38,899 on-ball events from Gwangju's 27 matches, checked against 110 K League 1 matches. The summer signings really did change what happens inside the box. What never changed was the road to the box — and it narrowed further whenever the team fell behind.
The sentence you hear most about Gwangju is “they survived on teenagers, then signed heavily in summer and still can't win.” The first half of that is confirmed by the data. The second half is wrong. The signings worked, and performance genuinely improved. Exactly one thing did not.
This piece narrows in on that one thing, seven times. It starts with the conditions that were already fixed before a ball was kicked, confirms what actually changed in July, breaks down why the change never became points, and arrives at a single number that did not move once in twenty-seven matches. It ends where the evidence shows that number is a choice, not a limitation — and then at the point where this data can no longer answer.
In September 2025 FIFA imposed a one-year registration ban on Gwangju; it lifted in June 2026. Players could be signed but not fielded, and several waited half a season. Park Won-jae (33), signed on 2 January, made his debut on 11 July.
Teenagers filled the gap. Park Seong-hyeon (28), 16, was on the pitch two days after signing a semi-pro contract, setting the all-time K League 1 & 2 record for youngest appearance (16 years, 8 months, 21 days). Gong Bae-hyeon (26), an 18-year-old centre-back, played a full 90 in round 2.
Minutes share = a club's total playing time taken by that age band. League pool of 110 matches (11–27 matches covered per club). Twelve players aged 19 or under appeared in the league all season; four of them were Gwangju's, and Gwangju alone accounted for 53.3% of all under-19 minutes played in the league.
Under-19 minutes share fell from 12.98% during the ban to 1.05% after it — a 12.4-fold drop. Rounds 17 through 24 were eight consecutive matches at 0.0%, and neither Gong Bae-hyeon (26) nor Kim Yun-ho (25) played a single minute across the final eleven fixtures. That they disappeared the same week replacements arrived says the selection was scarcity, not policy.
⚠ This is circumstantial. Injuries, opponent profiles and fitness are not in the data. What the data shows is that the timing lines up exactly.
The common read is “desperate for points, so they bought big men and went direct.” Fridjonsson (11) is 196cm, van Groonsven (4) 194cm, Aidyl (18) 188cm. The picture fits. The data says the opposite.
After July, Gwangju cut aerial duels by 25% (29.2 → 21.8 per match) and roughly halved the share of box shots that came straight off a cross, from 35.2% to 19.4%. Average pass length (17.8 → 18.2m) and long-ball share (11.4 → 11.6%) barely moved. They did not go direct — they went shorter, inside the box.
The roles bear this out. The only genuine aerial target is Fridjonsson (111 aerial duels, 59% won), and he was not a summer signing — he had been there since round 1, and his minutes actually fell afterwards. Aidyl (18) wins 32% of his aerials, which is not a target man's number. Ba Loua (50), with 29 crosses and 4 aerial duels, is a winger.
Restricted to the 10 clubs Gwangju played in both halves of the season — 25 matches — so opponent identity is held constant. ⚠ Opponent form and home/away balance are not controlled.
Shots inside the box rose from 3.6 to 6.0 per match (+67%), expected goals per shot from 0.091 to 0.110 (+21%), and the share of shots on target from 32.4% to 40.4%. A wide crosser, a box finisher and a ball-playing centre-back did what that combination should do.
Expected points per match rose from 0.72 to 0.96, up 33%. Actual points went from 0.47 to 0.50. The gap between the two widened, from −0.25 to −0.46.
To test whether that is simply bad luck, I built 88 rolling 12-match windows across ten clubs and computed the same gap for each. Gwangju's second-half figure of −0.46 sat in the bottom 4.5% — and more tellingly, seven of the eight worst windows in the entire league were Gwangju's. Short-run misfortune scatters across teams. This does not.
Expected points from a Poisson simulation on each match's xG. Actual results taken from official match reports, not from event data. In the matches they were outplayed, they got roughly what they deserved. Everything was lost in the seven matches they were level or on top.
The leakage narrows to three places.
Red cards, by contrast, are not the explanation. All three came in two matches, and the other 25 matches still yield only 0.52 points per game.
That is the result. Now the structure.
While twelve players debuted, the attacking signature changed and the box improved, final-third entries stayed at 56.3 then 56.9 per match. The share of middle-third possession that gets pushed forward went 0.11 → 0.11 — identical to two decimal places.
Thirds are the pitch divided lengthwise into three, with attacking direction normalised. Figures are per match, rounds 16–27.
The share of successful middle-third passes that reach the final third — Gwangju's 10.4% ranks 12th of 12. Yet their volume of middle-third passing, 198.3 per match, ranks 7th. They touch the ball plenty. It just doesn't go forward.
110-match league pool. Coverage per club ranges from 11 to 27 matches; the small-sample clubs move around more.
One popular diagnosis can be dismissed first. “The centre-backs hog the middle third, so nothing goes forward” does not hold. Incheon (56.4%), Seoul (55.8%) and Pohang (56.1%) all give their defenders as much of the middle third as Gwangju do (54.8%). The correlation between defender share and team progression rate is r = +0.20 — effectively none.
What separates the clubs is whether those defenders play forward.
| Club | Defender share | Defender progression | Team progression |
|---|---|---|---|
| Gangwon | 53.2% | 16.4% | 17.2% |
| Seoul | 55.8% | 16.2% | 16.5% |
| Incheon | 56.4% | 13.6% | 14.0% |
| Ulsan | 50.3% | 13.2% | 13.6% |
| Jeju | 51.6% | 12.9% | 14.0% |
| Jeonbuk | 48.6% | 9.9% | 11.7% |
| Pohang | 56.1% | 9.8% | 10.8% |
| Gwangju | 54.8% | 8.7% | 10.4% |
The top three ball-users in the middle third account for 43.3% of it at Seoul, 41.6% at Incheon and 37.9% at Gangwon. At Gwangju it is 25.7% — the most dispersed in the league. Yet those three Gwangju players progress at 13.8%, which is fourth-best in the division. The ball never concentrates on the people who move it forward; the other 74% of it circulates at around 9%. (Concentration vs team progression rate: r = +0.58.)
One check was needed first. Gwangju play 46.7% of their middle-third passes while losing (league: 25.8%), and only 3.7% while ahead (league: 18.8%). With a distribution that skewed, any average is distorted. So I recomputed everything split by score state.
The result was not what I expected. League teams raise their progression rate when behind — defenders go from 10.6% at level to 14.2% when losing, +3.6 points. Gwangju's midfielders behave normally, 12.2% to 15.0%. Only Gwangju's defenders go the other way: 10.0% down to 7.4%.
Sample: league defenders 11,024 passes level and 5,378 behind; Gwangju defenders 1,426 and 1,419. Score state evaluated at the moment of each pass.
Seven of Gwangju's nine defenders progress less when the team is losing. Kim Jin-ho (27) −7.5, João Pedro (17) −5.5, Ha Seung-un (9) −2.8, van Groonsven (4) −2.8, Min Sang-gi (39) −2.1, Kim Yong-hyeok (24) −2.1, Gong Bae-hyeon (26) −1.4 percentage points. Only Kwon Seong-yun (22) (+7.1) and An Yeong-gyu (6) (+0.5) go the other way. Individual temperament does not line up in one direction like that.
And this has a familiar face elsewhere in the season. In the first fifteen minutes after half-time while already losing, Gwangju concede at 2.07 times their own losing-state average. In the phase where they have to chase, they cannot get forward and they open up behind. Both numbers look like one thing: there is no plan for the chasing phase.
One last check: are Gwangju's defenders unable to play forward, or do they choose not to?
Compare only passes attempted at the same distance from the same lane and Gwangju's defenders match the league or beat it — 27.1% against 18.0% for 25–35m passes started centrally. The technique is not the problem. The difference is in which direction they decide to hit.
Defenders' 15–25m passes started in the middle third. Gwangju n = 1,130 · Ulsan 932 · Seoul 453 · Gangwon 409. This distance band is roughly 39% of all defender passing.
When league teams lift their progression rate while losing, pass length (16.2 → 16.4m), the mix of distances and the target lanes all stay essentially the same. What changes is how often the same pass gets through — 10.7% to 14.7% at 15–25m, 59.3% to 73.4% beyond 35m. Metres gained per pass triples, +0.62 → +1.81. In other words, the players ahead of the ball push up and offer. In that same band Gwangju go the other way, 5.6% down to 5.0%.
The weight of a single metric is easiest to judge by moving only that metric. Hold everything else fixed and shift only the defenders' progression rate to the league average.
⚠ This is arithmetic, not a projection. Opponent response and structural knock-on effects are not modelled. Read it as the leverage this one metric carries, not as “this is how much they would improve”.
By the same logic, Gwangju's current strengths make poor improvement targets. Leading the league in blocks at 17.19 per match is not a good sign — it means that many shots were allowed. Clearances say the same thing. At this club the number that has to rise is not how much they stop, but how much they send.
Seven narrowings later, one question remains.
Those three to five seconds after a Gwangju defender takes the ball in the middle third. If nobody was ahead of him, it is an off-ball problem. If somebody was and the ball still went sideways, it is a decision-and-instruction problem. The remedies are completely different — and event data cannot tell them apart, because a player who never touches the ball does not exist in this dataset.
What the analysis has done is fix where to look. Nobody needs to rewatch twenty-seven matches: the moment a defender receives in the middle third, three to five seconds forward, and starting with the losing-state passes — 1,419 of them. The job of analysis is not to hand over an answer. It is to narrow the search.
Other things this piece cannot claim.
The summer signings improved what happens inside the box, and the data confirms it. What did not improve is the road to the box. That road closes in the three seconds after a defender receives in the middle third — and it closes further when the team is behind.