Thirteen games, twenty duos,
one match point.

Overview

The format, read off the rosters

Nobody published a bracket we could cite, so the shape of the event is reconstructed from who appears in which lobby. Two 20-duo groups played 10 games each. The top seven of each went straight to the Grand Final; ranks eight to seventeen dropped into a 10-game Survival lobby; the bottom three of each group went home. Survival’s top six filled the last places in the final — 7 + 7 + 6 = 20. The cut was brutal at the bottom of it: ZETA Koyota + ZETA yuma missed the Grand Final by one point, 368 against the 369 that took sixth.

The race to 350

The Grand Final ran on match point: pass 350 points, then win a game and the tournament is over. That turns the leaderboard into two separate questions — who is above the line, and who is capable of winning the next game. For four straight games those were different sets of people.

Game by game

Games 9 to 12 were won by duos with 158, 314, 129 and 210 points — every one of them below the line. By the 13th game there were 9 duos on match point, and one of them finally won it.

Standings

Every point on this page is recomputed from the games with the scoring model the parse recovered — 3 points an elimination plus a placement curve of 60/50/45/40/35/30 tapering to 3 at fifteenth and nothing below. It reproduces 79 of the 80 team-days in the capture exactly.

Grand Final · 22 August

Damage is the duo total across the lobby, as recorded per game in the capture. The capture also carries an alive field per game whose unit we have not been able to characterise against the replays, so it is not displayed anywhere on this page.

Farming

Reload gives you respawns, and respawns make an accusation possible that no normal Battle Royale format allows: that a duo found somebody soft and kept killing the same person. This panel tests that. The short version is that it did not happen — and the test is more interesting than the answer.

What the four lobbies looked like

A repeat kill is an elimination of a victim that same duo had already eliminated earlier in the same game. A respawn kill is one taken within 45s of that victim’s previous elimination in the same game. “Pairs” is how many distinct team–victim combinations were observed and therefore tested.

The respawn clock, measured

Take every repeat kill — the 1,241 of them defined above — and measure the time back to that victim’s previous elimination. Pooled across all four lobbies the shape is very particular: nothing at all under fifteen seconds, a hard spike at 30-45s, then a long tail. The spike is the floor of the format — respawn, re-equip, walk back into it. The tail is ordinary Battle Royale: you died early, then you died again at the end.

Scope, precisely: this is the repeat-kill sample, not every case of a player dying twice in a game. In each lobby gapHistogram sums exactly to repeatKills (267 · 281 · 378 · 315), which is how we know what it counts. The parse gives us no elimination-level event list for this event, so a “every double death” version of this chart is not something we can compute or check — and we are not going to describe this one as if it were that.

Lobby medians: Group A 220s · Group B 201s · Survival 222.5s · Grand Final 232s. The Grand Final’s quartiles are p25 43s, p50 232s, p75 706s — the median gap is almost four minutes, which is what makes the 30-45s spike worth looking at separately.

Was anyone actually hunted?

For each duo and each victim, work out how many kills you would expect if that duo picked their targets at random from everybody outside their own roster — Expected(T->V) = elims(T) x deaths(V) / deaths of all players outside T's roster. Compare it to what happened, score the difference with a Poisson p-value, then correct the whole set for the fact that 721 pairs were tested at once (Benjamini–Hochberg). Uncorrected, plenty of pairs look damning. Corrected, none of them do.

Who died most in the Grand Final

Being the most-killed player is not evidence of being targeted — the “spread” column is why. The lobby’s biggest victim was killed by 14 different duos, and no single duo took more than 36% of anyone’s deaths.

ELITE

Two ELITE duos played the Esports World Cup. One went out in the group stage. The other qualified directly for the Grand Final, won a game in it, and finished 14th of 20. Both runs are below, with the result reported as it actually was.

josh + Eomzo — the run

Sixth in Group B on 354 points, which was a direct ticket to the Grand Final: no Survival lobby, a day off while 20 other duos fought for six seats. In the final they scored 344 and finished 14th — the middle of the pack, and ahead of the duo that had won Group B outright.

The scoring profile

ELITE took 25.3% of their Grand Final points from eliminations (87 of 344), against a field average of 32.7% — 6th lowest of 20. That is a placement-led profile: they were paid more for being alive than for fighting. Same shape in Group B, where elimination points were 25.4% of the total. It is a real read, and it is a read — the measurement is the share, the interpretation is ours.

pxmp + EpikWhale — Group A

The other half of the org’s trip. XP42 EpikWhale + ELITE pxmp finished 19th of 20 in Group A on 144 points with 19 eliminations — 2 places below the Survival cut at seventeenth, so their event ended on 19 August without a second day. 2 of their 10 games were top-ten finishes and their best was 6th; the other 8 are the reason the total is what it is.

Reported as it happened. Half the org’s duos at the event went out in the first round, and the page says so in the same place it says ELITE won a Grand Final game.

Method

Every number on this page, and where it came from

Those files live at the repo root, outside the web root, so their contents are inlined into this page at authoring time by scripts/build-film-ewc.mjs. Nothing on this page is fetched at runtime.

Two elimination counts, and why they differ

The replay parse counts 2,897 eliminations. The tournament board credits 2,816 across the same 43 games — a ratio of 1.029. Both are on this page and both are labelled. The parse counts the knock, the board occasionally counts the finish; the per-duo ratio is recorded as parserAccuracy in the farming file and never strays far from one. Standings tables use the credited count, because that is what the points were paid on. The farming analysis uses the parsed count, because that is what it is analysing.

The scoring model, audited

The scoring was not published in a form we could cite, so it was fitted: least squares over all forty group-stage team-days, which recovered 3 points an elimination and the placement curve above with zero residual. Applied to every team-day in the capture it reproduces 79 of 80 posted totals to the point. The one exception is the champions’ Grand Final row, which is posted 1,000 points high — a championship marker, not points scored. A model that fits 79/80 and misses only the row with an obvious extra constant is a model we trust.

One number we recompute rather than read

The farming file records an elimPointShare per duo, but it divides by the posted total. For the champions that total carries the 1,000-point marker, which drags their share down to 10.2% and pulls the field average with it. Everywhere this page quotes an elimination share it divides by the points the duo’s games actually earned instead: the champions come out at 30.7% and the field average at 32.7% rather than 31.7%. Every other figure on the page is used as the file records it.

The match-point model was wrong, on the short side

Before the final we simulated it: 200,000 runs, calibrated on FNCS Major 1 Summit — Finals Day 3 (8 games, 50 teams, 7 distinct winners against a modelled 6.74). It returned a median of 9 games, p90 11, and a longest run in 200,000 tries of 14. The real final took 13. That is inside the simulated range but well past the p90, and the sensitivity table says why: the model’s pace assumption was too fast. At a pace multiplier of 0.7 the median moves to 12 games and the p90 to 14. We are recording the miss rather than quietly re-fitting it.

The calibration event has its own room: FNCS Major 1 Summit. The sampleFinal array in the model file is one draw from the simulation, not the real final — the real sequence is the ledger in the Overview panel.

What is not here

Definitions used above

Analyst framing — the format reconstruction, the reading of the match-point race, and the claim that a placement-led profile is what ELITE ran — is interpretation laid over measured numbers, and is labelled as such wherever it appears. The measurements themselves are in the files listed above.