Match Analysis · K League 2 2026 · Re-examination

Cross → shot was half of it

Last month I narrowed Suwon Samsung's leak down to a single phase. Then I pulled 279 shot coordinates off the league's chalk boards myself and put the same question again — and a second answer came back. The cause I had named, meanwhile, had already been fixing itself. This is the record of me correcting my own conclusion.

Hyuckjin Han · Tactical analysis Aug 2026 Source: K League Data Portal only (supplied by Bepro11) 22 matches Chalk board coordinates — 279 shots Big 5 European leagues, 96 clubs

If this is your first piece of mine — it reads on its own, no prior article required. But what I got wrong and how I fixed it is the spine of the piece, so I start by summarising the earlier conclusion.

00Why I am reopening this

Suwon Samsung sit top of K League 2 in 2026 with the best defence in the division. And yet, among the clubs chasing promotion, they score the fewest goals. The earlier analysis asked where that shortfall is created, and answered it by setting 20 matches against 96 clubs from the Big 5 European leagues.

That conclusion still stands. But three things in it were wrong or incomplete.

The order below: define the problem (01) → the basis of the earlier conclusion (02) → first correction (03) → asking again with coordinates (04) → second correction (05) → third correction (06) → the revised prescription (07) → the limits of this piece (08).

01The problem — 1st at creating, 6th at cashing in

First, does the problem exist at all? Place Suwon against the other 17 K League 2 clubs in standard-deviation units (z-scores) and the shape shows up at a glance.

Input · output · efficiency — against the league average (z-score)

K League 2 2026, 17 clubs · through R21 · 0 = league average, one step = one standard deviation

The team that produces more crosses, more final-third passes and more key passes than anyone in the division ranks 6th for goals. The problem is real. The next question is which phase loses it.

02The earlier conclusion — locating the leak

I split the road from cross to goal into three gates and set each conversion rate beside the 96 Big 5 clubs, K League 1 and K League 2.

The gap opens at the first gate only; the two behind it overlap. So I wrote: "not a team that cannot finish, but a team whose crosses do not become shots." Across the 96 European clubs I also confirmed that cross volume is unrelated to goals (r = +0.03), while the rate at which crosses turn into shots is strongly tied to them at r = −0.68.

All of that still holds. The trouble was the sentence that came next.

03First correction — I read "same as the European average" as "no problem here"

So I took the same 20 matches and, this time, split them from the inside rather than comparing outward. Here are the gate conversions in matches won and matches lost.

Look at the second row. The share of shots taken inside the box is essentially the same whether Suwon win or lose (0.638 vs 0.626). Its correlation with goals is −0.02 — nothing at all. Suwon rank 2nd in the league for shots in the box (+1.52σ), and that metric explains none of their results.

What separates them is whether the shot goes towards the goal. Shot → on target splits 0.399 in wins against 0.261 in defeats, correlating with goals at +0.46 (t = 2.22).

† I do not use the +0.81 on "on target → goal" as evidence. Goals are a subset of shots on target, so the measure is substantially tautological.

In raw volume it is starker still. In the matches they lost, Suwon produced more crosses, more shots and more shots inside the box. The only column where the wins lead is shots on target.

Which means there are two cut points, not one

One funnel, two questions

Above = compared with everyone else · below = comparing Suwon's wins with Suwon's defeats

04Asking again with coordinates — position, or state?

We have reached "shots inside the box that do not go towards goal." There are two possible reasons.

• If it is a matter of position — the ball is arriving in bad places, and the shots in low-conversion matches should actually be worse located
• If it is a matter of state — the locations are fine but the conditions at those locations are not, and shot positions should be the same

Until now there was no data to answer this. So I separated the shot markers on the portal's chalk boards by colour and extracted their coordinates. 22 matches, 279 Suwon shots. Checked against the official team totals in 20 of those matches, 18 come within two shots.

I split the matches by shot-on-target conversion and compared the shot locations.

The in-box share is 62% against 63% — the same. And the distances run the other way: shots in the low-conversion matches are closer to goal.

That is the opposite of what I expected. Conversion is lower in the matches where the shots are nearer (share of shots within 12 m against on-target conversion: −0.48, t = −2.32). And in-box share has no relationship with conversion at all (−0.15).

One caution: "closer shots are worse" must not be read as causation. A high share of close shots is better read as a signal that the opposition box was crowded — the more you hammer at a team that has dropped off, the closer and the more blocked your shots become. Read that way, the table does not say "stop shooting from close range"; it says "pushing more bodies into a packed box is not the answer." Which is the same story as 37.5 crosses in the defeats against 23.7 in the wins.

And then this number

Where did the 279 shots come from — distance to goal

K League 2 2026, 22 matches · Suwon Samsung · coordinates extracted from chalk boards

05Second correction — the route I blamed was already being fixed

In the earlier analysis I blamed the delivery route for degrading cross quality. The argument was that when a lot of passes go straight from the back line to the flank without passing through midfield, the opposition defence never has to turn its head, which makes it easy to close down the crosser. In the measurements, that kind of route accounted for 34% of the total.

This time I rebuilt the 21 pass matrices in round order.

Share of back-line passes that skipped midfield — by round

Of forward passes from the back (GK · DF), the share going direct to the front without passing through midfield · R12 is Suwon's bye, the league having an odd number of clubs

And yet the output did not improve to match.

Fixing the route took crosses from 24.1 to 35.6 and shots from 10.9 to 18.8. But goals per shot went the other way, 0.113 to 0.099. They create more and convert less.

In a multiplicative structure that means one thing. If you raise one term and the product does not rise, another term is low enough to swallow the gain. And the previous section established that the term in question is not position. What remains is the composition of the arrival.

06Third correction — set pieces were mixed into the 554

I went back to the R7 chalk board against Gimpo and re-counted it filter by filter.

If set pieces were excluded, open play alone would have to reach 29; it reaches 22. Set pieces are inside the cross total. (The gap between 27 and 29 looks like overlapping markers merging.)

And one more thing. The portal's "set-piece cross" is broader than corners plus free-kick crosses. There are 5 of them, against 3 corners and 0 free-kick crosses. The coordinates put the other two on the touchline — and Suwon took 24 throw-ins that day. Long throws into the box are being counted as set-piece crosses.

07So how does the prescription change

Two cut points mean two prescriptions. And they are different in kind.

That last row is the most practical thing in this piece. Three items drop out of the prescription.

1. Crossing less — across the 96 European clubs, cross volume is unrelated to goals (r = +0.03). This is a conversion problem, not a volume one
2. The crossers' technique — cross completion already ranks 5th in the league (+0.67σ). The same players produce crosses from corners that the division respects. It is not ability, it is conditions
3. Getting into the box more — the share of shots taken inside the box is identical in wins and defeats (0.638 vs 0.626). Box entry is not this team's problem

One thing is left. What state the players in the box are standing in at the moment the cross arrives. Seven shots from inside 6 m in twenty-two matches says it most directly — they get into the box but not deep into it, and so they shoot from whatever space is left over rather than from a good position.

In the matches they lost, Suwon delivered 53% more crosses and took 62% more shots inside the box. And scored 47% fewer goals. Doing more is not the answer.

08The limits of this piece

This is a corrected piece, so I will note in advance where it is likely to need correcting next.

• Sample. 20 matches, of which 4 are defeats. From the win/draw/loss comparison I read direction only, never magnitude. The same goes for the 6-versus-4 low/high conversion split
• Multiple comparisons. This analysis produced more than twenty correlations without correction. The t = 2.22 on shot → on target is significant only as a single test. It stands only because the win/loss group difference and the low/high conversion contrast stand beside it as supporting evidence
• Proxy metric. "Shot → on target" is only a proxy for the composition of the arrival. Whether shot quality is low because of box occupation, finishing ability or opposition density, this data cannot separate. Measuring it properly needs the number and arrangement of players in the box at the moment the cross arrives — and that is positional-data territory
• Coordinate extraction error. These values come from colour separation on images. Overlapping markers merge, leaving 2 of the 20 matches three shots short
• The set-piece coefficient is unfixed. The 17.2% in section 06 is one match. It needs several more open-cross boards
• Correlation is not causation. "Closer shots are worse" in particular should be read as a proxy signal for congestion, never inverted into a prescription
• Not opened yet — the player level. Everything here is team-level. I have reached "they shoot from inside the box and miss the target" without knowing who does it. That is the next piece

09Why rewrite rather than move on

The earlier conclusion was not wrong. It looked at one layer and wrote as though that layer was everything. Transcribing the observation "the same as the European average" into "no problem here"; presenting a season average of 34% as the current state; calling a number with set pieces in it an open-play figure — all three only became visible after digging further into the data.

And the corrected version is stronger. Two cut points make the prescription more precise than one did; ruling out a cause candidate narrowed the model; and the fact that the problem I identified was already improving turned out to be evidence that this team is correcting itself.

Analysis does not finish in one pass. Re-testing your own conclusion takes longer than reaching it, and is usually worth more. This piece is the record of that second pass.

The next test is at player level. Open up shots on target and shots in the box per player from the same data, and it becomes clear whose story "inside the box but not on target" actually is. When it does, I will correct this piece too.