Two Seasons in One

By the last match of the first half of the season (R17, 12 July, 2-0 v Gimpo), Busan IPark had 11 wins, 3 draws and 2 defeats from 16 matches — 36 points. 2.25 points per match: stretched over 34 rounds, a pace in the mid-70s. The press at the time called Busan the league leaders.

From the very next match they were a different team. A 0-1 defeat at Seongnam in R18 began an eight-match winless run (2 draws, 6 defeats). A 2-0 win over Gimhae in R26 ended it, but they lost again at Gimpo in R27. Ten matches since the halfway mark have brought 5 points, 0.50 per match — relegation form. In the September table Busan are 6th.

This piece asks one question. Which one is the real Busan? To answer it, results (points) have to be separated from process (the quality of the shots). In both halves, the results sat a long way from the process — just in opposite directions.

Re-take every shot by both teams at its xG probability and you get the points a team could expect from each match — expected points. In the first half of the season Busan’s expected points were 23.1. They took 36.

The gap of +12.9 is by far the largest of K League 2’s 17 clubs. Second-placed Chungnam Asan are at +4.3. Replay those 16 matches 20,000 times with the same shots and Busan reach 36 points or more in only 0.3% of them. The first 11 matches alone (9 wins, 1 draw, 1 defeat, 28 points) were close to impossible from 14.7 expected points — 1 time in 20,000.

Where did it come from? Busan’s 204 first-half shots were worth 25.1 xG and produced 34 goals (+8.9). Opponents’ shots were worth 23.4 xG and produced 17 goals (−6.4) — on top of which Busan players scored 3 own goals. Both ends beat expectation; the scoring end slightly more.

And the scoring surplus did not belong to one man. Gabriel +3.4, Minhyeok Kim +1.8, Hwi Son +1.4, Gaon Baek +1.3, Jinhyuk Kim +1.1 — spread evenly across five or more players. The top scorer, Cristian, was right on expectation (−0.6). One player’s finishing is something you can keep counting on. A surplus spread across many players at once is usually variance — the signature of luck.

The second half is the mirror image. In the nine matches with event data Busan’s expected points were 11.2; they took 5. That is 6.2 short — the 3rd-largest shortfall in the league. Replay those nine matches with the same shots and 5 points or fewer comes up 3.5% of the time.

This time it is almost all at the scoring end. 115 shots, 11.4 xG, 6 goals (−5.4). Conceding: 13 goals from 13.2 xG of opponents’ shots — exactly on expectation. A team that scored nearly nine more than expected in the first half scored more than five fewer in the second.

So did the process stay the same? No — the process got worse too. Just not by as much as the results. Non-penalty expected goal difference per match fell from +0.30 to −0.19. In expected points per match that is 1.44 → 1.25, a drop of about 0.20. Actual points per match fell from 2.25 to 0.56 (matches with event data), a drop of 1.69.

Put simply: in the first half Busan took 0.81 points per match more than their process earned, and in the second half 0.69 fewer. Measured by shot quality, about 88% of the fall is luck turning and about 12% is a genuine drop in performance (a split that leans on a single xG model). Neither version is the “real Busan”. On process alone, Busan have been a 1.37 expected-points-per-match team all season — upper-mid-table.

Where did the drop in performance come from? Split into phases: open-play chains only (set pieces and penalties removed), per match.

The attack is a paradox. In the second half Busan got into the final third more often (32.2 → 36.0 times). The share of build-ups reaching the final third went up too. But the size of each entry (xG) went from 0.042 to 0.027 — down about 35%. Open-play xG fell from 1.35 to 0.97, and open-play goals collapsed from 1.75 to 0.33 per match: 3 open-play goals in nine matches.

The defence is the reverse shape. Opponents got in less often (39.8 → 34.6), but each entry grew from 0.027 to 0.033. Opponents’ open-play goals doubled, from 0.56 to 1.11 per match.

⚠ Part of this change is made by the score. Official possession rose from 49.4% to 54.3%. Teams that are behind have more of the ball, get in more often, and take less dangerous shots against a deeper block. That is why the next section splits by score state.

In the first half of the season Busan spent 40% of match time in front (667 minutes). In the second, 15% (136 minutes). Time spent behind rose from 12% to 29%.

It is not that they stopped scoring first. Suwon Samsung in R19, Seoul E-Land in R20, Daegu in R23, Suwon FC in R24 — Busan scored first in all four. They took one point from those four matches. Points dropped from winning positions: 11 in the second half alone (4 in the first).

Split the whole season by score state and Busan’s character shows (compared across 212 league matches).

• Leading — they stop going forward. Attacks starting in their own third reach midfield 55.1% of the time, 15th in the league. Opponents reach the final third 40.6 times per 90, the 3rd most allowed.
• Level — Busan at their best. xG per final-third entry 0.036, 1st in the league. At the back, opponents score 0.41 open-play goals per 90 — the fewest in the league.
• Trailing — they create, but cannot turn it round. Open-play xG 1.35 per 90, 2nd in the league. Yet in matches where they fell behind at any point they take 0.40 points per match, 14th of 17.

The second half is the stretch where time in front turned into time behind. A team that gave the ball away when ahead now holds it while behind — so much of the “more entries, less danger” attack in the phase table is a result of that shift. Matched for score-state mix, opponents’ entry → shot rate was 22.6% in the first half (25.1% expected from the state mix) and 24.4% in the second (25.5% expected). The first-half defence was a little tighter than expected; the second half returned to average.

Jooseong Woo, a veteran defender who joined Busan this season after spells at Gyeongnam and Daegu, was a fixture in the first half. He started 14 matches between R1 and R16. His event locations put him on the right (average width 76 in Busan’s attacking direction — left-sided defender Sungjin Jeon sits at 14). That matches reports of him moving to right-back early in the season.

Then, at Jeonnam in R16 (5 July), he was replaced at half-time with Busan 0-2 down. From R17 to R27 he was missing from the matchday squad for 11 consecutive matches.

What the data can say is the difference between when he played and when he did not, split by his on-pitch spells (using substitution times).

The difference sits at the back. With him on the pitch Busan conceded 0.60 per 90; without him, 1.13. Opponents actually got into the final third more often when he played (35.5 vs 31.3) — the way of defending was different. Let them in, then stop the shot (entry → shot 20.8% vs 26.2%). Exactly the shape of Busan’s first-half defence.

⚠ But there are three reasons not to read this table as cause and effect. First, most of the OFF time falls in the second half — it overlaps completely with the luck turning and the score-state shift seen above. Second, splitting the same way by midfielder Dongsu Lee, whose minutes also fell over the same period, gives almost the same picture (opponent entry → shot ON 19.9% · OFF 27.0%). The two effects cannot be separated. Third, the first match without him, R17 against Gimpo, was a 2-0 win with a clean sheet.

His role was shared out. Right-sided Hyunbeom Ahn went from 39 to 90 minutes per match, and average width moved right for Jinhyuk Kim (58 → 70) and Hoik Jang (56 → 63). In early August a “Plan B back three” was reported (SPOTV News, 6 Aug). The make-up of the right side of the defence itself changed.

Several things happened to Busan at once in the second half. Here, in order, is only what the data and reporting confirm.

Cristian’s (the top scorer’s) departure came after the eight-match winless run. He played throughout it. His 22 second-half shots were worth 2.7 xG and produced 2 goals — within expectation. So Cristian is not a cause of the fall; he is a condition for the remaining seven matches (R28–R34). The only sample without a player who took 20% of the team’s shots is R26 and R27.

1, 2 and 3 are two sides of the same story. 5 is the clearest sign of what genuinely got worse, and 4 is where it spilled into points.

Representative matches are not picked by hand. The series’ four rules are applied as they are — ① the match closest to the season average ② the most Busan-like match (shots per entry) ③ the match where the result most outran the process ④ the match where the process most outran the result.

③ comes from the first half, ④ from the first match of the second. The rules split the season in two on their own.

Busan’s 2026 season looks as if it split in two, but by shot quality it is closer to one. 1.37 expected points per match — far more than that in the first half, far less in the second.

Some things did genuinely change in the second half. The size of each attack fell by 35%, time in front turned into time behind, and four matches in which they scored first left one point. It coincides with the change in the make-up of the right side of the defence.

Three things to watch in the remaining seven matches (R28–R34):

• xG per entry chain — 0.042 in the first half, 0.027 in the second. If it recovers, the fall was mostly luck.
• Holding a lead — points after scoring first. Second half: one point from four matches.
• Attacking without Cristian — the player who took 20% of the team’s shots this season is gone. The sample so far is two matches.