Playtest Oct 25 · 25 October 2026, evening

Three on the box

Three testers on one team, a lobby full of bots, the Oregon server. Three testers on one team against a lobby of bots: School, then Mall, then Slum. The testers' team took School on the score limit, nothing crashed, and the server held a steady 60 ticks a second.

Build
2c3482b0
Lobby
A · US West (Oregon)
Lobby time
32 min
Maps
3 in rotation
3testers
3at once, peak
2matches
290kills
140by humans
0human-on-human
8,544shots fired
25 %accuracy
21revives
0 · 0crashes · kicks

Humans in the lobby: 21 min with 2 or more.

Classes

Who did the killing

Share of human kills by which class's gun did it. The pick screen doesn't log picks yet, so this is the honest proxy for what people played.

Share of human kills by class: Skinny 40.2 % · Mid 38.5 % · Thicc 21.3 %Skinny: 40.2 % of human kills · 49 kills · 2,491 shotsSkinny 40 %Mid: 38.5 % of human kills · 47 kills · 2,818 shotsMid 38 %Thicc: 21.3 % of human kills · 26 kills · 2,757 shotsThicc 21 %
  • Skinny49 kills · 2,491 shots · 31 % of shots
  • Mid47 kills · 2,818 shots · 35 % of shots
  • Thicc26 kills · 2,757 shots · 34 % of shots

Weapons

The gun table

Shotguns and the sniper live at the top of kills-per-shot by design; the three main rifles are judged on accuracy and kills per hundred shots.

Weapon statistics for Three on the box, human players only
WeaponClassShotsAccuracyHeadshotKillsKills / 100RangeShare of kills
AK-69Mid2,02927.2 %11.2 %472.316 m
Noscope 3000Skinny32636.2 %32.2 %4212.921 m
B.A.R.Thicc99829.8 %11.8 %161.611 m
ChauchatThicc1,18621.8 %19.4 %100.811 m
PEW-90Skinny93924.5 %12.2 %50.510 m
Double BarrelSkinny18517.3 %15.6 %21.110 m
Meme Launchereveryone0——2—— m
Dual BlickiesSkinny1,04123.7 %15.0 %00.010 m
MID-16 BurstMid57419.5 %16.1 %00.013 m
AYO-12Thicc32819.5 %18.8 %00.08 m
Desert BeagleThicc24518.8 %15.2 %00.011 m
BoomstickMid21520.5 %13.6 %00.07 m
Pocket Pistoleveryone2611.5 %33.3 %00.013 m

Human shots only. Accuracy = hits the server confirmed ÷ shots. Headshot = share of hits. Kills / 100 = kills per hundred shots. Range = typical hit distance. Other kills: Explosions 5 · Melee 0 · Radiation (Rad Bomb) 1.

Abilities& gadgets

Not counted yet. The server doesn't log ability or gadget use in this build; the next one does, and this section fills in.

The server

How the box held up

One cloud machine in US West (Oregon) running the lobby headless at 60 ticks a second. Ping is measured from every connected player every ten seconds.

  • 60 Hzserver tick, never below 58; slowest frame typically 29 ms
  • 49 msmedian ping · 9 in 10 under 62 ms · 99 in 100 under 65 ms
  • 10 KB/sper player each way during play
  • 0 %packet loss almost all session
  • 33 %of shots: client and server agreed on the hit (lag compensation rewind ≈ 0 ms)
  • 0 · 0crashes · kicks, across 6 joins

Notes

What we found. What shipped.

Found

  • Two guns did nearly all the testers' killing: the AK-69 and the Noscope 3000.
  • Bots got stuck on the Dead Mall garage ramp: 8 of the night's 11 stuck events.
  • Gun levels came too fast: one match could take a gun to level 4.
  • One tester spawned as the default class every life: the class screen's pick never reached the server.

Shipped after

  • Gun levels follow a longer curve, a match pays at most 2,000 weapon XP, and a kill on a human pays three times a bot kill.
  • Your pick sticks: if the timer runs out you keep the class and loadout on screen, every life.
  • Bots step off whatever they were stuck on.

Matches

  1. SchoolScore limit
  2. MallScore limit · 10 min

Counted from the Lobby A server log, build 2c3482b0. Weapon damage values come from the game's weapon table. No gameplay data about individual players is published anywhere. The numbers as JSON.