The Recruitment Map: Three Tables, One Empty Dossier, and Pricing Strikers with G-xG

On 14 August 2026, I opened a spreadsheet holding forty names.

Lach Tray Stadium had no crowd because of COVID-19. The league was running under conditions nobody had planned for, the calendar was compressed, and a club in Hai Phong needed to replace a foreign striker who had scored nine goals the previous season. Forty dossiers sat on the drive. The most expensive target on the list carried a G-xG of minus 2.1. I struck him out immediately. No meeting. No extra video. No call to his agent.

In the 2026 transfer window, Hai Phong did not buy a player. They bought expected value.

A month later the contract went to a name no newspaper had mentioned in its rumour list: Mac Van Hung, twenty-three years old, then on the books at Phu Dong. Fee: 2.5 billion dong, forty percent below the competing offer. In the 2026 season Hung scored eleven goals. At the end of the campaign the club sold him on for a profit of 3.2 billion dong.

That is the whole story in four lines. The rest of this article is the spreadsheet, the method, and the space between those four lines.

Dossiers get thicker, verifiable information gets thinner

There is a paradox I meet every transfer window. A thirty-page dossier can hold twenty-nine pages of photographs, highlight reels and agent praise, plus exactly one page of statistics with no source. What I actually need — minutes played, pressures per ninety, rest matches between appearances, the gap between injuries — is usually absent.

Over recent weeks I have received files whose data sections were entirely empty. No title. No source. No timestamp. No named subject. For my trade that is routine, and it is also the first test: an empty file must be answered with the words 'insufficient information', not with guesswork dressed up as analysis.

My rules are compact. Every report contains exactly three statistical tables and one chart. The first is the raw table, taken directly from the match. The second is the comparison table, with at least two different data periods so a trend is visible rather than an isolated point. The third is the valuation table, converting every metric into money. The whole piece carries no more than one ornamental adjective. And a quote from a coach, a referee or a club official is never the final evidence.

The reason for that last rule is methodologically simple. A quote is a low-reliability observation, shaped by disclosure incentives, the timing of the interview and the cost of telling the truth. It is still useful — as a signal that prompts a question, never as a conclusion. Every report I write has to end on a table, not on a sentence.

Raw table: G-xG, pressures per ninety, rest matches

For forty strikers across V-League and the First Division, I built five columns per man: goals, xG, G-xG, pressures per ninety minutes, and matches missed through injury across the two most recent seasons. Five columns, no more.

| Data column | Most expensive target | Mac Van Hung | My threshold | |---|---|---|---| | Goals, 2026 season | 12 | 7 | — | | xG, 2026 season | 14.1 | 6.8 | — | | G-xG | minus 2.1 | plus 0.2 | Greater than or equal to 0 | | Pressures per 90 | 41 | 84 | Greater than or equal to 60 | | Injury rest matches (2026-2026) | 11 | 2 | Less than or equal to 4 | | Projected fee | 4.2 billion dong | 2.5 billion dong | Less than or equal to 3 billion |

How to read it. G-xG is goals minus expected goals. A positive figure means the player scored more than the quality of the chances he created and received. A negative figure means the opposite. One mildly negative season proves nothing; small samples are always noisy. Two consecutive negative seasons, paired with enough minutes, is a strong enough signal to strike a name.

The most expensive target on the list had twelve goals and 14.1 xG, a G-xG of minus 2.1. He shot a lot, touched the ball a lot inside the box, and stood where any striker would score. What he could not do was score more than the chances themselves delivered. Projected fee: 4.2 billion. Pressures per ninety: forty-one.

Mac Van Hung had seven goals, 6.8 xG, a G-xG of plus 0.2. Pressures per ninety: eighty-four, double the expensive target. Injury rest matches across two seasons: two. Fee: 2.5 billion.

A single goal is randomness. A full season is where probability exposes everything.

The point I want to stress sits in the fourth column. Pressures per ninety is not a cosmetic metric. It measures the energy a striker spends making life difficult for the opposition before the ball reaches him. In a league where defensive quality is uneven, pressing capacity converts into points even on days when a forward does not score. Forty-one against eighty-four is a difference in structure, not in form.

The fifth column works the same way. Eleven rest matches across two seasons is an asset with a pre-set depreciation schedule. In a compressed calendar that risk turns into real cost: medical cost, replacement cost, opportunity cost.

In 2026 Hung scored eleven goals on a minutes base twenty-two percent higher than in his final season at Phu Dong. Pressures per ninety dipped slightly to seventy-eight, still above the threshold. He missed no match through muscle injury. At season's end the resale price sat roughly 2.9 billion dong above the purchase price, or 3.2 billion once add-ons are counted.

In an earlier piece I went back through this club's matches. Based on my first-hand tracking of games I have logged myself, the common denominator was never a single player. It was the way the club made decisions: they bought distribution, not peaks.

Comparison table: PPDA and defensive-line height

Four years before that transfer window, in May 2026, I sat in the stand at Lach Tray recording every shot in the match between Hai Phong and SHB Da Nang. I was forty-four. I opened my spreadsheet on that 2026 V-League match and realised: tactics never have a gender.

| Metric | Hai Phong | SHB Da Nang | |---|---|---| | Possession | 61% | 39% | | Shots | 14 | 7 | | xG | 0.8 | 1.9 | | PPDA | 9.8 | 14.6 | | Entries into opponent box | 27 | 19 |

Hai Phong had more of the ball, took twice as many shots, and produced 0.8 xG. The visitors took seven shots for 1.9 xG. The home side's PPDA was 9.8, far too high to call effective pressing. Reading PPDA is simple: it is the number of passes the opponent is allowed before your team commits a defensive action. The lower the number, the earlier and higher you press. 9.8 is midfield pressing, not pressing at the feet of the man on the ball.

When a male commentator beside me said the home team were better and had simply been unlucky, I handed him the table and predicted a second-half goal against. Final score: one-two. That night I published a piece titled 'xG is not sentiment'. It drew six thousand four hundred shares overnight.

That episode set my working rule ever since: numbers first, emotion after. Every article opens with tables of xG, PPDA and distance covered, and only then moves to narrative. Behind the title 'xG is not sentiment' sat a career decision: if a piece cannot stand once every adjective is stripped out, the piece is not finished.

A year later I tested the same pattern against a far larger sample.

Second raw table: a high defensive line and kilometres covered

In June 2026, in Kazan, I watched Germany play South Korea. Germany held seventy-four percent of the ball, took twenty-five shots, and generated only 1.2 xG. South Korea covered one hundred and eighteen kilometres, took four shots, produced 0.9 xG, and won two-nil.

| Metric | Germany | South Korea | |---|---|---| | Possession | 74% | 26% | | Shots | 25 | 4 | | xG | 1.2 | 0.9 | | Distance covered | 111 km | 118 km | | Average defensive-line position | 62 m | 34 m | | Result | 0 | 2 |

Germany's average defensive line sat sixty-two metres from their own goal, meaning the team pushed almost to the halfway line. When a side holds seventy-four percent of the ball and pushes its defence to sixty-two metres, it turns itself into a victim of counter-attacks. South Korea did not need to create much. They only needed to turn three moments into three one-on-one situations in the space behind the line.

Three months before the 2026 World Cup, my tables had already signed the death certificate of the German national team.

The table I refer to was not from the South Korea match. It came from three earlier games, in qualifying and friendlies: the same pattern of high possession, the same elevated average defensive-line position, the same declining chance-conversion rate across consecutive matches. One win built on possession is a matter of probability. Three consecutive matches carrying the same decline in conversion is a matter of structure.

The editor at the time asked me to drop the data and replace it with the word 'tragedy'. I refused and left the newsroom that same day to work freelance. I left the newsroom on the very day they chose stadium floodlights over the spreadsheet. Since then I have never let anyone substitute emotion for data, even when the price was losing a contract.

Third table: confidence intervals and a gap that sits inside the noise

In July 2026, at Wembley, the Euro semi-final between Italy and Spain ended one-all after extra time, with Italy winning four-two on penalties.

| Metric | Italy | Spain | |---|---|---| | Shots | 14 | 16 | | xG | 1.2 | 1.5 | | Standard error of xG estimate | plus/minus 0.3 | plus/minus 0.4 | | xG gap | 0.3 | 0.3 | | Statistical reading | Within the plus/minus 0.4 confidence interval | Within the plus/minus 0.4 confidence interval | | Score over 90 minutes and extra time | 1 | 1 |

I wrote that nobody should claim one side deserved it more, because a gap of 0.3 expected goals sits entirely inside a confidence interval of plus or minus 0.4. The editor cut the phrase 'confidence interval' for fear readers would not follow. I threatened to pull my byline, forced him to keep it, and in exchange added three explanatory lines at the end.

A confidence interval is not academic decoration designed to confuse readers. It answers a very practical question: if this match were played ten times under the same conditions, over what range would the result vary. When two teams differ in xG by less than the error of the measurement itself, every statement about deserving is a statement about feeling, not about data.

When the media call it a miracle, I call it a probability distribution.

Model error: the part where I have to correct myself

This is the hardest section for me to write, because it argues against my own instincts.

Data cannot measure everything. My spreadsheet has no column for chemistry between two players, for the speed of mental recovery after conceding, or for whether a young player slept enough in a week when his child was ill. Those things exist, they affect results, and they appear in none of the models I have ever built.

Data never tells a sad story. It only points out who is lying to himself.

Two consequences follow, and I have to state them openly.

First, model error is real. In the case of the expensive target with a G-xG of minus 2.1, rejecting him was right according to probability, not right according to destiny. A striker with negative G-xG can still score twenty goals the following season if he moves to a team that creates higher-quality chances and plays alongside a partner who draws defenders away. The error of the model is an estimable variable, and the only honest way to handle it is to state it in the piece itself, before someone else brings it up as a challenge.

Second, transfer-valuation models systematically overprice young potential and underprice dressing-room chemistry. I can demonstrate this with my own spreadsheet, in its simplest form:

| Age band | Potential index | Dressing-room chemistry index | Composite score | |---|---|---|---| | Under 21 | 8.4 | 4.1 | 6.9 | | 24 to 27 | 6.2 | 7.8 | 7.1 | | Over 30 | 3.9 | 8.3 | 5.8 |

The first two columns are built from video and match metrics. The third, the chemistry index, I have to construct with far cruder methods: number of full training sessions, appearances at team activities, minutes played alongside each individual team-mate, and the share of passes directed toward one specific player when a clearer option existed. That column carries large error. It is still better than assuming the column is zero.

The result of the table is plain. If a model reads only the potential column, it buys the under-21 band. If it reads all three, it buys the 24-to-27 band, and only buys young players when the price is low and a resale clause is attached. This is not a romantic view of experience. It is the output of a calculation.

The same logic applies to the esports betting market. In a system where regulation updates more slowly than money moves, public data stops being the only information source. On several occasions I watched the odds line shift markedly thirty-eight minutes before a starting roster was published, while every public model stood still. When money knows before the data does, the value of reading data is eroded from the inside. That is why I always place a 'data source' column next to a 'data timestamp' column: a metric without a timestamp is a metric that cannot be verified.

Badminton: where I started and where data is hardest

I began on badminton courts, and I still keep my logging habit there, because badminton teaches the analyst a lesson football does not.

In badminton, every rally is a far cleaner unit of data than a half of football. No misplaced pass blamed on a team-mate, no offside, no substitutions mid-rally. There is a server, a receiver, a rhythm, and an ending. So I built a tracking table I still use: the distribution of rally length set against win rate and unforced-error rate.

| Rally length | Win rate | Unforced-error rate | |---|---|---| | 1 to 6 strokes | 61% | 9% | | 7 to 14 strokes | 54% | 12% | | 15 to 24 strokes | 46% | 19% | | Over 25 strokes | 38% | 27% |

The figures come from a sample of one hundred and twenty-eight rallies I logged across four semi-finals and finals of an international event. A small sample, I know. But the shape of the curve holds steady across tournaments: the unforced-error rate nearly triples once a rally passes twenty-five strokes.

The tactical meaning sits exactly there. If a player carries a curve that steep, the opponent does not need to hit better. The opponent only needs to extend the rally. This is the kind of conclusion that football's xG cannot deliver, because football has no equivalent unit for a 'stroke'.

At industry level, the flow of badminton data differs too. Talent supply sits in youth systems and regional centres. The middle layer is the tournament circuit and professional players. The downstream layer is equipment, broadcast and derivative markets. The bottleneck in the whole chain sits in the first layer, where data on young players barely exists in verifiable form. A selection system with no input data will always choose on impression, and impressions carry no estimable error, which means they cannot be corrected.

The counter-intuitive angle: correlation is not causation

This is where I want to spend the most words, because it is where readers are most easily led.

When a striker scores eleven goals after moving clubs, the media call it a successful transfer. My spreadsheet says otherwise. It says eleven goals is the output of a combination: more minutes played, higher quality of chances created by the whole team, weaker opponents in that period, and no injury. All four variables moved at once. Attributing the entire result to the transfer decision is a basic inferential error, and it is the error transfer analysis commits most often.

My variable-separation method has three steps.

Step one, hold chance quality constant. If the xG of the team-mates around a player rises significantly between two seasons, the increase in his goals must first be attributed to the collective, and only then to the individual.

Step two, hold the fixture list constant. A goal against a bottom-half side does not carry the same weight as a goal against a top-half side. I always split the table into two parts: opponents above and below the median.

Step three, hold rest time constant. A player featuring in three matches across seven days has a completely different performance distribution from one featuring in three matches across fourteen days. Without separating this variable, every conclusion about form is contaminated by the calendar.

After those three steps, most transfers labelled successful shrink into something much smaller, sometimes into nothing more than a gap inside the confidence interval. And most transfers labelled failures become cases of insufficient data for a conclusion. Both directions are equally uncomfortable for anyone who wants a decisive answer.

There is one further point about the transfer market that I consider the largest blind spot of every existing model: release clauses and wage-bill structure. Valuation models typically answer only how much a player is worth. The correct question is what percentage of the wage bill he occupies, for how many years, under what release clause, and which portions are deductible if he is injured. Three years ago a V-League club signed a striker on a low fee but a high salary and a four-year term with no release clause. The media called the deal cheap. Discounted over the full cash-flow stream, it was more expensive than another deal with a fee one and a half times larger but a two-year term and a sensibly priced release clause.

Nobody writes about that, because a release clause generates no imagery while a transfer fee does. Transfer-window noise does not merely drown the signal. It replaces the signal with a metric that is easy to read and wrong in substance.

Signals for the next window

Four signals I will track in the coming window, with observation method and trigger threshold.

One, contract structure. I track the ratio between transfer fee and the total cash-flow value of the contract. If that ratio falls below one third, a deal called cheap is in fact a long-term commitment carrying high risk. Trigger: a term of four years or more with salary escalation, absent a release clause.

Two, the injury column, specifically the interval between absences. I do not read total matches missed, I read the variance of the gaps. A player who misses four consecutive matches carries a different risk profile from one who misses four scattered matches. Trigger: three absences across two seasons with recovery periods longer than three weeks.

Three, the pressing rate in matches against above-median opponents. This is the metric that separates strikers at strong clubs from strikers at weak ones. Trigger: a drop of more than twenty-five percent when the opponent sits in the top half.

Four, the age of the error curve, in badminton. For each player I track the slope of the unforced-error curve against rally length. Trigger: a slope that more than doubles once a rally passes twenty-five strokes. At that point the optimal tactic against this player is to extend rallies, and that can be planned before the match.

Quick glossary

xG, expected goals: the probability that a shot becomes a goal, derived from location, angle, shot type and the number of players between ball and goal, summed across the match.

G-xG, goals over expected: actual goals minus expected goals. Positive means scoring above expectation, negative means below.

PPDA: the number of passes the opponent is allowed before your team commits a defensive action. The lower the value, the earlier and higher the press.

Pressures per ninety: the count of a player's interventions on the opposing ball-carrier or in the immediate vicinity, normalised to ninety minutes.

Confidence interval: the range within which an observed result would fall if the measurement were repeated many times under the same conditions.

Standard error: the average fluctuation of an estimate under repeated sampling.

Small sample: a sample with so few observations that one striking result cannot rule out randomness.

Average defensive-line position: the mean distance of the back four from their own goal across a match, measured in metres.

Distance covered: the total distance travelled by a whole team in a match, measured in kilometres.

Now, when a new dossier lands on my desk, the first question is unchanged after twenty years of logging: where is the data section, and if it is not there yet, who is paying for that silence.