Lost on Dead Balls

Gimcheon finished the 29-round stretch on 4 wins, 18 draws, 7 defeats — 30 points, eleventh of twelve. The draw rate of 62.1% is not merely the highest in the league. It is half again as high as the next club.

The character of those draws matters more than the count. Fifteen of the eighteen finished 1–1 or 0–0. Across the whole 29 matches, 24 — 82.8% — ended within a single goal. These were not shoot-outs that happened to level out. They were quiet.

And the team was not hanging on. It was ahead. Gimcheon scored first in twelve matches and won three of them. Counted as goals rather than matches, twelve conceded goals erased a lead; in the ten of those matches that were not subsequently won, the club dropped 21 points. Hold those ten and the season reads 51 points — second place.

So: where did those goals come from?

28 scored, 36 conceded, a goal difference of −8. Split that by how the goals were created and it stops looking like one team. The classification below is taken verbatim from the official match-report goal descriptions, not inferred.

This club did not lose its season in the run of play. It lost it when the ball was stationary.

On open play alone the record is 26 : 23. Those are not the numbers of an eleventh-placed side. The whole of the −8 is the −11 on dead balls.

Volume is not the problem. Gimcheon committed the fourth-most fouls in the league yet faced fewer corners than the league average. Opponents were not being allowed deep. What failed was conversion.

Reconstructed second by second, the eight set-piece goals conceded (penalties excluded) break down brutally: seven were finished within three seconds of the dead ball, and four were scored on the first contact. This is not a second-ball problem. It is a first-contact problem. Across 111 corners faced, the first Gimcheon defensive touch was identified 108 times — on average 6.1 units from their own goal line, which is the edge of the six-yard box, and 42 percent of those touches were clearances.

Cut the season by score state and the defence looks competent. Over the full 29 matches Gimcheon conceded 36 against 41.1 expected — five goals to the good. That entire surplus is surrendered in one state.

Level and behind, this team concedes less than expected. Only when ahead does the relationship invert. And the goals that went in were not good chances: while leading, the mean expected value of a conceded shot was 0.294, lower than the 0.355 of everything else. Bad chances went in.

What kind of bad chances? Nine of the fourteen (64 percent) were headers, or came from a cross or a headed assist. Across the other 22 goals conceded, that figure is 27 percent (Fisher exact, two-sided, p = 0.030).

One popular explanation dies here. "They sit back when ahead" is not in the data. The mean position of a Gimcheon defensive action is 28.5 when leading, 29.1 when level, 29.6 when behind — effectively identical. The block height does not move. What moves is possession: leading, the team plays 386 passes per 90 against 462 when level, a drop of 16 percent.

If the defensive set-piece record is bad, the attacking one is worse. 147 corners produced one goal — a conversion rate of 0.68 percent against a league average of 1.77 percent, and the lowest of the twelve clubs. Seoul took exactly the same number of corners, 147, and scored twice.

League-wide, 24 of the 26 corner goals scored in 2026 were headers, and 25 of 26 were indirect. Corners are a heading contest. Here is where Gimcheon puts the ball.

The consequences are measurable at every step. The delivery reaches a teammate half the time (50%). A shot follows within twelve seconds a third of the time (34%). But the first contact goes to the opponent 53 percent of the time, and the single most common first contact in the whole sample is an opposition clearance, 44 times. The shots that do come are poor: 54 corner-origin shots at 17% on target, against a team average of 29.5 percent, and the lowest expected value per shot of any origin.

A league that scores from corners with headers, a team that wins 48.5 percent of its aerial duels, and a delivery pattern that puts 56 percent of the ball into the best-defended square of the box.

Which raises the obvious question, and it is the one everybody asks about this club first. Is this just what happens when you are the only team in the league that cannot sign a foreign target man? That is the next section.

Here is the explanation every Korean analyst reaches for, and it is a good one. Korean football has a shortage of tall centre-forwards. K League clubs solve it the way clubs everywhere solve a shortage: they buy it abroad. Gimcheon cannot. Foreign players cannot serve in a military unit, and the club has fielded zero in every season from 2022 to 2026. So of course it loses the heading contest.

To test this I built a second dataset: K League 1 and 2, 2022 to 2026, four evenly spaced rounds per league-season, 248 matches, 3,762 player-seasons with height, age and position, taken from the matchday squad lists. Nationality was inferred from names and then sanity-checked against the quota: no club exceeded its allowance in any season, and the minimum of zero was always Gimcheon.

A threshold for "can threaten in the air" needs a basis. The mean height of a K League centre-back — approximated as the taller half of defenders, to exclude full-backs — is 187.5 cm. Rather than pick a round number below that, I checked it against outcomes: for the 30 players in the 2026 sample with 15 or more aerial duels (1,204 duels in total), a weighted regression gives

win rate (%) = 0.790 × height − 94.0

which crosses 50 percent at 182.4 cm and 55 percent at 188.7 cm. This series therefore uses two lines: 185 cm, at which a forward can contest a centre-back, and 188 cm, at which he starts to win.

Foreign players are 14.5 percent of all player-seasons in the sample. Their height advantage is not spread evenly across the pitch — it exists in two positions and is negative in a third.

The same pattern read from the other side: of every player in the sample standing 185 cm or more, 42 percent of the forwards are foreign, against 13 percent of the defenders and 5 percent of the midfielders. A group that is 14.5 percent of the league supplies four in ten of its tall forwards — roughly a three-fold over-representation. Clubs do not import height in general. They import a centre-forward who can head the ball.

And the dependence is accelerating, in step with the rules. The Asian quota was abolished for 2025, the squad limit rose to six, and for 2026 the holding limit was removed entirely with five allowed on the pitch at once in K League 1.

There is exactly one regular domestic target man of 188 cm or more in the entire division: Lee Ho-jae of Pohang.

That is the true size of the shortage, and it is worse than the phrase "target men are scarce" suggests. It is not that they are rare. It is that the domestic supply is one player, and the other six places in that category are filled from a market Gimcheon is legally barred from.

So the first half of the hypothesis is correct in every particular. The league does import height, it imports it specifically at centre-forward, the domestic pool is essentially empty, and the dependence has roughly doubled in two seasons.

If the mechanism explains Gimcheon's aerial record, the club should be short of tall forwards. It is not. In 2024 it held four forwards of 185 cm or more — joint most in the league, with no foreign players. In 2025 and 2026, three, against a league average of 2.3.

Holding a player is cheap, though. Playing him is the test.

Gimcheon gives 9.2 percent of its minutes to forwards of 185 cm or more — third in the league, above average, with no foreign players at all. Lee Gun-hee (186 cm, 2,077 minutes) played more minutes than any other tall forward in the division.

And the decisive test: across the twelve clubs, the number of tall forwards a team holds does not predict its corner goals — the correlation is −0.19, pointing the wrong way, and with aerial duels it is 0.04. Daejeon scored five corner goals, the league's best, with two tall forwards. Pohang scored one with five.

If the club has the height and still loses the heading contest, the next place to look is whether the height is on the field when it matters. Gimcheon took 147 corners in 2026. Here is who was available for them.

The tallest attacking player at the club was on the field for under a third of its own corners.

Jung Jae-min is 192 cm. He played 577 minutes across 14 appearances — an impact substitute. There is nothing inherently wrong with that; a 192 cm forward is not automatically a starter. But it does change the shape of the problem. The constraint that matters here is not the transfer market. It is the team sheet.

Which makes the practical conclusions above stronger rather than weaker. You cannot blame a resource you are barred from buying. You can ask why the one you already own watched 101 of your corners from the bench.

Everything above makes a case. This section is about the checks, because a football analysis that only reports what survived is not showing its work. Two popular explanations for Gimcheon's season died under testing, and so did one of my own findings.

All three are worth reading not for the conclusions but for the shape of the test. In each case the pattern was visible at the aggregate level and dissolved when the right control was applied.

This is the explanation that follows naturally from Part 1 of this series. Eighteen-month terms, two intakes a year, players arriving and being discharged mid-season — the organisational cost seems obvious.

It is testable, because the rest of the league is a control group. I used 24,894 player-match records from K League 1, 2024 to 2026, and ran three checks.

The third is the direct one. Between 2024 and 2026, 42 players appeared at least six times for Gimcheon and at least six times for another club. Comparing a player against himself removes individual ability entirely.

The spread is wide — Won Ki-jong is +17.1, Kim Tae-hyun −6.8 — but the players with the largest samples sit within three points of zero, and the central tendency is a small positive. There is no evidence that going to Gimcheon makes a player worse.

And the 2026 change runs the other way from the story. The share of minutes taken by the eleven most-used players rose from 52.8 percent in 2025 to 70.0 percent in 2026, against a league average of 68.5. The squad did not churn more. It settled.

We established above that Gimcheon concedes 1.16 goals per 90 above expectation while leading, and nowhere else. The intuitive reading is that the team drops deep, invites pressure and gets punished. That is checkable: if a side retreats, its defensive actions happen closer to its own goal.

The block does not move. The team that concedes 2.58 per 90 while leading defends at the same height as the team that concedes 0.74 while level.

What does change is possession. Leading, Gimcheon plays 386 passes per 90 against 462 when level — a 16 percent drop. It is not retreating. It is giving the ball away, which is a different problem with a different fix.

This matters practically. "Stop dropping so deep" is a instruction that would have had no purchase here, because the deep block was never the change. The change was that the team stopped keeping the ball.

Working through the 496 crosses Gimcheon faced, a clean pattern appeared. When leading, crosses into Gimcheon's right side rose from 7.60 to 11.23 per 90 — up 48 percent — while the left side did not move. The lead-erasing goals leaned the same way, five on that side against two on the other. It looked like a specific, coachable weakness, and I wrote it up as one.

Two checks took it apart.

First, conversion. Volume went right, but goals did not. Across the season the concession rate was 4.1 percent per cross on the left and 2.8 percent on the right — pointing the opposite way from the volume, and a binomial test returns p = 0.32. There is no demonstrable left-right difference in danger at all.

Second, stratification. Splitting the 496 crosses by phase showed where the increase actually came from.

The dominant phase rises symmetrically. The apparent right-side bias was the sum of three small residual categories — second-phase set pieces, turnovers, throw-ins — each worth 0.3 to 0.5 crosses per 90. And opponent by opponent it is not even consistent: of the seven clubs against whom Gimcheon led for 30 minutes or more, five produced more right-side crosses and two fewer, with Daejeon (+8.63) and Pohang (+7.86) carrying most of the effect against Ulsan's −10.84.

A pattern that is real in the aggregate, absent in the dominant sub-phase, and inconsistent across opponents is not a weakness. It is an artefact of adding things together.

Four explanations were removed along the way: no foreign target man, squad rotation, sitting deep, and a right-side defensive weakness. What survives is narrower and more actionable than any of them.

None of those require a signing this club is forbidden from making. All of them are things a coach who receives his squad rather than choosing it can still work on.

The problem was never who was playing. It was what they did in the air and in the box.