在单位正方形里放 n 个采样点。从原点量起、边平行于坐标轴的每一块矩形，占了多少面积，就该分到多少比例的点；你的分数，是所有这类矩形里最大的那个偏差。把它压到最低。
把正方形当成一帧要渲染的画面，这 n 个点就是你全部的采样预算。哪块矩形分到的点比面积应得的多，是预算浪费在同一处；少了，是那块的细节被漏掉。
格点和随机撒点都会在某些矩形上系统性偏置；低偏差构造（Hammersley、van der Corput）压得低得多。但对每个具体的 n，没人知道还能压到哪里。
n ≤ 21 的最优解已由 Clément、Doerr、Klamroth 与 Paquete 在 2025 年证明（Proc. Amer. Math. Soc. Ser. B 12: 78–90），所以本站从 n = 22 起；再往上，每一个 n 都是开放的。
最大误差 D*
容器是边长 1 的正方形：左下角是原点 (0, 0)，右上角是 (1, 1)。被比较的矩形永远从原点量起，右上角可以落在正方形里的任何位置，所以一共有无穷多块矩形要同时满足，而不是某几块。坐标写成小数，例如 "0.25"，最多九位小数。
提交 points。每个坐标写成十进制字符串，例如 "0.25"，最多九位小数。榜上的数字是 D*，也就是所有矩形里最大的那个误差；分数越小越好。
{
"points": [
[
"0.008510638",
"0.008510638"
],
[
"0.025531915",
"0.408510638"
],
[
"0.042553191",
"0.208510638"
],
[
"0.059574468",
"0.608510638"
],
[
"0.076595745",
"0.108510638"
],
[
"0.093617021",
"0.508510638"
],
[
"0.110638298",
"0.308510638"
],
[
"0.127659574",
"0.708510638"
],
[
"0.144680851",
"0.058510638"
],
[
"0.161702128",
"0.458510638"
],
[
"0.178723404",
"0.258510638"
],
[
"0.195744681",
"0.658510638"
],
[
"0.212765957",
"0.158510638"
],
[
"0.229787234",
"0.558510638"
],
[
"0.246808511",
"0.358510638"
],
[
"0.263829787",
"0.758510638"
],
[
"0.280851064",
"0.033510638"
],
[
"0.29787234",
"0.433510638"
],
[
"0.314893617",
"0.233510638"
],
[
"0.331914894",
"0.633510638"
],
[
"0.34893617",
"0.133510638"
],
[
"0.365957447",
"0.533510638"
],
[
"0.382978723",
"0.333510638"
],
[
"0.4",
"0.733510638"
],
[
"0.417021277",
"0.083510638"
],
[
"0.434042553",
"0.483510638"
],
[
"0.45106383",
"0.283510638"
],
[
"0.468085106",
"0.683510638"
],
[
"0.485106383",
"0.183510638"
],
[
"0.50212766",
"0.583510638"
],
[
"0.519148936",
"0.383510638"
],
[
"0.536170213",
"0.783510638"
],
[
"0.553191489",
"0.021010638"
],
[
"0.570212766",
"0.421010638"
],
[
"0.587234043",
"0.221010638"
],
[
"0.604255319",
"0.621010638"
],
[
"0.621276596",
"0.121010638"
],
[
"0.638297872",
"0.521010638"
],
[
"0.655319149",
"0.321010638"
],
[
"0.672340426",
"0.721010638"
],
[
"0.689361702",
"0.071010638"
],
[
"0.706382979",
"0.471010638"
],
[
"0.723404255",
"0.271010638"
],
[
"0.740425532",
"0.671010638"
],
[
"0.757446809",
"0.171010638"
],
[
"0.774468085",
"0.571010638"
],
[
"0.791489362",
"0.371010638"
]
]
}{
"n": 47
}{
"points": [
[
"0.008510638",
"0.008510638"
],
[
"0.025531915",
"0.408510638"
],
[
"0.042553191",
"0.208510638"
],
[
"0.059574468",
"0.608510638"
],
[
"0.076595745",
"0.108510638"
],
[
"0.093617021",
"0.508510638"
],
[
"0.110638298",
"0.308510638"
],
[
"0.127659574",
"0.708510638"
],
[
"0.144680851",
"0.058510638"
],
[
"0.161702128",
"0.458510638"
],
[
"0.178723404",
"0.258510638"
],
[
"0.195744681",
"0.658510638"
],
[
"0.212765957",
"0.158510638"
],
[
"0.229787234",
"0.558510638"
],
[
"0.246808511",
"0.358510638"
],
[
"0.263829787",
"0.758510638"
],
[
"0.280851064",
"0.033510638"
],
[
"0.29787234",
"0.433510638"
],
[
"0.314893617",
"0.233510638"
],
[
"0.331914894",
"0.633510638"
],
[
"0.34893617",
"0.133510638"
],
[
"0.365957447",
"0.533510638"
],
[
"0.382978723",
"0.333510638"
],
[
"0.4",
"0.733510638"
],
[
"0.417021277",
"0.083510638"
],
[
"0.434042553",
"0.483510638"
],
[
"0.45106383",
"0.283510638"
],
[
"0.468085106",
"0.683510638"
],
[
"0.485106383",
"0.183510638"
],
[
"0.50212766",
"0.583510638"
],
[
"0.519148936",
"0.383510638"
],
[
"0.536170213",
"0.783510638"
],
[
"0.553191489",
"0.021010638"
],
[
"0.570212766",
"0.421010638"
],
[
"0.587234043",
"0.221010638"
],
[
"0.604255319",
"0.621010638"
],
[
"0.621276596",
"0.121010638"
],
[
"0.638297872",
"0.521010638"
],
[
"0.655319149",
"0.321010638"
],
[
"0.672340426",
"0.721010638"
],
[
"0.689361702",
"0.071010638"
],
[
"0.706382979",
"0.471010638"
],
[
"0.723404255",
"0.271010638"
],
[
"0.740425532",
"0.671010638"
],
[
"0.757446809",
"0.171010638"
],
[
"0.774468085",
"0.571010638"
],
[
"0.791489362",
"0.371010638"
]
]
}提交 points。每个坐标写成十进制字符串，例如 "0.25"，最多九位小数。榜上的数字是 D*，也就是所有矩形里最大的那个误差；分数越小越好。 · 验证器 v1.0.0