. Say the baseline from that tee is 4.0 expected strokes. The drive finishes in the fairway, 150 yards from the flag, and say the baseline there is 2.8.

4.0 minus 2.8 is 1.2. The ball is now 1.2 strokes' worth closer to being in the hole. Subtract the one stroke it took to get there. The drive gained 0.2.

Now the same tee shot pushed into the rough, 170 yards out, from a position worth 3.2. That is 4.0 minus 3.2, which is 0.8, minus the stroke. The drive lost 0.2. Note what happened there. The difference between those two tee shots is four tenths of a stroke, roughly what a decent putt from twelve feet is worth. Not a disaster, and nothing like the gulf that the fairways-hit column implies.

1. 1Record where the ball startsA tee shot on a 400-yard par 4. The position is only two things: the distance to the hole, and the lie, which here is the teeing ground.
2. 2Look up the baseline for that positionThe average number of strokes a player of the reference standard needs to hole out from there. Call it 4.0 in this example.
3. 3Play the shot and record where it stopsThe ball finishes in the fairway, 150 yards from the hole. Distance and lie again, nothing else.
4. 4Look up the baseline for the new positionFairway, 150 yards. Call it 2.8.
5. 5Subtract the new baseline from the old one4.0 minus 2.8 is 1.2, which is how much closer to holing out the ball now sits, measured in strokes.
6. 6Charge the shot that was usedSubtract one for the stroke played, and one more for every penalty stroke incurred. There were none here, so subtract one.
7. 7Read the answer1.2 minus 1 is plus 0.2. The drive gained two tenths of a stroke on the baseline.

The baseline values in this example are invented and round, chosen so the arithmetic is easy to follow. They are not tour data.

Penalty strokes fit the same formula without any special casing. Suppose that tee shot finds a penalty area. Under the Rules the player takes one penalty stroke and drops, and say the drop leaves him 240 yards out in the fairway, a position worth 3.5. The calculation is 4.0 minus 3.5, which is 0.5, minus two strokes, one played and one penalty. The tee shot cost 1.5.

That comparison is worth sitting with. The good drive was worth plus 0.2. The bad one was worth minus 0.2. The one in the water was worth minus 1.5. Golf's arithmetic is not symmetrical, and the rest of this article keeps returning to that fact.

Two properties fall straight out of the formula.

The first is that the sign of a shot has almost nothing to do with whether it looked good. A drive into the rough is positive if it is far enough down the hole. A drive in the fairway is negative if it is short enough. A putt that finishes eight feet past is heavily negative even when the stroke was struck perfectly, because the model prices where the ball ended up and does not care what you meant.

The second is that everything is denominated in strokes, which is the unit on the scorecard. A putt and a drive are now measured in the same currency and can be added together. Nothing before strokes gained could do that, and it is the property that makes the rest of the system possible.

Why the shot values on a hole add up to the score

This is the part that gets left out, and it is the reason the whole thing is trustworthy rather than merely clever.

Write the expectations along a hole as E0, E1, E2 and so on. E0 is the baseline from the tee. The last one is zero, because the ball is in the hole. The value of shot number i is E(i-1) minus E(i) minus one.

Add all the shot values on the hole together and watch what happens. E1 is subtracted in the first shot's value and added back in the second's. E2 is subtracted in the second and added back in the third. Every interior term appears exactly twice with opposite signs and cancels. What survives is the first expectation, minus the last one, minus the number of strokes played.

The last expectation is zero. So:

Strokes gained on a hole equals the baseline from the tee, minus the number of shots you actually took.

Take the worked hole above to its conclusion. The drive gained 0.2. The approach from 150 yards finishes twelve feet away, a position worth 1.6, so it gained 2.8 minus 1.6 minus 1, which is 0.2. The putt drops: 1.6 minus 0 minus 1, which is plus 0.6. The three shots add to plus 1.0. The player made three on a hole whose tee baseline was 4.0. The identity holds exactly.

Miss that putt and leave it two feet short, from a position worth 1.05. The first putt is 1.6 minus 1.05 minus 1, which is minus 0.45. The tap-in is 1.05 minus 0 minus 1, which is plus 0.05. The two putts add to minus 0.4, which is exactly 1.6 minus the two strokes taken. The hole now totals zero: a par against a tee baseline of 4.0.

Drive, tee to 150 yards in the fairway0.2

Approach, 150 yards to 12 feet0.2

Putt from 12 feet, holed0.6

Hole total1

Strokes gained per shot. Invented baselines, with round numbers chosen so the arithmetic is legible rather than to report tour figures. The three shot values add to the hole total, which is the tee baseline of 4.0 minus the three strokes taken.

Show the numbers

Worked example: what each shot was worth on one par 4

Item	Value
Drive, tee to 150 yards in the fairway	0.2
Approach, 150 yards to 12 feet	0.2
Putt from 12 feet, holed	0.6
Hole total	1

Run the same cancellation across eighteen holes and a round's total strokes gained equals the expected score minus the actual score. There is no residual, no unexplained term, nothing left over and nothing swept under the carpet. Every stroke on the card is accounted for exactly once, by exactly one shot.

That identity has an underrated consequence. Strokes gained total contains no information at all beyond your score. If the baseline is the field on the day, a player's strokes gained total for the round is simply the field's scoring average minus what he shot. Post 68 in a field averaging 71 and you gained three, which anybody could have worked out by looking at the leaderboard. Anyone quoting strokes gained total as though it were an advanced measurement is quoting the scoreboard with extra steps. All of the actual information is in the split.

The four categories, and where the lines are drawn

Because shot values add, they can be sorted into buckets and each bucket adds up on its own. The standard split has four, and every shot in a round falls into exactly one of them.

Off the tee takes tee shots on par 4s and par 5s. Just those.

Approach the green takes every shot played from more than 30 yards from the hole that is not one of those tee shots. That includes par-3 tee shots, the second shot on a par 4, and on a par 5 both the go-at-it second shot and the wedge third after a lay-up.

Around the green takes every shot played from inside 30 yards of the hole that is not on the putting surface. Chips, pitches, greenside bunker shots, hacks out of the collar.

Putting takes every shot played from the putting surface, at any distance at all.

Tee to green is not a fifth category. It is the first three added together, which is the same thing as the total minus putting. It exists because it answers the question people actually want answered, which is how well did this person hit the golf ball.

Two of those boundaries deserve attention, because they behave differently.

The green boundary is physical and unarguable. The ball is on the putting surface or it is not, and no judgement is involved. Note that the category follows the ball, not the club. A ninety-foot putt across the whole green is putting. A ball six inches into the fringe is around-the-green even when the player takes the putter out and rolls it.

The 30-yard boundary is a convention. It could have been 25 or 40 and nothing fundamental would change. It sits there because the shot changes character somewhere around that distance: inside it a player is choosing a landing spot and a roll-out, outside it he is making a full or near-full swing at a number. The consequence is that a pitch from 31 yards is an approach shot and a pitch from 29 yards is a short-game shot, which is arbitrary at the margin and rarely matters in aggregate.

• 1Strokes charged against every shot played
• 30Yards from the hole where approach becomes short game
• 4Categories every shot is sorted into, exactly once
• 14Off-the-tee shots on a par 72 with four par 3s

Read across the four numbers and you have a complete account of a round, which is why they have replaced the old box score almost everywhere in tournament golf. The old columns still get printed. Nobody with access to the new ones argues from them.

Why approach play dominates

Ask which category separates good players from great ones and the answer is approach play, consistently and by a distance. There are three structural reasons and one empirical one, and the structural reasons are the interesting part, because you can derive them without any data at all.

Volume. On a par 72 laid out in the ordinary way, four par 3s, ten par 4s and four par 5s, the off-the-tee category gets exactly fourteen shots. One per par 4, one per par 5, and that is the entire supply for the day. The approach category gets the four par-3 tee shots, the second shot on each of the ten par 4s, and on each par 5 either a second shot at the green or a lay-up second and a wedge third. That is already north of twenty before a single shot goes wrong, and every green missed from beyond 30 yards adds another. More shots means more opportunity to be different from everybody else, and the category with the most shots has the most room to accumulate a difference.

The gradient. The value of precision depends entirely on how steep the baseline curve is where the ball comes to rest, and the curve is nowhere near uniform. Out at 280 yards from the tee it is nearly flat: twenty yards of extra accuracy moves the expectation very little, which is exactly why hitting fairways is worth so much less than the old statistic implied. In the band where approach shots finish, the curve is at its steepest. A ball at fifteen feet rather than thirty-five is a materially different proposition, because the make rate across that range changes fast. Approach play is the skill that decides which part of the steep section you are putting from, and it therefore converts precision into strokes at the best exchange rate on the golf course.

The ceiling. Look at the formula again. The most a shot can possibly gain is the baseline where it started minus one, which is the value of holing out from there. That is a hard ceiling, and it shrinks as you get nearer the hole.

Tee shot, 400-yard par 43

Approach, 150 yards from the fairway1.8

Chip, 20 yards from light rough1.5

Putt from 12 feet0.6

Putt from 2 feet0.05

Each figure is the baseline for that position minus one, which is the value of holing out from it. Baselines are invented and round, chosen to make the shape clear rather than to report tour figures. There is no matching floor: a shot can lose far more than it can gain.

Show the numbers

The most a single shot can gain, by where it is played from

Item	Value
Tee shot, 400-yard par 4	3
Approach, 150 yards from the fairway	1.8
Chip, 20 yards from light rough	1.5
Putt from 12 feet	0.6
Putt from 2 feet	0.05

A putt from twelve feet can gain six tenths of a stroke at absolute best, and three-putting from there costs about 1.4. The entire range of outcomes on that putt is two strokes wide, and nearly all of the probability sits in the middle of it. An approach shot from 150 yards has a range running from close to two strokes gained down to a penalty drop, and the outcomes are genuinely spread across it. Dispersion is what creates separation between players. The green is the place on a golf course where outcomes are most tightly bounded, which makes it the worst possible place to look for a lasting difference between two professionals.

The empirical finding sits on top of all three. The published research using this data consistently reports that ball striking, and approach play in particular, accounts for more of the difference between tour players' scoring than putting does, and that a player's approach performance is more stable from one season to the next than his putting performance. The mechanism for the instability is not mysterious. A green's behaviour changes with mowing height, moisture and the day's foot traffic, and a putt's outcome turns on a fraction of a degree of face angle at impact. The same stroke produces different results from week to week in a way a 150-yard 8-iron simply does not.

Finally, and this is the correction that most annoys traditionalists, strokes gained moves credit back to where it was earned. Under the old box score, a player who hit it close all week and holed a perfectly ordinary share of putts was described as putting out of his mind. His putting number under strokes gained is computed from the distances he actually faced, so it comes out near zero, and the surplus lands on the approach shots that set up all those short putts. That is not a quirk of the method. That is the point of it.

The baseline is a choice, and the choice changes the answer

Everything above assumes a baseline exists. Which one you pick is a real decision, and two numbers built on different baselines are not comparable no matter how alike they look.

The first design is a fixed baseline, pooled from a very large multi-season sample of tour shots. Its virtue is comparability: a shot from 150 yards in the fairway is valued the same way in April as in October and at every venue in between, so numbers can be added across a season and set against previous years. Its vice is that it has no idea what the conditions were. Play a week on firm greens in a gale with the rough grown out and every player in the field posts negative numbers. Those numbers describe the golf course, not the golfers.

The second design is a field-relative baseline, where the standard is what the field itself did that day. Course setup and weather drop straight out, because they land on everybody equally.

Field-relative has one structural property worth stating plainly: it is zero-sum. If the standard is the field's own play, the strokes gained of every player in the field must add to exactly zero. One player's gain is somebody else's loss, by construction. A fixed baseline forces nothing of the kind, and an entire field can finish underwater together.

What field-relative does not remove is field strength. Beating a weak field by four shots produces the same number as beating a strong field by four shots. Anybody using field-relative strokes gained to rank players across a season has to model field strength separately, and that is an entirely different piece of work from the shot-value arithmetic.

In practice the published tour numbers combine both ideas. Individual shot values come from a pooled baseline built out of previous seasons of shot data, and the round is then centred on the field for that event, so the day's conditions cancel while the shot-level values stay on a consistent scale. The consequence for a reader is simple. Establish which construction you are looking at before drawing any conclusion, because the two answer different questions. "Did he play well?" wants a fixed baseline. "Did he beat the people he was playing against?" wants a field-relative one. A brutal week under a course setup designed to protect par will push those two answers in opposite directions.

One more thing about the baseline, because it trips up nearly everybody. Par is not the baseline and never was. Par is a label a committee attaches to a hole. The baseline is what players actually do from that position. A 480-yard hole has the same tee baseline whether the card calls it a long par 4 or a short par 5, and a player who makes five on it produces the identical strokes gained figure either way. Only the word next to the number changes. This is why strokes gained comparisons work perfectly well across courses of different pars, and why "under par" and "gained strokes" are not the same statement.

Strokes gained explained for the club golfer

The arithmetic does not change one bit at handicap level. The baseline does, and getting that wrong is the single most common mistake amateurs make with this.

Measure a 15-handicap against a tour baseline and every category comes out negative. That is not a finding. It tells the player he is a 15-handicap, which he already knew. Worse, the comparison is distorted in a specific and misleading direction, because the gap between a club player and a professional is not uniform across the game. A two-foot putt is a two-foot putt for everybody. A 200-yard second shot to a firm green is emphatically not. Against a tour baseline, an amateur's putting will therefore look like the strongest part of his game almost regardless of how he putts, purely because that is where the ceiling is lowest and the gap is smallest.

Which explains a fair amount of the confident nonsense talked in clubhouses about who is a good putter.

The fix is a baseline built from players of the same standard, and handicapping already supplies the machinery to define that peer group, since the way a handicap index is calculated exists precisely to sort players onto one common scale. Several shot-tracking systems ship handicap-appropriate baselines for exactly this reason.

The data collection is far lighter than people assume. For every shot, record two things before you play it: the distance to the hole, and the lie. That is all. The finish of one shot is the start of the next, so a round is one line per shot plus the hole-out, and the categories then sort themselves without any further input from you. A rangefinder and a notepad will do it. A phone app will do it more comfortably.

What most amateurs find when they do this is not what they expected. The biggest single source of lost strokes at most handicaps is not a category at all. It is the small handful of shots that cost more than a stroke each: the tee shot out of bounds, the second penalty from the drop zone, the bunker shot that stays in the bunker, the topped fairway wood that travels forty yards.

The arithmetic of that is unforgiving. A shot that loses 1.5 needs about thirty shots each gaining 0.05 to cancel it out, and there are not thirty such shots available in a round. The distribution of shot values is heavily skewed to the left, because the gain on any shot is capped at the baseline minus one while the loss has no floor at all. Removing the left tail is arithmetically far cheaper than extending the right one, and it is usually a matter of club selection and target choice rather than technique, which is to say it is available immediately and for nothing.

Two cautions. The first is sample size. A season of weekend golf produces a few hundred approach shots, which is enough to see a large signal and nowhere near enough to see a small one. If two categories come out close, treat them as identical. The second is more encouraging. An amateur does not strictly need a peer baseline at all. Take any fixed baseline, even a rough one you assemble yourself, hold it constant, and track your own categories against it over time. The absolute level will be meaningless. The movement will not. That is the cheapest useful version of this entire exercise and it costs nothing but a notebook.

What the model cannot see

Strokes gained is the best measurement golf has, and it has real, structural blind spots. Anybody quoting it seriously should know all of them.

Lie quality is not captured. This is the largest one by some distance. "Rough" is a single word covering a ball sitting up on top of a mown first cut with a clean strike available, and a ball sitting down in thick grass growing towards the target with no chance of a clean strike. The baseline hands both the same expected value. Within one round, the difference between those two positions is worth more than most players gain from their entire short game.

The error also lands twice, with opposite signs, which is the part people miss. If your drive finishes in rough but in a perfect lie, the model has undervalued the drive, because it believes you left yourself a harder shot than you did. It has simultaneously overvalued the approach played from there, because that approach was easier than its starting position implied. Those two errors do not cancel within a hole, and they do not cancel within a tournament. They mostly wash out over a season, since nobody is systematically lucky with lies across two hundred rounds. Some newer implementations grade lies more finely using tracking data, which shrinks the problem without removing it.

The green is modelled as a flat disc. Putting baselines are usually a function of distance alone. A twenty-footer straight up a gentle slope and a twenty-footer downhill across a ridge are the same input to the model and very different propositions on the ground. Green speed also shifts through the week as the greenkeeping changes, and through the day as the surface dries out and gets walked on.

The difficulty of the target is invisible. Approach values come from where the ball finished relative to the hole, not from how hard the hole location made the shot. A pin cut four paces from the edge with a run-off behind it is a completely different problem from the same yardage to a middle pin, and the per-shot model does not know which one was played. A field-relative construction absorbs some of this in aggregate, since everybody plays the same pin, but the individual shot value does not.

Wind, elevation, stance and temperature are all absent for the same reason. Same distance, same lie category, entirely different shot.

It prices outcomes, not decisions. Strokes gained has no opinion on whether the shot attempted was the right one. A player who takes on a carry he had no business taking on, and pulls it off, collects the full value. A player who lays up correctly and makes bogey collects a negative number. The same baselines can be used to build a genuine strategy analysis, by comparing the expected values of the options available rather than the realised value of the one chosen, but that is a separate calculation and it is not what a strokes gained line reports.

It localises a problem, it does not diagnose one. The number tells you the leak is in approach play. It does not tell you whether the cause is distance control, start line, club selection or a swing that falls apart out of a particular lie. Finding that out means reading the shot-level data underneath the number, which is a specialist job, and it is a large part of why analysis has become a department rather than a hobby across professional sport.

The mean hides the shape. Two players with identical approach numbers can be entirely different golfers, one metronomic, the other alternating brilliance with catastrophe. Across four rounds of stroke play those are close to equivalent. In matchplay, or holding a lead on a Sunday, they are not remotely the same person. An average is the right tool for a season and the wrong tool for an afternoon.

This last confusion is not unique to golf. It is precisely the argument that follows expected goals in football around every weekend, and the one that dogs baseball's wins above replacement: a baseline-relative accounting of what already occurred keeps being read as a verdict on a player's quality, or worse, as a forecast.

What it cannot tell you about pressure

Look at the formula one more time and notice what is not in it. Distance. Lie. Strokes used. That is the complete list of inputs. There is no term for the leaderboard, the hole number, the day of the week, the size of the gallery or the size of the cheque.

A twelve-foot birdie putt on Thursday morning to move from 47th to 42nd is worth exactly what the same putt is worth on Sunday afternoon to win a major. The model cannot distinguish them and was never built to.

That is a design choice rather than an oversight. A version that weighted shots by their importance would immediately stop adding up to the score, and the exact telescoping described earlier is the property that makes the whole system trustworthy. You can have importance weighting or you can have an accounting identity. Not both.

Can strokes gained be used to test whether performing under pressure is a repeatable skill? In principle yes, and the design is obvious. Define a set of high-stakes situations, compute each player's strokes gained inside them, compare it with that same player's number outside them, and check whether the difference persists from one season to the next. In practice the sample is punishing. A player genuinely contends in a handful of events a year, and the closing holes of those events contain a few dozen shots between them. The confidence interval around a few dozen shots is wide enough to accommodate almost any story you would like to tell, which is why the argument never resolves and why both sides can always find a season that supports them.

There is a subtler problem underneath the sample size, and it is more interesting. Under pressure, the shot a player attempts changes. He aims twenty feet away from a tucked flag, takes an extra club to cover the front bunker, hits three-wood from a tee where he would otherwise have hit driver. His strokes gained falls. Whether that measures a failure of nerve or a correct change of strategy is genuinely unclear, and the model, which has no view whatsoever on strategy, is quietly scoring the second one as though it were the first.

So what a strokes gained line tells you about a player under pressure is what happened. Not what will happen next. The distinction sounds pedantic right up until somebody builds a Ryder Cup pairing on it.

Reading a strokes gained line without being fooled

Five checks, in the order that matters.

Which baseline produced it. Fixed or field-relative. Two numbers from different constructions are not comparable, and a fixed-baseline number is not comparable across two weeks played in very different conditions. If the source does not say, be suspicious of the source.

How many rounds went into it. A four-round tournament is a few hundred shots split four ways, and the around-the-green category is far and away the smallest of the four, since a player only plays those shots when he has missed a green. A week's figures describe a week. Season-length samples, and preferably longer, before you believe anything about a category.

Tee to green against the total. If a player's total is being carried by his putting, the safe expectation is that the total falls, because putting is the least stable of the four. If it is being carried by approach play, it is far more likely to hold. This one check explains most of the hot-streak stories that quietly stop being written three weeks later.

The tail, not just the mean. The average of a category tells you about the season. The count of shots that cost more than a stroke each tells you what is likely to happen on a Sunday afternoon with a one-shot lead, and those two things are not the same number.

What zero actually means. Zero is not a bad round. Zero is exactly the standard the baseline describes, and on tour that standard is the best players in the world. A club golfer measured against a tour baseline will post large negative figures in every column and learn nothing at all from any of them, which is the most common way this measurement gets misused.

Strokes gained did not make golf harder to talk about. It made it possible to talk about a round without the previous shot lying about the next one. The categories, the competing baselines, the arguments over which construction to use, all of it is bookkeeping stacked on one honest question, asked before and after every shot: from where the ball is right now, how many strokes should this take?

Everything else is a consequence of asking that question twice and subtracting.