WHAT THE DATA SAYS · 04

Jeonbuk defend with possession.
The same possession works against them going forward.

Jeonbuk Hyundai make fewer defensive actions than almost anyone in K League 1, and concede the second fewest goals. That is not because they defend badly — it is because there is less to defend. Divide by exposure and their defensive volume is third in the league; their pressing intensity is higher than their opponents’. Yet the same possession inverts going forward. Half of Jeonbuk’s expected goals come from slow build-up, and slow build-up returns only 37% of what it is worth. All 12.2 of the goals they have lost sit there. The first answer this piece found — “they win with less of the ball” — turned out to be a scoreboard artefact, and the wrong turn is left in.

K League 1 2026 · Jeonbuk HyundaiR1–R27 · 27 matches38,079 raw eventsK League Data Portal · as of 5 Sep 2026

WHAT THIS PIECE FOUND

  1. Possession by result (W 51.6% / D 56.8% / L 59.4%) is an illusion. Strip the scoreboard out and the sign flips (60.9 / 56.4 / 54.2).
  2. Defensive actions rank 11th in the league — but 3rd per 100 opponent passes. PPDA is 5.3 against their opponents’ 6.2.
  3. Fast-break phases return 1.90× their xG. Slow build returns 0.37×. Ulsan’s slow build, same league and same model, returns 0.78×.
  4. The share of shots that reach the target is identical — Jeonbuk slow build 33.5%, Ulsan slow build 34.2%. Only the share that goes in is halved (12.1% vs 23.7%).
  5. When opponents sit deep Jeonbuk create more (17.2 shots, 53.3 passes into the box) — and take 1.33 points a match. When opponents come out they take 10.2 shots and 2.22 points.

01Two tables that do not agree

What the standings leave out

Jeonbuk are second in K League 1 — 27 matches, 11 wins, 9 draws, 7 defeats, 42 points. They beat Pohang 2-0 at home yesterday and pushed Ulsan down a place.

That is the standings. Spread the league’s 45 season metrics across all twelve clubs and a second table appears — one the standings do not explain.

123456789101112league rank →Shots13.37Shots in box8.48Off target6.07Passes434.81Short passes236.15Sideways passes300.07Crosses6.22On target4.63Long passes32.52Key passes5.85Tackles6.19Corners4.59Goals1.26Conceded0.85Recoveries58.67Interceptions6.63Ground duels6.26Dribbles1.7Blocks12.59Fouls10.11
Jeonbuk’s league rank across 20 metrics; the number beside each dot is the per-match value. Shots, shots in the box, passes, short passes, sideways passes and crosses are all first — goals are fifth. Lower down the same chart runs the opposite anomaly: fewest blocks and dribbles, 11th for ground duels and interceptions, 10th for recoveries — and the second-fewest goals conceded. (League search values are completed actions, not attempts.)

Three contradictions at once. They shoot most and rank fifth for goals. Their defensive actions sit near the bottom and they concede second fewest. They lead the league for passes and crosses and complete the fewest dribbles.

Why has the team that knocks hardest and is hit least drawn nine times?

02The first answer was wrong

Clearing away the possession illusion

The first answer was simple. Split possession by result and you get 51.6% in wins, 56.8% in draws, 59.4% in defeats — “the more ball they have, the less they win”. Too clean to trust.

Possession causes results, but it is also a trace of them: a team that concedes first ends up with the ball while chasing. So take only the window in which nobody is ahead yet — everything before the first goal.

Full-match possession → pass share before the first goal50%52%54%56%58%60%62%64%W 11 matches51.660.9first goal 40′D 9 matches56.856.4first goal 77′L 7 matches59.454.2first goal 31′○ full match (scoreboard effects included) · ● before the first goal (game state removed)
Hollow circles are full-match possession, filled circles the pass share before the first goal. The sign flips. Jeonbuk in fact held the most ball in the matches they won (60.9%). The high possession in defeats (59.4%) was not a cause — it was the mark of conceding on 31 minutes and chasing for the next hour.

The first answer was wrong. It left a question behind, though: the mean first goal in drawn matches arrives on minute 77, twice as late as in wins (40). In those matches Jeonbuk did not have less of the ball — nothing happened at all.

03The real variable is the first goal

Breaking the deadlock

Who scored first27 matches split by who scored firstJeonbuk first · 14 matches10W2D2L2.29pts/matchOpponent first · 8 matches1W2D5L0.62pts/matchNo opener · 5 matches5D1.00pts/matchHalf-time scorestate at the restartAhead · 8 matches7W1DLevel · 14 matches3W8D3LBehind · 5 matches1W4LBar length = matches · dark ink win / grey draw / rose loss
Left, who scored first; right, the half-time score. Jeonbuk hold what they have (8 matches ahead at half-time: 7 wins, 1 draw, no defeats) and cannot turn a deficit around. The problem is that 14 matches — more than half the season — went into the second half level, and eight of them stayed that way.

What Jeonbuk lack is not the ability to hold a lead or to survive. It is the ability to break the deadlock first. Which narrows the question: why does the team that shoots most fail to break it?

04First, why is the defence good?

There is an answer — and it is the clue to the attack

Jeonbuk make the fewest blocks in the league, rank 11th for ground duels and interceptions and 10th for recoveries, and concede the second-fewest goals. Read on its own that says “they do not defend and do not get punished”. But defensive actions should scale with how long the opponent has the ball. Divide by that exposure and the picture changes.

Defensive actions (raw)Per 100 opponent passesrank by volume per matchrank once divided by exposure2 Daejeon1 Daejeon5 Gimcheon2 Gimcheon11 Jeonbuk3 Jeonbuk8 Seoul4 Seoul9 Incheon5 Incheon1 Gangwon6 Gangwon10 Pohang7 Pohang4 Ulsan8 Ulsan3 Anyang9 Anyang12 Gwangju10 Gwangju6 Jeju11 Jeju7 Bucheon12 BucheonApproximation: opponent passes/match = 2 × league mean − own passes. Jeonbuk measured (§4): 45.8 vs 35.8.
Left, rank by raw defensive actions per match; right, rank per 100 opponent passes. Jeonbuk climb from 11th to 3rd. Ulsan (blue) move the other way, 4th to 8th — they see less of the ball than Jeonbuk. The direction each line travels is a read-out of how much possession that club has.

The measured figures say the same thing without the approximation. Jeonbuk make 45.8 defensive actions per 100 opponent passes; their opponents make 35.8 per 100 Jeonbuk passes. For the same exposure Jeonbuk defend 28% more. PPDA — passes allowed per defensive action, lower meaning heavier pressing — is 5.3 for Jeonbuk against 6.2.

The second wrong answer this piece deleted. An earlier draft said Jeonbuk “do not press especially high”, on the grounds that the mean position of ball recoveries is almost identical for both sides (39.7 against 39.5). That measures where the ball is won, not how often. Jeonbuk win it in similar places but far more often for far less exposure. Judging pressing from a single mean coordinate was the error.

So the low defensive numbers are not a weakness signal. They are the shadow of possession: with the ball, there is nothing to defend. That is the good news. The problem is that the same possession runs backwards at the other end.

05Same possession, opposite result

The centre of this piece — split by phase, Jeonbuk are two teams

Possession phases were rebuilt from the event stream: consecutive events by one team form a phase, and the number of passes and its duration give it a character. Jeonbuk’s phases really are long — 6.32 passes and 32.5 seconds against their opponents’ 4.78 and 27.8, with 30% of phases running to eight passes or more (opponents 20%).

And yet slow build and fast breaks produce completely different returns.

Expected goals and actual goals, by phase typeSolid bar = goals scored · tick = that phase type’s xG · past the tick means more than expectedJeonbuk · 27 matchesFast breakxG 9.5Goals 181.90×MiddlexG 9.2Goals 60.65×Slow buildxG 19.2Goals 70.37×Ulsan HD · 26 matches — controlFast breakxG 12.5Goals 201.60×MiddlexG 17.5Goals 140.80×Slow buildxG 11.6Goals 90.78×Fast break ≤ 3 passes and ≤ 15s · slow build ≥ 8 passes · the rest is middle. Threshold sensitivity in the appendix.
Fast break (≤3 passes, 15s)MiddleSlow build (8 passes+)
The solid bar is goals scored; the tick marks that phase type’s expected goals. Past the tick means more than expected. Jeonbuk’s fast breaks return 1.90×, their slow build 0.37×. Ulsan — same league, same date, same portal xG model — return 0.78× from slow build. Jeonbuk make half their xG in slow build (Ulsan 28%), and all 12.2 of their lost goals are there.

Written out, the split is +8.52 goals on fast breaks, -3.19 in the middle, -12.17 in slow build. Jeonbuk’s fast break (1.90×) is better than Ulsan’s (1.60×). Their transition game is among the best in the league. Only the slow build collapses.

06Where it leaks

The shots reach the target. They just do not go in

“They cannot score from slow build” splits two ways — either the shots miss the target, or the ones on target get stopped. It is the second.

Shots → on target → goalsThe share that reaches the target is the same. Only the share that goes in differs.Jeonbuk Fast break74Shots32On target43.2%18Goals56.2%56.2%on target → goalJeonbuk Slow build173Shots58On target33.5%7Goals12.1%12.1%on target → goalUlsan Slow build111Shots38On target34.2%9Goals23.7%23.7%on target → goal
Jeonbuk fast breakJeonbuk slow buildUlsan slow build
The share of shots that reach the target is effectively identical — Jeonbuk slow build 33.5%, Ulsan slow build 34.2%. The difference is the last step: on-target shots become goals 12.1% of the time against 23.7%. Inside the same squad it is 56.2% on fast breaks against 12.1% in slow build — a 4.6× gap.

Which means the on-target shots from slow build are shots a goalkeeper can save: narrow angles, straight at the keeper, struck under pressure. An xG model reads the location and the situation type but not how many defenders are standing in front of the ball. Part of this gap comes from there.

So these numbers are not evidence that “Jeonbuk’s players cannot finish”. They mix the model under-rating how hard slow-build shots are with Jeonbuk genuinely producing poor chances, and this data cannot separate the two. What it can say is that if the bias were the whole story, Ulsan’s slow build should be just as low — and it is 0.78×. Model bias alone does not explain it.

07And Jeonbuk do not choose that

The opponent’s block height decides the phase mix

The 27 matches were split in three by the average position of the opponent’s defensive actions — a measure of how deep they sat.

The deeper the opponent sits, the harder Jeonbuk knock and the less they win27 matches split in three by the average position of opponent defensive actions (9 each)Output per matchLow blockopponent defensive actions, mean X 33.617.2Shots53.3Passes into the boxMid blockopponent defensive actions, mean X 39.611.1Shots38.1Passes into the boxHigh blockopponent defensive actions, mean X 45.510.2Shots29.9Passes into the boxPoints per match1.333W 3D 3L · Conceded 1.221.112W 4D 3L · Conceded 0.782.226W 2D 1L · Conceded 0.56The left-hand column knocked hardest and took the fewest points.
ShotsPasses into the box
Top panel, output per match; bottom panel, points per match. (Two different scales, so two separate axes.) In the nine matches where opponents sat deepest Jeonbuk knocked hardest — 17.2 shots and 53.3 passes into the box — and took 1.33 points. In the nine where opponents came out they knocked least — 10.2 shots, 29.9 passes into the box — and took 2.22 points. They also conceded least there (0.56 against 1.22).

Volume and results move in opposite directions — and the volume is set by the opponent, not by Jeonbuk. Against a deep block their phases stretch to 7.56 passes and 38.4 seconds (against a high block, 5.52 and 28.9), and the share of eight-pass phases climbs to 39%.

One thing worth checking — “don’t opponents sit deeper against Jeonbuk because they are title contenders?”
The Ulsan dataset gives a control. Mean block height against Jeonbuk is 39.56, against Ulsan 40.49 — a gap of 0.93, Cohen’s d of 0.17, which is negligible. Comparing the ten clubs both sides played, only six of ten sat deeper against Jeonbuk, and the correlation between a club’s strength and how deep it sat is r = 0.175, essentially none. Opponents treat the two teams much the same. What differs is the output.

08Where it breaks

Block height × phase type

Goals ÷ xG — opponent block × phase type1.00× = scored exactly as expected. Lower means fewer.Fast breakMiddleSlow buildLow block2.49×phases 24 · xG 2.41 · Goals 60.64×phases 36 · xG 4.67 · Goals 30.23×phases 75 · xG 8.65 · Goals 2Mid block2.31×phases 20 · xG 2.60 · Goals 60.39×phases 28 · xG 2.60 · Goals 10.21×phases 43 · xG 4.80 · Goals 1High block1.34×phases 26 · xG 4.47 · Goals 61.04×phases 22 · xG 1.93 · Goals 20.70×phases 30 · xG 5.71 · Goals 4The deeper the block (upwards), the worse slow build gets and the better fast breaks get — but those chances grow scarcer.
Lay the two axes over each other and the gradient completes. The deeper the block, the further slow build falls — 0.70× to 0.23× — while fast breaks improve from 1.34× to 2.49×. But those chances shrink from 26 phases to 24, and slow-build phases swell from 30 to 75. More and more of the load goes down the worse route.

09Same player, different phase

The loss is attached to the phase, not the player

Season totals make it look as though the squad contains a few poor finishers. Split by phase, it does not.

Same player, different phase — goals − xGRight of zero means more goals than expected0-2-1+1+2Lee Seung-wooShots 50Fast break 9Shots 4Goals +2.47Slow build 27Shots 3Goals -1.47MotaShots 39Fast break 9Shots 1Goals +0.05Slow build 19Shots 0Goals -2.43Lee Dong-junShots 37Fast break 7Shots 1Goals -0.26Slow build 17Shots 0Goals -1.88Kim Seung-seobShots 26Fast break 5Shots 0Goals -0.36Slow build 13Shots 1Goals -0.71ThiagoShots 26Fast break 8Shots 3Goals +1.66Slow build 14Shots 1Goals -1.49OberdanShots 17Fast break 8Shots 1Goals +0.24Slow build 7Shots 0Goals -0.54Team total — Fast break +8.52Goals · Middle -3.19Goals · Slow build -12.17Goals
Fast breakSlow build
The axis is goals − xG; right of zero means more goals than expected. Mota and Lee Dong-jun have 36 shots and no goals from slow build — and score to expectation on the break. Lee Seung-woo (+2.47) and Thiago (+1.66) carry the team in transition. The same player becomes a different player depending on the phase.

10One tool only

84% down the flanks, fewest dribbles in the league, 648 crosses

Where they enter — 784 entries into the final thirdJeonbuk attacking direction. Darker band = more entries through that channel.45%Left channel · 35316%Central channel · 12539%Right channel · 306Final thirdattacking directionOpponents enter centrally 19% of the time (Jeonbuk 16%)What crosses produce24.0 attempts per matchAttempts648Led to a shot97of attempts 15.0%Goals7of attempts 1.1%168 completed (26%) — most in the league
Of 784 entries into the final third, only 16% come through the centre and 84% down the flanks (opponents 19% central). On the right, what crosses produce: 648 attempts, 97 shots (15.0%), 7 goals (1.1%). Crossing is not the problem in itself — 15.0% reaching a shot is close to double the opponents’ rate. The problem is that it is almost the only key that fits when the door is shut — 46 completed dribbles is the league’s lowest.

Split by phase, the dependence has an address. On fast breaks 12% of shots come straight from a cross; in slow build it is 25% (Ulsan’s slow build: 15%). The more the box fills up, the more Jeonbuk lean on the cross.

The goal breakdown says the same. Open play 26 · Penalties 6 · Own goals 1 · Corners 1. One goal from 124 corners, and six of 34 goals (18%) from the penalty spot.

11Two matches

The cover and the counter-example

R14 at Bucheon — 26 knocks, one that nearly opened the doorJeonbuk attacking half · circle size = xG · green = on target · grey = off target or blocked · final score 0-0attacking direction85′ Lee Seung-woothe only big chance of the matchxG 0.350xG0.050.150.3578% possession · 811 passes · 64 crosses · 7 corners · 26 shots · 2.46 xG · one shot above 0.30 xG · 17 below 0.10
R14 away at Bucheon. 78% possession, 811 passes, 64 crosses, 26 shots. And 0-0. Exactly one shot was worth 0.30 xG or more, and the best moment of the match was Lee Seung-woo’s 0.350 on 85 minutes. Twenty-six knocks; the door rattled four times and nearly opened once.
R14 · away at Bucheon

Creating without opening

0-0

They took 78% of the ball and had 26 shots. One big chance. In this dataset it is the most Jeonbuk-like failure of the season.

Possession78%
Passes811
Shots26
xG2.46
Opp xG0.55
Big chances1
R27 · home v Pohang

They won on the day they created least

2-0

Yesterday, at home. Pohang out-created them on xG (1.12 to 0.97). Two goals arrived within two minutes — 14 and 16 minutes into the second half — and the match was over.

Possession54%
Passes535
Shots12
xG0.97
Opp xG1.12
Big chances1

What separates the two is not how well Jeonbuk played. It is which phase was available to them.

12Conclusion

And what is still unknown

Jeonbuk defend with possession. That is verified — third in the league for defensive volume once divided by exposure, and a lower PPDA than their opponents. The low defensive numbers are not a weakness; they are the shadow of the ball.

But the same possession runs backwards going forward. The longer they hold it, the more their shots come from slow build, and slow-build shots do not become goals. When the opponent sits deep Jeonbuk create more and take less. Nine draws have cost 18 points, and the gap to Seoul at the top is 17.

Jeonbuk do not need more entries. They need another key for when the door is shut.

Open questions

Appendix — data, verification, limits

Parsed directly from the internal responses of the K League Data Portal (portal.kleague.com). No screenshots, no manual entry. All 27 matches: 38,079 raw events with coordinates and xG, 1,079 × 77 per-match player rows, and 45 season metrics for all twelve clubs. The league comparisons were re-collected for every club at the same cut-off (5 Sep 2026).

Cross-checks

CheckResult
Standings ↔ match reports, re-summedexact (27 matches, 11W 9D 7L, 42 pts, 34-23)
Event shots ↔ league searchexact (361 = 361)
Event shots on target ↔ league searchexact (125 = 125)
League search ↔ match reports (corners)exact (124 = 124)
Event tackles ↔ league searchexact (167 = 167)
Event fouls ↔ league searchexact (273 = 273)
Coordinate normalisationmean shot end-X 100.0 for both sides; each half passes separately
45 league-search metrics, duplicate checkall 45 distinct by header+sum signature (0 duplicates)

Five limits

Sensitivity

Slow build, goals ÷ xG — does it hold when the phase definition changes?Five thresholds plus a time-based split. Jeonbuk stays at or below half of Ulsan in every one.0.2×0.4×0.6×0.8×1.0×≤3 passes·15s / ≥8 passes0.370.78≤2 passes·10s / ≥7 passes0.490.87≤4 passes·20s / ≥9 passes0.350.82≤3 passes·12s / ≥10 passes0.320.80≤5 passes·20s / ≥8 passes0.370.78time-based, 45s+0.510.75
Jeonbuk slow buildUlsan slow build
Slow-build goals ÷ xG recomputed across five thresholds and a time-based split. Jeonbuk stay at or below half of Ulsan in every one.

Portal data issues found

Reproducing

Collection spec in 울산HD/포털_수집_사양_K리그1.md, dataset notes in 전북현대/README_데이터수집.md, the full hypothesis test in 전북현대/가설검증_보고서.md. Every number on this page comes from report_src/data.py, which gen_data.py generates from the CSVs in data/.