Skip to content

Minesweeper Data ​

Minesweeper data are the metrics used to evaluate board complexity and player performance. The existing metrics can be roughly divided into single-game data and historical data.

  • Single-game data are the data of one game. Single-game data are further divided into basic data and derived data.

    time, bvs and stnb are the three most widely accepted single-game metrics for evaluating racing strength, and are called the three major parameters.

  • Historical data are distilled from all of a fixed player's single-game data over an entire minesweeper career. At present the most popular kinds of historical data are the LSF system, the pb system and the player line system.

    • The LSF system is a standard that rates a player's level based on the data of the country's top players.
    • pb is a player's historical best result on a specific bv. Combat power, typical time, typical stnb, the medal system and the head-to-head mechanism can all be derived from pbs.
    • The player line is the set of a player's strongest replays that do not beat each other. Coarse tiers, fine tiers and the mutual-knock mechanism can be derived from player lines.

Basic Data ​

Basic data are data specific to each replay that do not depend on any other data. Subdivided further, basic data can again be split into properties of the board itself (board data) and the player's input data.

Properties of the board itself include bv, op, is, the various zini and cellx. The player's input data include the various cl and ce, path and time. Strictly speaking, the placement of mines is the most basic property of the board itself, and each click and the mouse trajectory are the most basic player input data, but these are too tedious to summarize and are generally not discussed.

Board Data ​

bv, op, is ​

bv is short for 3BV, 3BV is short for BBBV, and BBBV is short for Bechtel's Board Benchmark Value. bv is the most classic parameter for measuring the complexity of a board.

op is short for Opening; in Chinese it is 空.

is is short for Island; in Chinese it is 岛. To avoid confusion with the English word "is", it is sometimes written isl.

Wudi's column explains in detail what bv, op and is mean and what role they play in board difficulty, so it is not repeated here.

zini ​

If bv measures the lower bound of the number of NF operations needed for a board, zini measures the lower bound of the number of FL operations. The concept and algorithms of zini were proposed in 2009 by Elmar Zimmermann and Christoph Nikolaus on the international forum, and are named after their surnames. Because finding the optimal FL solution for a board is an NP problem, i.e. it cannot be computed at a small cost, they instead looked for fast approximation algorithms. They proposed three greedy algorithms.

The core greedy idea is to find the operation with the greatest gain at each step. Given that all mine positions are known, every cell in a given board state has exactly one operation that can produce gain:

  • Flag around this cell and chord on it; the gain is bv opened by the chord - number of flags - 1. If the cell is not yet opened, then taking the operation needed to open the cell into account, the gain is further reduced by 1. bv already opened and flags already placed are not counted. Chording directly is equivalent to a flag count of 0.

Starting from the beginning of the game, compute the gain of every non-mine cell at every step. If any cell has non-negative gain, choose the cell with the greatest gain and perform the corresponding operation; if all cells have negative gain, finish the remaining bv with left clicks. The number of operations used to finish a game this way is its zini. The three zini algorithms differ only in details.

gzini is short for "Greedy ZiNi": when several cells share the same greatest gain, it always picks the topmost-then-leftmost one. rzini is short for "Random ZiNi": when several cells share the same greatest gain, it picks one at random. The random sweeping is repeated until n consecutive games fail to set a new record for the fewest operations. hzini is short for "Human ZiNi": it first opens all op with left clicks and then behaves like gzini.

Obviously rzini still contains luck and is unstable, so it is generally not used as a metric. gzini and hzini are not ordered, and neither is unbeatable. Players have already beaten both gzini and hzini on beginner and intermediate (i.e. finished with fewer operations), but nobody has achieved this on expert yet. Both Minesweeper Arbiter and Metasweeper report gzini and hzini.

In recent years other variants of zini algorithms have appeared, the most notable being Pttazini (PTTACGfans's ZiNi).

cellx ​

cellx is the number of cells showing the digit x. x ranges from 0 to 8.

Input Data ​

time ​

time is the time used to finish a game and is the most important metric in minesweeper.

Historically there have been two kinds of time. One starts counting from 1 second and belongs to Minesweeper Clone. The other starts counting from 0 seconds and belongs to Minesweeper Arbiter. The mainstream timing today starts from 0 seconds, but because the earliest Winmine started from 1 second, some concepts still follow the 1-second convention.

  • Saolei.wang displays data timed from 1 second.
  • (Very important, a must-read for insiders and bizarre trivia for outsiders) Traditionally, the concepts of seconds and sub are defined on the premise that timing starts from 1 second. For example, an intermediate time of 10 seconds means that, with timing starting from 1 second, the integer part is 10, which corresponds to rtime from 9.00 to 9.99. Expert sub40 means expert seconds sub40, i.e. expert seconds <= 39, i.e. rtime<=38.99.

Whichever timing convention is used, rtime (Real Time) is unambiguous and always means time counted from 0 seconds. The time mentioned on the rest of this page always means rtime.

path, cpath, epath ​

path is the distance the cursor travels while finishing a game. The unit of path is pixels or cells, where 1 cell = 16 pixels.

cpath, short for Click Path, only considers the distance between the positions where click events happen and ignores the loss from curved mouse trajectories.

epath, short for Effective Path, only considers the distance between the centers of the cells of effective click events and ignores the difference in where inside a cell the mouse clicked.

cl, ce ​

cl is short for "Clicks" and means the number of mouse clicks. ce is short for "Efficient Cl" and means the number of effective clicks. Broken down by click type there are also lcl, rcl, dcl, lce, rce and dce, where the prefix l is short for Left (left button), the prefix r is short for Right (right button) and the prefix d is short for Double (chord).

The difference between cl and ce is ineffective clicks. The reasons for ineffective clicks with the left, right and both buttons are:

  • left click landing on a number or a flag; cancelling a left click; clicking a just-opened op during a continuous sequence because of slow reaction.
  • right click landing on a number; misflagging; cancelling a misflag; wrongly cancelling a flag; re-flagging after wrongly cancelling a flag.
  • chord landing wrong; cancelling a chord; a probing chord that fails.

The main disagreement between standards is about ineffective right clicks. Minesweeper Arbiter considers ce to be all operations that change the board, so misflags and cancelling flags both count as effective clicks. Open Minesweeper and Metasweeper are stricter and require an operation to be correct to count as ce.

bvdone, opdone ​

bvdone means completed bv and opdone means completed op.

luck, pluck ​

Both parameters are based on a probability engine. For every operation that opens a cell, the engine computes the safety probability (0-1) from the current board; multiplying the safety probabilities of all operations gives luck, which represents the theoretical win rate of a game. pluck is defined as -log10(luck).


Derived Data ​

bvs, rqp, qg, ios, rv ​

bvs is short for "bv/s"; the formula is bvs=bv/time, meaning the number of bv solved per second, and larger is better. bvs is the most classic speed metric, second only to time in importance. Generally a smaller bv gives a better time, while a larger bv gives a better bvs, so other parameters are needed to balance them.

rqp is short for "Rapport Qualité Prix"; the formula is rqp=(time+1)/bvs. The reason for adding 1 to the time is that the timing convention was still 1-based when this parameter was invented. If that 1 second is ignored when time is large, rqp is approximately time^2/bv. rqp values the role of time heavily, but it deviates when time is small, and the deviation is already serious when time<60.

qg is short for "Quality Grade", a strength parameter defined by Guo Jinyang from a large amount of data; the formula is qg=time^1.7/bv. Compared with rqp, qg has the advantage of emphasising bvs more and removing the deviation at low times.

ios is short for "Index of Speed"; the formula is ios=log(bv)/log(time).

rv is short for "Rirun Value"; the formula is rv=sqrt(yybv*rqp+0.25)-0.5. Here yybv is the bv used for wishful thinking, for example 100 for expert and 30 for intermediate.

ces, cls ​

Similar to bvs, ces=ce/time and cls=cl/time. These two parameters represent the player's operation speed; cls focuses on the overall operation rate, while ces represents the conscious operation rate.

corr, thrp, ioe ​

corr is short for "Correctness"; the formula is corr=ce/cl. Because ineffective clicks are in most cases not caused by misaiming (a player who misaims a lot can hardly finish a game), corr can actually reflect, to some extent, how much thought goes into the operations. Mindless chording sprees and combos that completely ignore potential op both lower corr.

thrp is short for "ThroughPut"; the formula is thrp=bv/ce. For FL, thrp reflects overall flag-and-chord efficiency. For NF, thrp reflects overall opening-breaking efficiency.

ioe is short for "Index of Efficiency"; the formula is ioe=bv/cl. ioe reflects overall operation efficiency.

Obviously we also have the following relations:

  • corr=ces/cls
  • thrp=bvs/ces
  • ioe=bvs/cls
  • ioe=corr*thrp

For NF, the only way to improve operation efficiency is to improve awareness of op; for NF the meanings of ioe, thrp and corr are much the same, and higher means stronger opening-breaking awareness. For FL, corr mainly reflects operation precision and the success rate of probing chords, while thrp reflects the efficiency of the optimization strategy. Because reasonable probing chords can produce gain, FL players should not over-pursue corr.

cpo, coo ​

For NF, a low thrp in a single game may be caused not only by insufficient opening-breaking awareness but also by a bad board. In plain words, the more op a board has, the lower the thrp. To offset the influence of op fluctuations on thrp, two parameters were invented.

cpo is short for "avg Ce wasted Per Opening"; the formula is cpo=(ce-bv)/op. coo is short for "avg Ce cost On each Opening"; the formula is coo=cpo+1.

mov, sao, iome ​

mov is short for "Movement"; the formula is mov=path/time, reflecting the overall mouse speed.

iome is short for "Index of Movement Efficiency"; the formula is iome=bv/path, reflecting the solving strategy. A low iome means a tendency to solve bv nearby, while a high iome means a tendency to look for optimal solutions over a wide area.

sao is short for "风骚度" (flair); the formula is sao=mov*ces, reflecting the overall intensity of operations.

zinis, znt, zne ​

In the framework that uses bv as the measure of board complexity, bvs represents speed and thrp and ioe represent efficiency. bvs, thrp and ioe correspond to zinis, znt and zne in the framework that uses zini as the measure of board complexity:

  • zinis=zini/s
  • znt=zini/ce
  • zne=zini/cl

For different zini algorithms, just change the corresponding zini prefix, for example gzinis, rznt, hzne and so on.

completion, estime, stnb ​

completion means completeness; the formula is completion=bvdone/bv.

estime is short for "Estimated Time", meaning the projected time; the formula is estime=time/completion.

stnb is short for "尸体牛逼" (how awesome a corpse is) and measures how impressive a corpse is; the formula is stnb=sqrt(completion)*constant/qg. The constants for beginner, intermediate and expert are 36, 162, 435; the purpose of the constant is to make stnb roughly equal across the three levels at the same skill level. It is precisely because of this constant that stnb is more popular than qg and is one of the most important "three major parameters".

The constant can also be computed as 73.5*mode^2-94.5*mode+57. This applies to custom counters that do not support branch logic. Before 2023-09-13 the constants were 47.299, 153.73, 435.001, and the formula for computing the constant was 87.420*(mode^2)-155.829*mode+115.708.

numbers, obg ​

numbers is the sum of all numbers on a board; the formula is numbers=0*cell0+1*cell1+2*cell2+3*cell3+4*cell4+5*cell5+6*cell6+7*cell7+8*cell8.

obg measures the player's reaction speed to bv; the formula is obg=10000*bvs/numbers, where 10000 is purely to keep the value from being too small.


The LSF System ​

The top and LSF rankings use best-five results, anchors and relative strength to rate players. See top and LSF Rankings for the current rules and tier system.


References ​

  1. Guo Jinyang, "A Brief Discussion of Strength Parameters" (实力参数浅谈), Saolei.wang
  2. Lin Jinfan, "A Dictionary of Minesweeper Technical Metrics" (扫雷技术指标解析词典), Saolei.wang
  3. Guo Jinyang, "Advanced Minesweeper Technical Metrics" (进阶扫雷技术指标), Saolei.wang
  4. Zhou Ke, "A Brief Analysis of 3BV and ZiNi" (3BV与ZiNi浅析), Saolei.wang
  5. Zhao Shuang, "Minesweeper Terminology, 2nd Edition" (扫雷术语(第二版)), Saolei.wang
  6. Elmar Zimmermann, "new benchmark: 3bv for flaggers", Authoritative Minesweeper