Grid Sage Forums

Grid Sage Forums

  • March 28, 2024, 12:51:04 PM
  • Welcome, Guest
Please login or register.

Login with username, password and session length
Advanced search  

News:

LINKS: Website | Steam | Wiki

Pages: 1 [2]

Author Topic: High Scores and Stats  (Read 86680 times)

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #25 on: February 22, 2019, 06:18:03 PM »

Beta 7 Stat Summary
(As usual, leaderboards were reset with the new release. The full archived Beta 7 leaderboards can be found here, with score sheets for all runs accessible at the bottom.)

Time to look at non-anonymous player data! Like Beta 6 this was an especially long cycle, spanning months of Beta 7, 7.1, 7.2, and 7.2LE as I also took time out to do some other dev things, so we now have 6,145 score sheets to go through... Well, technically we're only looking at 5,208 (87.3%) of those since the rest scored below the 500-point threshold, a similar ratio to what we've seen before.

606 unique named players are included this time, a 14.6% increase compared to Beta 6. That's 11.6 qualified runs per player, a 27.4% increase.

The Beta 7 win rate was actually significantly lower at 6.6% (down from 8.2%), primarily because speedrunning was a big thing in the last release, but not so much this time. For Beta 6 I graphed the wins by turn count and found that about half of all wins were speed runs, so technically excluding those our win rate has actually increased somewhat. There were 344 wins in all by

Valguris (126), MTF (17), Zyalin (15), lsend (14), GJ (12), PlasticHeart (11), kiedra (10), Raine (10), Pimski (9), Tone (8), wilk (8), Decinym (6), Terminus (6), whitenitro0 (5), badjay (4), Main.Cuck (4), Malth (4), cptwinky (3), GridBugBear (3), JackNine (3), Kyzrati (3), Mhorre (3), octapi (3), Omewes (3), Puzzlebark (3), tiniclx (3), UlyssesB (3), Ape (2), IceBox (2), JakeThyCamel (2), MojoPoop (2), Owlsley (2), PhenomPhear (2), Rumbl3 (2), Aeon2927, Alexbot, Amphouse, bugsniper, ConfusingDalek, Cowmind, FIRE_C4T, GOR, Heavyrisk, Howard, JC, Joel, joke_LA, McSi, moonshine, NoNeedToExplain, physalin, Quelan, RNGesus, Sherlockkat, slh, Some Random Guy, SpiritFryer, SSMR, TBExtent, tournTrione, zidji, Zinc, zxc.

Once again Valguris racks up an insane number of wins xD

And of special note, Pimski earned Cogmind's second highest ever score (241,581!) using so-called "Garrison looping," repeatedly traveling through garrisons and destroying their contents, returning to the main floor. There was only one place in the world where this was possible, and it also involved the most challenging garrisons, but through a combination of speed and guided weaponry, Pimski managed to take out 44 of them xD. As of Beta 8 some of the mechanics involved there have changed so that it's no longer possible, but we're already seeing more regular runs creep up towards 200k regardless... That said, it may never be possible to break the 123 Regions Visited record set by that run.

Last time I graphed wins by turn count, and now that everyone seems to have speedrunning out of their system I thought it'd be interesting to take a look at win distribution by real time:



Most of Valguris' runs tend to fall in the 0.5~1.0-hour range, hence the extra high surge there, otherwise it's a fairly even distribution overall. I split out his runs in the graph to give a better representation :P

PlasticHeart has also provided us with what he calls "mutation distribution," graphing what kinds of long-term benefits different winners are using:



As you can see the graph excludes speed runs since they tend to distort data. There's a fairly even distribution here as well, although it's interesting to see that nearly 40% of wins do so with no mutations at all. Of course, getting any of these requires deviating from the main path, sometimes fairly far away, so they aren't the kind of thing you'll just "happen across." The new FarCom introduced for Beta 8 will no doubt be popular. (The current "mutation" data for Beta 8 scores is also available here, by the way, although there's not yet a way to tell whether anyone's immediately lost FarCom and therefore didn't really take advantage of it.)

The big new thing for Beta 7 was RIF, or robot hacking, so that's definitely something to explore here.

Obtaining this ability requires entering a Garrison, which isn't too bad if you're prepared, but a lot of people were probably avoiding them before, so I imagine there was a distinct lack of experience... The data bears that out, with 3.6% of losses inside a Garrison, compared to 2.3% in Beta 6.

422 runs (8.1%) acquired RIF, two-thirds of them inside the -8 Garrison, although a fair but smaller number did it in -7, and relatively few later than that. RIF keeps getting improvements, some in the Beta 7 updates, more in Beta 8, and likely yet more in Beta 9, so the style is definitely coming into its own.

The average RIF run hacked only 9 bots, but the highest number hacked was 124 (by Joel). Players tended to hack combat robots a bit more than non-combat bots. The only hacks that went completely unused so far are tweak_propulsion, likely because it's pretty situational and there are generally better options, and report_prototypes, report_schematics, and report_analyses, probably because their results are both uncontrollable and can be achieved through other means anyway, thus Relay Couplers are too valuable to spend on them.

Among RIF runs, only an average 3.27 Relay Couplers were attached, although Joel attached 43 in a single run. And wow, apparently I took the second spot in this category with 41 attached (and the third spot with 24 xD). I did rather like RIF, and streamed a couple runs with it, and plan to do some more, having really barely scratched the surface so far when it comes to robot hacking... 131 (31%!) of RIF users never attached a single Relay Coupler, probably because they were still learning, or unprepared for that Garrison ;)

I've compiled a chart of hacks by the number of times they were used, excluding anything with fewer than 10 uses total:



Hacks that can target both combat and non-combat bots are not marked, but some of those are especially popular as well, such as overload_power, a great hack for making a mess. parse_system is popular of course, but it only provides meta information and plot stuff so isn't as meaningful here. That was something which could always be done even before RIF.

formatsys_high is by far the top hack, the good old assimilation hack which is expensive but if you've got the coupler power might as well use it! overwrite_iff is the cheaper temporary version, useful for weakening enemy groups when a newly formatted ally would end up just dying anyway.

I'd like to compare this data to what we get from Beta 8 to see if there are any shifts as players learn more about the different hacks, and perhaps take advantage of RIF more often as it's both been buffed and works well with FarCom.

I was curious about gunslinging stats since Beta 7 made that guaranteed, plus more players might want to take advantage of it as a result. Sure enough, the average amount of gunslinging doubled in Beta 7, increasing from an average 1.68 secondary targets to 3.03. (I did a gunslinging run myself.) kiedra led the pack with a whopping 104 secondary targets in a single run. By comparison my own gunslinger run only had 29, but that's still pretty high ;)

Melee secondary attacks were also buffed, and while they weren't recorded in Beta 6 so we can't compare, according to Beta 7 data there were an average of only 0.51 per run. That's fairly low, but then most people prefer taking on targets one-on-one with their melee builds. That said, PhenomPhear had a run with 32 (!) secondary melee attacks. To be clear, this is attacking more than one robot with a single melee volley.

Comparable data that we do have for Beta 6 includes follow-up attacks, i.e. hitting a single target with more than one melee weapon. There was a definite surge there, where the average run went from 0.90 follow-up attacks to 2.00. The averages in all these are rather low since many don't use this mechanic at all, but it's extremely effective. Back in Beta 6 the follow-up record in a single run was 284, but that was topped in Beta 7 by GJ with 527! That's quite a beat-down...

Meta stats:
  • 89.9% of players use fullscreen (30.9% of those use borderless windowed mode)
  • 7.3% use hjkl for movement
  • 12.8% prefer ASCII
  • 21.8% don't touch the mouse
  • 1.5% of players have rebinded at least one command
  • Smallcaps (the default) is the most popular font choice (65.9%), followed by Terminus (16.2%), and X11 (7.8%) (10.1% use some other font)
  • 1080p is the most common resolution (53.8%), with 28.4% of players using something lower, and 17.8% higher
  • 1.5% percent of players are using a render filter, fewer than in Beta 6. Most of those are activating low-contrast mode.
  • 72.6% of players are using Tactical HUD mode
« Last Edit: April 19, 2019, 01:19:41 AM by Kyzrati »
Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #26 on: November 03, 2019, 10:46:00 PM »

Beta 8 Stat Summary

It's the end of an age! This is the final stat summary based on the system we've had in place for past several years, as we migrate to the new database-driven system from Beta 9 onwards. That system is not yet available, and I'm not sure how the future stat analysis will be handled, but the data is being uploaded (assuming you've opted in via the options menu) and in the meantime you can already explore the online data for your own runs via the URLs appended to your scoresheets.

The full archived Beta 8 leaderboards can be found here, with scoresheets for all runs accessible at the bottom.

For Beta 8 we have a total 7,225 runs that made the 500-point threshold (86.2% of a total 8,381 runs).

Sadly there was a server issue in June leading to about 10 days in June during which scoresheets could not be uploaded, and I estimate we likely lost approximately 300 non-anonymous submissions. And although the main problem was eventually fixed, in the subsequent months a number of scoresheets failed upload for some reason and we couldn't figure out why or even precisely what, only much later discovering that it seemed to be some of the +/++ extended run wins that were failing, and some others as well for unknown reasons. So although this leaderboard's share of run data is quite large because the 8/8.1/8.2 period lasted quite a while, let it be noted that we're unfortunately missing about 4% of the runs, including some wins. (As per my announcements regarding this issue, however, some players proactively provided me with their scoresheets for inclusion, so some of them were recovered and added retroactively.)

699 unique named players submitted at least one qualifying Beta 8 run, or 10.3 per player, an 11.2% decrease from Beta 8, although Beta 8 itself was a significant jump from 7.

Beta 8's win rate continued on a downward trend, from 8.2% in Beta 6 to 6.6% in Beta 7 to the latest 4.9%. This is due to the number of new players remaining steady whereas a lot of the old guard capable of reliably winning runs taking a break during the unusually long Beta 8 cycle (also back in Beta 7 they were speedrunning, which leads to far more deaths :P).

354 wins were submitted by alice_fexa (28), 3.14 (25), Pimski (15), Trione (12), Naughtyusername (12), Dullahan (12), lsend (11), Alexbot (10), Terminus (9) PlasticHeart (9), Mitaro (9), Horse (8), Valguris (8), cptwinky (8), Xii (7), Main.E (7), Solar Sloth (7), MTF (7), Leebears (6) muxecoid (5), whitenitro0 (5), Palikka (5), Fod (5), Zyalin (5), Gaidara (4), Owlsley (4), Howard (4), Joshua (4), TBExtent (4),Quelan (4), Raine (4), Torako (3), JSLIN (3), eldritch (3), Puzzlebark (3), Malth (3). There were (2) wins each by TheBestTreadsBuild, SNACKS, fpsbossfight, Cheeriohz, Styles, Kyzrati, inSANE, zzxc, Rumbl3, octapi,Sorctek, AlmondRoll, Sylvia, eb, tiniuclx, JC, aoemica, Amphouse, and (1) by Benjamin, MojoPoop, mat, Daniexpert, TheEpicPerson, Nikola Kolodziejczyk, IAMCRAIG, Nikolayag, Cracklepappy, cinereoux GridBugBear, Omewes, Tone, l.n.c.a., Vect, ConfusingDalek, Mutter, PerrySimm, SSMR, JackNine, Suslik, Cookie, Gobbopathe, Maiker Moran, QudTheCat, ctrl_alt_x, Cowmind, happylisk, BCN, Bert, RNGesus, Wayward Satellite, KirdTheApe, wilk, 277, d4void, Gitida,Finestep, Marcus Aseth.

alice_fexa took the lead this time in total win count, and Pimski once again broke Cogmind's high score record with a whopping 355,452.

PlasticHeart's "winners mutation distribution" graph is back, growing increasingly complicated as yet more combinations are possible due to new strategic options:



According to this, two-thirds (63.8%) of winners obtained access to FarCom, one of Beta 8's big new features, although 11.3% of those definitely stole prototypes from the Exiles and later had their access rescinded. An additional greater number no doubt also lost access, unknown because the old scoresheets didn't include any explicit indication of stolen prototypes (we'll have that data for Beta 9), so the true number of runs actually using FarCom throughout might be half or less.

I've been curious about the total number of runs by individual players, something we haven't looked at before, so I put together that data for Beta 8:



A little more than half played only one or maybe a few runs, those who generally just try out what's new in the latest Early Access release then go back to whatever else they were playing, with the second largest category being the core group of players, many who keep coming back for more and do quite a few runs aiming for not only the latest content but also extra achievements and lore. At the highest end are some pretty crazy figures!

The top three run counts by individuals:
  3.14              405
  alice_fexa        343
  Naughtyusername   225


Alongside the run counts I also tallied the average number of evolutions by all the runs in a given category, showing that the players who did more runs also performed better on average.

But the averages were only so telling so I thought it'd be even more interesting to look at the single best run from each player and average only those runs, finding a much clearer disparity between the groups, and a clearer connection between number of runs and how far players are capable of reaching. (Of course depending on the player and their style that could also be attributed to sheer luck in some cases, though I don't think that accounts for the majority of the difference, based on relative scores.)

Exiles

Beta 8 added the first new map in a while, so of course a lot of players are going to be interested in visting it, especially considering it's in the early game. Almost one-third (31.6%) of runs (2,285) visited the Exiles in their lab, which is quite a lot! It's near the beginning so not too hard to reach, although getting there does require passing through the Mines, which is where 9.7% of runs ended this time around. This is up from 5.2% in Beta 7, when the Mines only had their other regular benefits rather than being a route to the EX-prototypes and their FarCom system. (Note that leaving the Exiles lab forces players to pass through a second Mines area as well, which is yet another chance to die there.)

Funnily even 2.7% of runs ended in Exiles itself! This seems like it should be more rare because it's a friendly map and although dangerous things can happen there, you're protected from them, but this value comes mostly from players who are intentionally attacking them which doesn't always go so well considering how powerful they are relative to how weak Cogmind is at the start. Still, by now it's clear that if there's a potential benefit to be had by being evil, a portion of players are going to do whatever it takes to get it :P

Trackers/DC

In other branch activity, I was curious how the addition of a new deadly enemy, Trackers, affected the number of people using "DC" compared to before. Of the runs that routed through the necessary map, 17.6 used DC in Beta 7, compared to a much smaller 10.1% in Beta 8. Admittedly this doesn't capture the portion people who didn't visit the necessary map in the first place specifically because they didn't intend to do this, but that portion won't be too significant because there are great benefits to be had there even without "DC". The shift could also be attributed to a factor mentioned earlier, the old guard (more likely to use this strategy) playing less of Beta 8 while a greater number of newer players don't even know the feature exists, but we also have anecdotal evidence from the community that suggests this strategy is being relied on less often as a result of its new consequences.

RIF

Strategically speaking FarCom actually goes pretty well with RIF, which was just introduced in Beta 7, so it might be worth taking a look at if/how RIF usage changed.

687 runs (9.5%) entered at least one Garrison, and 194 (2.7%) died inside one, down from the 3.6% in Beta 7, meaning either fewer people entered Garrisons since RIF was no longer the New Thing, or those that did got better at tackling Garrison challenges. 436 (6.1%) of runs actually used a RIF Installer, down from the 8.1% of Beta 7, so indeed fewer were likely interested in getting this functionality. (I think this particular data would be somewhat more accurate if we only checked whether each player acquired RIF on any of their runs, but that's more complicated with the set I have so I'm not doing it that way.)

The average RIF run hacked only 9 bots, the same average found as in Beta 7 (split evenly between combat and non-combat targets). aoemica hacked the most in a single run, 146! Among RIF runs an average of 2.87 Relay Couplers were attached in a single run(fewer than the 3.27 in Beta 7), compared to aoemica's 66 couplers on one go! Somewhat surprisingly 171 RIF-capable runs didn't attach a single coupler, an even higher ratio than last release.

I mentioned in the Beta 7 stats that I'd like to see if/how relative bothack usage changed from one release to the next, but it really hasn't shifted much aside from a noticeable rise in find_chute uses as Waste-hunters discover how useful that one is :) (also good for slow bots to use in avoiding an undesirable visit to Crusher land).

HCP

With the Hcp. Storage Unit finally making its way back into Cogmind, I thought it'd be interesting to confirm that it's affecting the average and upper storage capacity. That does pan out in the data...
             Largest    Average     Most      Average
             Inventory  Capacity    Carried   Carried

Beta 7 avg   12.5       9.1         11.3      7.3
Beta 8 avg   15.9       11.3        14.3      9.5

Beta 7 max   60         35          48        34
Beta 8 max   62         41          62        40


Pay2Buy

Not included in the regular stats presented here, we've had a total of 105 Pay2Buy runs uploaded since that mode was introduced in April. The mode is still accessible and as of now a total of 50 unique players have both played the mode and uploaded their data. The leaderboard for the event period is here, including access to the original announcement and many more event-specific stats. There were 39 players who participated in time to be included for the leaderboard.

Meta
  • 91.0% of players use fullscreen (30.4% of those use borderless windowed mode)
  • 5.7% use hjkl for movement (41.2% prefer mouse, 11.3% use arrows, 41.8% use numpad)
  • 11.0% prefer ASCII
  • 21.3% don't touch the mouse
  • 1.6% of players have rebinded at least one command
  • Smallcaps (the default) is the most popular font choice (67.5%), followed by Terminus (17.0%), and X11 (7.3%) (8.2% use some other font)
  • 1080p is the most common resolution (53.8%), with 28.9% of players using something lower, and 17.3% higher
  • 1.3% percent of players are using a render filter. Most of those are activating low-contrast mode, followed by brightness tweaks as the second most popular adjustment.
  • 72.9% of players are using Tactical HUD mode
« Last Edit: March 28, 2022, 02:45:21 AM by Kyzrati »
Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #27 on: January 08, 2021, 07:41:01 PM »

The Beta 9 stat summary is finally here! And it's big. Really big. So big that it's on the blog :P

Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #28 on: April 01, 2022, 03:29:38 AM »

Beta 10 Stat Summary

Once again a major new Cogmind version has been released, and that means it's time to look back at some of the stats for the previous version! Beta 11 was the longest ever cycle in Cogmind development history (being the huge overhaul that it was), meaning Beta 10 was active for quite a while and therefore includes a lot of runs, though that release itself was not uniquely interesting from a data perspective considering that much of the work that went into it was of a cosmetic nature, including for example the ambient audioscape and QoL-related features. By comparison I'm sure the future Beta 11 stats will be very interesting when those come out, given how much was changed or added, but that's for next time on the blog. Here we'll be using the older forum-based format to explore some of the usual player stats, as well as of course poking around here and there looking anything that might be relevant to Beta 10 in particular.

As usual the leaderboards have been reset and the full archived Beta 10 leaderboards can be found here, with scoresheets for all runs accessible at the bottom.

For Beta 10 we have a total 14,338 runs that made the 500-point threshold (96.7% of a total 14,820 submitted runs) to be included in the leaderboards and results here. That count includes 12,545 regular runs, which we'll focus on for the purposes of the stats below, as well as 1,793 runs completed under special events or challenge modes, which I'll write a bit about separately.

1,320 unique named players submitted at least one qualifying Beta 10 run, or 10.5 per player, returning to earlier levels after having dipped a bit in Beta 9.

Difficulty

Beta 10 was the second release after adding the difficulty selection menu, so the change in player distribution wasn't as stark as that seen between Betas 8 and 9, though we can still see a bit of a downtick in the portion of players using non-Rogue modes.



However, the number of players in all categories increased for Beta 10, so the change in distribution here was at least partially a function of a more hardcore roguelike players joining the community (+60.7% vs. +20.3%/+12.2% for Adventurer/Explorer). There was also likely a smaller number of players gradually shifting out of other settings (especially Adventurer) to play Rogue, having gained more experience and wanting to seek extra challenge.

Wins

The win rate in Beta 10 was 4.1%, a new low down from 5.7% in the previous release (and only slightly below Beta 8), though there's generally too much noise to read into such a broad value. As a better comparison to other releases, for my own purposes I looked at the number of Rogue-level players who won at least one run, mainly to more easily filter out runs where the player was doing weird stuff or not aiming to win, and excluding the other difficulty settings since they can use quicksaves (and aren't what the content is balanced around anyway).



There we see a fairly consistent rate across all releases from the past few years, solidly in the 12~14% range. I imagine the win-one-run rate won't fall too much in Beta 11 (those who can already win will find the new version challenging, but eventually figure out how to deal with the new content), although next time we do these stats I'll be very curious to see if the general win rate falls further, which is likely given that the difficulty of Complex 0b10 itself was increased.

516 wins were submitted by GJ (56), Jezoensis (47), CrossedViper (28), MTF (17), cptwinky (13), Wyvric (11), 3.14 (11), aperiodic (10), Savant (10), Alexbot (9), blelm (9), ChromeKnight (9), derk (9), ktur (9), LU-1G1 (9), Perseus (9), zzxc (9), Phaethon (8), Farfid (7), lsend (7), Palikka (7), thingamajigger (7), alice_fexa (6), Michael (6), Sanctumkeeper (6), cavemonk (5), KTG-V2 (5), leiavoia (5), OR-2NL (5), sentrybuster (5), WayneKnight (5), Exp. Geonium Core (4), Floating Point Exception (4), Pimski (4), the cat (4), Torako (4), Andru7428 (3), Azken Zutik (3), Boris (3), Cookedpoo (3), Demise (3), Finestep (3), Kenzurith (3), kerapace (3), Leebears (3), Mares (3), Tone (3), whisper (3), xerxia (3), youngster (3), Buffy (2), Coronach (2), Cracklepappy (2), Dawn (2), Decinym (2), Hermelin (2), KTG-V3 (2), MoonDog (2), muxecoid (2), Nikolayag (2), NoNeedToExplain (2), Peace (2), pointone (2), rath (2), Saya (2), Terminus (2), Valguris (2), Vectis (2), Xeram (2), and (1) by AF_C2_5001, albenzo, aoemica, Avari3l, buthix9, cinereoux, Cinquain, Cog, Crorem, CV-BEE, DaddyWalrus, darlorth, DEZA, Dhoby_Ghaut, Fleshy Vegetable Bird, FR-35H, gammafunk, Geraldon, Gitida, Gobbopathe, HawkeyeFierce, Hilbert, Howard, Infomantis, JackNine, Joel, Joking_Phantom, Joshua, kasaphescent, Krahie, Kyzrati, malcneuro, Matie Bal, MinorBread, MrPotato28, Nalzok, Needle, PerrySimm, Peter, PlasticHeart, raddakar, ShuSh, SirMrDrProf, soymachine, spruce, StarSirius, Stryker, szymekc, The Phoenix Tome, TheEpicPerson, Tmvn, Trione, Veritas, viper, Yosh, Zailor.

The Return of GJ inevitably means GJ is a strong contender for win count leader, though we also see pretty high numbers from other players, too. Jezoensis began with quite a few Adventurer mode wins before moving on to challenging Rogue mode (and getting some wins there, too), and CrossedViper played a relatively huge number of runs (287!).

Of significant impact when it comes to our stats this time around is that core/OG players are somewhat underrepresented because many among that group are patrons and therefore played Beta 11 prerelease versions throughout the majority of the Beta 10 leaderboard period. For example I first realized this when I noticed aoemica has only a single Beta 10 win on record, but on closer inspection submitted 53 wins while helping test Beta 11 prereleases :D

Among the runs included above, MTF set a new record high Cogmind score of 423,663 (scoresheet) by sterilizing 10 different maps (mostly in the late game) and earning 140,168 of that purely from bonus points for high-alert combat kills, destroying 3,721 robots in all.

Another run of particular note set an even higher record, a whopping 1,012,980-point run by Azken Zutik, excluded from our stats only because it was played under the Gauntlet challenge mode. That run spanned a mind-boggling 91 hours, made possible by approaching the challenge as a bothacking/RIF build that led an army capable of protecting Cogmind from almost all attrition, repeatedly visiting Garrisons at the same depth to restock. 370,220 points came from allied kills alone. Even the 582 regions visited resulted in a bonus of 116,400 points, which is just... crazy xD

Very impressive stuff!

While we're on sterilization, that was the new mechanic added for Beta 10 as a sort of final cutoff for destruction, added especially for players otherwise endlessly crushing 0b10 on a single map using easier settings such as RPGLIKE mode or the Explorer difficulty. Only 17 players sterilized any maps at all, across 29 runs, so a pretty small number. The majority of those only sterilized one map, maybe two, aside from MTF's 10-map sterilization run and a 5-map sterilization by NikolayAg. (Since then I also attempted a "sterilization run," but quit after my 4th sterilization going into the late game because I was streaming it and my strategy was taking too long :P)

I'm sure we'll be seeing the frequency go up significantly in Beta 11, if only because this mechanic was further reused in DSFs, as well as being much easier to trigger there, and even offering a decently reliable strategy.

Cogmind had seven win types prior to Beta 10, with that release adding two new ones for a total of nine. Both of the new endings and their requirements involve the extended game and are therefore very spoiler-heavy, so I'll put this (small) section behind a spoiler tag:

Of course we saw lots more of these achieved during the Beta 11 prereleases as well.

During Beta 10 I've heard yet more questions about the "length of a winning Cogmind run," something which is often compared across different roguelikes.

We know that it can vary greatly depending on goals and strategy, since Cogmind is designed as a "wide" game rather than being outright long, and you can have anything from 20-minute speed run wins to 10-hour marathons, or the more reasonable/likely/common 100~200-minute runs in between, but let's look at some real data on that for fun...

Again it depends on strategy and goals, so our statistical results are of course skewed by whichever types of players have greater representation. A more accurate approach would be to also look at the build types used here, and the number of optional areas visited, but that's probably not especially helpful beyond the general range we're looking to visualize.



This graph is composed of 181 w0/non-extended wins and 335 extended wins, and we can see that the shape of each is similar, with peaks at the lower end and trending downward into longer and longer runs, ending with another spike consisting of outliers doing special runs. Obviously extended runs are going to increase the average somewhat, though it's not an especially huge increase--extended wins averaged 5.4 hours, while others averaged 4.7 hours (the median for each category was 5 and 4 hours, respectively). As expected, a little over half of "standard" (non-extended) wins fall within the 2- to 4-hour range.

Mechanics

Because Beta 10 added six new RIF abilities, I thought it'd be interesting to see whether that had any impact on adoption of that strategy, but a comparison between the last two versions shows there was no meaningful effect (stats only include those runs which reached a depth of at least -6, which was about 44% in both versions):
                                        Beta 9      Beta 10
Used only 1 RIF Installer               6.0%        6.4%
Used 2 or more RIF Installers           9.1%        8.2%
RIF builds hacked at least 3 robots     65.9%       61.0%


However, the top 100 RIF runs by bot hacks applied increased pretty significantly, up from a median of 89 hacks in Beta 9, to 117 in Beta 10. So RIF builds are riffing harder. The median number of RIF Installers used in those runs was 7 in Beta 9, vs. 8 in Beta 10.

I also took a quick look at the most common primary classes used across a run, but didn't see any significant changes between the in-depth Beta 9 analysis and more recent Beta 10 runs. We'll almost certainly see more changes there headed into Beta 11, now that the "Hauler" concept has been refedined, so we'll look at those again then. This time around there were 76 unique build combinations (out of a theoretical maximum of 208), as opposed to the 67 from Beta 9.

Special Events

Beta 10 included some special events, as with previous versions (Beta 9 was especially prolific in this regard--see stats), though not quite on the same scale. The big one was the lore-centric ARG, and we also had a quick AFD mode called "Volatile" which wasn't even explicitly announced, so I imagine not as many people even actually played that one. Of course players are also still able to go back and play most prior events if they want to, and some do that, especially with RPGLIKE mode, for example. So let's take a look at what was played, and how much...

249 unique players (17.6%) played at least one special event run, with 74% of them playing more than one run. There were 1,746 special event runs in all (13.89% of all runs).



Sure enough, not many actually tried out Volatile mode, it being both unannounced and pretty short in duration. Another thing you'll notice is the much higher win rate under RPGLIKE, not exactly a surprise since it adds a healing mechanic which overall makes Cogmind an easier game. And of course the dominant event of the version was Forbidden Lore, being the major one held during Beta 10, and also requiring multiple runs to fully explore its content. That requirement was somewhat offset by the community working together in some cases, but it was still a run-heavy mode due to the drip of puzzles and metaknowledge it would offer with each run. Forbidden Lore also had a lower win rate since more players would be likely to simply self-destruct once they had what they were after, if it looked like their next clue was in a place they could no longer feasibly reach on their current run.
« Last Edit: April 01, 2022, 04:39:01 AM by Kyzrati »
Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #29 on: March 07, 2023, 09:19:50 PM »

As part of my articles on design of the Beta 11.2 "Polymind" event/special mode from late last year, I analyzed a bunch of relevant stats you can check out over there: Special Mode Design, Polymind Part 2: Players, Stats, and the Future.

Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #30 on: March 21, 2023, 12:43:24 AM »

Big move from Beta 10 to Beta 11, calling for a big stat article filled with graphs, charts, and numbers :)

Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon

Kyzrati

  • Administrator
  • True Cogmind
  • *****
  • Posts: 4266
    • View Profile
    • Cogmind
Re: High Scores and Stats
« Reply #31 on: March 12, 2024, 08:53:12 PM »

Beta 12 Stat Summary

Beta 13 is out, leaderboards have been reset, time to examine some stats from the previous version! Our last huge stat review was done after Beta 11, in 2023 here on the blog, and that being an important milestone with regard to balance, that particular review was quite extensive. This time I'll be looking at somewhat fewer categories, but it's still nice to see where certain aspects stand, especially any newer features.

Beta 12 leaderboards have been archived here, with scoresheets for all runs available here.

During Beta 12 "Encrypted Comms" (March 2023 ~ Feb 2024), 790 unique players submitted 8,592 runs that met the minimum threshold for inclusion. 296 (3.4%) of those were played in a special or challenge mode (a lower percentage than last time (8%) since no new modes were released during Beta 12), leaving us with 8,296 runs for our data analysis. (There are also many prerelease runs by patrons, but those are excluded as well, since the version is still in flux at that point.)

Wins

The overall win rate was 4.7%, so fairly middling as far as historical stats go.

Among the 7,143 Rogue-level runs (86.1%), 97 (16.5%) unique players earned at least one Beta 12 win, seemingly confirming that the increase in that metric following the Beta 11 rebalance is here to stay. Maybe it will reverse down again next time in Beta 13 with the influx of new and returning players now that we have map zooming and the new upscaled UI layouts? We'll see, but for now it looks like Beta 11 did indeed have a sizeable impact on winnability, perhaps making it a bit easier for some (?)... or our sample size is still too small to draw conclusions like this, what with different folks coming and going over the long term :P



(This is an updated graph from our Beta 11 stats--the Beta 12 prereleases would show a similar comparison, but it's not being included here.)

This time around I compiled a new category of win data people have been curious about: win types. As of Beta 10 there are 9 different win types, and according to Beta 12 records, 11 Rogue difficulty players have at some point earned every single win type at least once (there might be more who did not play Beta 12 in particular). Adding in Adventurer/Explorer runs does not change the count, which makes sense since it's generally the more hardcore players going up against every single ending (and winning).

Some other Rogue-specific win type data (remember that despite this including historical runs from earlier versions, our data is only looking at those who at least played one qualifying Beta 12 run):



Thus 32.0% of Beta 12 players have one at least one run at some point before, and 15.6% have at least two different types.



w0 is the most common, of course, and although we we can't really know how many "w0 of shame" that includes, a respectable near-half of runs are alternative wins requiring at least one late-game branch visit. Some of the win types are clearly rare, either because they're difficult to discover, or are simply really challenging (leading in many cases to a w0 instead when the primary goal was not achieved).

Among the unique winning accomplishments that aren't explicitly represented by a win type, special shoutout to Runia for being the first and only player to have successfully managed to fire the Drained L-Cannon and even use it to get an extended ++ win. Some video and summary/notes by Runia here--EXTREME EXTENDED GAME SPOILERS.

Scrap Engine

One of the main new features of Beta 12, even though not yet fully integrated plot-wise, was the already mechanically-complete easily-acquired Scrap Engine. We have a few basic stats on that.

472 runs (5.7%) used the Scrap Engine (SE) at all, and a total of 127 unique players (21.6%) tried it at least once. 71 winning runs used the SE, or 15.0% of all SE runs, most of them using it heavily as opposed to for a short time. Naturally these same folks generally created and modified greater numbers of constructs.



(Construct creation does not include modifications to existing constructs, a much greater number by around a factor of 20.)

Garrisons and DSFs

One of the other major enhancements of Beta 12 was the new Garrisons, so that probably means more Garrison visits and use of related mechanics.

18.3% of runs entered at least one Garrison in Beta 12, compared to 12.0% in Beta 11. 4.0% entered three or more (2.6% did so in Beta 11).

But continuing to follow our general RIF stats in this version, we can can see how garrison visitors were somewhat less interested in bothacking this time around, thus were clearly eager to enter Garrisons specifically to see the sights and/or acquire the other new benefits.



Speaking of new benefits, at least 1.2% of runs acquired the Encrypted Comm Array. The true number is likely higher, but it wasn't until Beta 13 that I added a way to more easily analyze that in the scoresheets, so I'm basing it purely on the number of runs in which it was actually used to rendezvous with allies (technically not a requirement to ultimately benefit).



One of our other small maps-between-maps, DSFs are also relatively new, having only been added in Beta 11. Last time I reported that 10.5% of runs visited at least one, and interestingly that level continues to hold, with 10.9% of runs visiting a DSF in Beta 12. It appears we can say with confidence that DSFs have their uses and are doing just fine--I know I've ducked into them on occasion either for fun, or to avoid trouble or get a quick build together.

Strats and Miscellany

Traps, reborn as a strong supporting strategy as of Beta 11 and gaining significant popuarity from then on, continue to enjoy the same popularity with 445 extractions by the top 20 trap users in Beta 12, compared to 428 before. The average number of traps and overall Trap Extractor use remained similar. Traps good.

Among the many uses of traps is convoy interception. 21.7% of runs that made it to -7 (lowest convoy depth) reported at least one interception. Now technically the player may not be the one who triggers it, but for some folks we can see it's quite clearly deliberate, especially those who intercept numerous convoys, and just happen to be heavy trap users, one of the most reliable ways to accomplish that task...



Notice how many of the leading convoy interceptors are also on their way to winning runs. Pure coincidence, I'm sure ;)

From the data I think everyone's come to terms with the Beta 11 fabrication overhaul and is making healthy use of the system. 35.2% of runs built at least one part, only a little lower than Beta 11, and an average of 1.38 parts were built (1.6 in Beta 11, 2.5 in Beta 10). (Note: As usual fab stats are only looking at runs that made it to at least -6.) 12.4% of runs built a robot, 0.53 on average.

20.6% of runs carried at least one Authchip, significantly higher than the 11.8% of Beta 11, and 46% of those runs used at least one of them to build something, not as high as in Beta 11 when everyone was still getting familiar with the system and how they wanted to make use of it, but significant enough to see that a good portion of players know what they want out of Authchips and are packing them in advance with a possible goal in mind.

Also with respect to Fabricators... only 4.6% of runs that reached -6 used some amount of fabnet, down from 7.9% in Beta 11 when it was introduced. And only 3 runs hit the cap (15%), down a lot since last time as well. So it does look like the previous release was more folks who wanted to try out a new feature (after even discovering it anyway--I'm sure many don't know it exists), and now it's settled down to being just another tool that might get some use in the right kind of run. MTF autooverrided the most bots in a single run with 40, and at only a 6% peak throughout, so it also helps to spread out lower numbers over a greater stretch of depths (he didn't even start until -3, then had a fabnet up for the remainder of that run which went on to an extended win).

One of the activities that's been somewhat more common in the Cogmind community lately, either as a goal in itself, or something to survive after causing too much trouble for some other reason, is sterilization. Forcing Complex 0b10 into a complete floorwide meltdown can be quite a challenge, rewarding if pulled off, and let's face it, is pretty funny when you get to walk around an empty floor with hopefully enough cooling to loot the place before being forced out :D

Surprisingly 354 runs (4.4%) sterilized at least one floor in Beta 12. Less surprisingly, none of the top 10 sterilization runs went on to win...



Despite not quite reaching the end, leiavoia did top the overall leaderboards within their epic sterilization run for the sheer score achieved by doing this, even without a win. As you can see some sterilization-oriented runs also tend to be on the longer side--it takes a while to both set up for and destroy that many bots while avoiding a swift death!

17.2% of the runs that completed two sterilizations did go on to win, however. 10.1% of the runs that only completed a single sterilization ended up winning, more than twice the overall win rate. Sterilizers are serious, 0b10.

(My own last run ended up sterilizing -3/Research in order to escape after I was blocked into a corner for plastering one too many walls with robot guts. It was a win, too, although not included in these stats because that's Beta 13...)

And finally, following up on one of Beta 11's new strategic options that I covered in more detail last time: the Exo build. Only 11 runs used Exo in Beta 12, or 0.1% of the total, compared to 82 (0.8%) in Beta 11. It was a new thing back then, so not surprising there was more uptake in the previous release. 95.2% of those which entered and survived S7 did not integrate (compared to 73.1% in Beta 11). It's usually a last resort sort of thing for most people, or something to fool around with, since it either offers less challenge, or a very different kind of challenge (it's not like everyone easily survives using it :P). I mentioned before that it might get further tweaks, but so far it seems about where I'd like it to be insofar as it serves its purpose in terms of lore, balance, and strategic options.
Logged
Josh Ge, Developer - Dev Blog | @GridSageGames | Patreon
Pages: 1 [2]