I've been playing my fork of minesweeper, collecting data, and trying out making variants to try to understand the game more. my version is here: https://ernop.github.io/minesweeper-friendly/

The more I play the more I realize that high scores, fast times etc are actually composed of many factors; within them some I still care about but others are pretty irrelevant.

For example, see this image. It shows the time to complete an intermediate size board, by 3BV, a technical measure of how complex a board is to solve based on the layout of the fixed number of mines; you can prove that this bears some relation to the minimum and likely average number of clicks humans have to make to solve a level.

Notice that all my fastest times are on boards which have very low 3BV? Note 2: the colors relate to the decency of the playtime of that game. There are other charts which show that over time I am slightly improving, at least in aggregate.

So, getting a high score at least requires (the skills of logic and fast hands) which you know are important. But for someone with say, best in 99.9% of players, then you can say that his actual score achieved is mainly going to depend on the random 3BV rules which he is subject to. And so you'll play 100 games and perhaps only the 1 or 2 best will be your real chances to get a new record; the rest is just sitting there, basically wasting your time.

This is similar to this concept of "Jagged" progress in AI. I first came to this realization from playing against super-strong all-around go bots; they'd beat every human; but I realized that that ability doesn't mean at all or anything close to a claim that they are superhuman at any one part of the game. i.e. they easily could have been rather bad at the opening, but so superhuman at other parts of the game that they'd still never lose. Some of these are actually detectable with experiments, although I didn't see any being done. So, be wary of assuming if something is best overall, that any single one of its styles or choices is best AT ALL. You can't infinitely discard it with this idea but you can at least justify challenge.

So given that "getting a high score" really is composed of lots of things (thinking fast, solving fast, having memorized patterns, executive choice ability, physical mouse speed accuracy, as well as being lucky to be playing well when you get a high quality chance via a good 3BV), what are we even doing playing all day? that's not at all the way to get ahead.

And also it would be very against the spirit of things to write a search program for the absolute lowest 3BV levels and then only play those! (you could likely win in just 1s even though the world record might be 10x that)

So the implementation of the actual game is more like: at the start, you don't know anything about a board's 3BV, and the more of the board which opens, the more you know. i.e. you can feel or see that things seem to be going well, and that the worst complexity cases aren't happening, that you're in a "high-roll" and "above-pace" or on a gold split effectively.

But what about strategies like "insincerely playing a level and then intentionally dying fast once you suspect strongly enough it's not a world-class low 3BV"

And so as with so many things, true randomness is hard to use in a game.

And this seems to also be another instance of that old principle, that it's very hard and probably impossible to truly tell the difference between "reward-hacking" and "real work" or between "rentseeking" and "economically productive activity". Just as in speedrunning, there are so many degenerate strategies.

Perhaps some are just due to the removal of formerly pervasive natural additional restrictions on people which now no longer exist. i.e. in sports, medicine/recovery improvements allow and rationalize much greater and more intense training, which in the old world just weren't optimal; i.e. with a low population on earth and limited recovery powers, it might have been optimal not to train often nor to train at max intensity.

But even in the video game world, defining the rules of competition for a category in a speedrun is not trivial.

Mouse speed

I sometimes get stiff or can't move as fluidly as at other times. This is obvious from the tendencies in my max click-rate over time or by session;

My view now

This kind of insight, that "trying to get the record" is freeing because rather than operating at the top of the funnel, i.e. just playing over and over, if I realize that many of aspects of doing well are just uninteresting (waiting for the right seed) then i am free.

I often conclude that, also, I should just focus on the true limitation. This frees me to do things like

In factors, there is a very strong limitation on the player interacting with an object which requires him to be located near it. This was put into place as a nudge to early players to focus on automation. However at one point in lategame I observed I was spending say sometimes much more than 10% of my total game time simply running from place to place, or even just running those last 1/2 screen gaps (taking 2s) before being allowed to interact.

And even worse, the game played a very annoying and echoing sound when this happened, AND there was no viz indicator of being close enough to avoid this! So I just installed a mod which removed this limit and I feel totally fine about it.

Same with many cases; difficulty can have multiple components; but if one is not interesting, i don't think its wrong to remove it.

My rule for minesweeper - addiction resistance and focus ownership

mentally I call this "the game which you must constantly end early, losing good runs, to show you are the master, not it". That feeling of "just one more game" or "I can't quit now, this in-progress game is too good!" is actually your enemy and if you want to thrive, whenever you feel annoyed with the game or remember you should do something else, you should do so, immediately close the game, even if you were on WR pace.

And cases where you think "I'll quit after I win just one more time" are absolute classic opportunities to show yourself who's really boss. And even if you fail, at least you'll just know flat out that you are not at present the master of your time and focus.

The Future

This also leads me to wonder what really is interesting about this? I can see the following as important and fun optimization targets for this game

• the logic- both in super fast recognition of the common patterns,
• the logic- also being able to self-challenge to puzzle out the really hard logic puzzle
• at some point you realize that taking such time means you won't get a high score so you then think about doing early-guessing, or optimal-guessing
• they you realize that this is a slightly anthropic choice and that you can make it advantageously
• i.e. that guessing for max safe is not the same as guessing for max-win-this-board by getting more info at some slight cost being a possible real case.
• mouse control—accuracy. this also varies by coffee.
• mouse movement—in general, sometimes i
• mouse control—not misclicking. at certain times of day/coffee infusion states/other I am a massive misclicked!
• agents power and control—some people have bad desks, bad mice, etc and just never fix it. this is actually part of the liq test that is life! and the ability to identically fix your life, right now, get up now and do it
• So, I've thought of making a keyboard shortcut version which would remove the mouse requirements and directly test how quickly you could mentally prioritize among the main choices in the game: v0 where is safe v1 do i see some good patterns here in my immediate focus region (the region where my local cache of mine locations is still in RAM) v3 this region is sucking and being slow; even though I have moves I can find here, there are other regions where i can make progress much faster,

So the steps to improvement are (at least these) * solving * recognizing * executive choice to switch regions * finally too, once you notice you ARE on r pace, now you get to deal with stress and self-judgment and get experience there too! bonus!

Links

Note: my personal implementation of minesweeper is here

https://ernop.github.io/minesweeper-friendly/

It stores your full play history in localstorage (so save it if you move comps) and it's very useful for looking for trends. I've got about 4000 games stored so far; I can only dream what future agents can gather from this! Start collecting data now!

The version I have also tracks session info in various ways.

At the limit of video game skill, you probably mostly don't practice in-game I think this is rather related too. I need to flesh this out more, but basically I think that of the ways to improve now mainly I should separately be optimizing my mouse-input speed via actual physical therapy etc. 2nd part, make a bit of training software which would let me see my real weaknesses in exec choice, switching behavior, and pattern application, to improve that. I don't even know right now what is going on, why my production rate of good times is based on my quantified skill at the sub-skills, etc! So Wild

My records

Year ago I believe I hit sub 70 for expert; just yesterday I got my first "modern times" 71s for it. The below is just about 10 days since I made this; long ago I put in maybe 100x total plays or more over years and multiple periods starting as a teenager. It is amazing how quickly it comes back to you!

I never used to play intermediate but do a lot now: