Tuesday, June 6, 2023

Game 491: Adventure Dungeon (1983)

Adventure Dungeon
Independently developed; published in March 1983 CLOAD magazine (a cassette magazine)
Released 1983 for TRS-80.
Date Started: 3 June 2023
Date Ended: 5 June 2023
Total Hours: 6
Difficulty: Very Easy (2.0/5) in the sense that every game is a "winner"; Hard (4.0/5) to get the top score and survive all dungeons.
Final Rating: (to come later)
Ranking at Time of Posting: (to come later)
Adventure Dungeon is an interesting title that commenter Dungy unearthed a couple of years ago by going one-by-one through old disk and cassette magazines. It was written by David Lo of Port Coquitlam, British Columbia. Lo was a modestly frequent contributor to CLOAD, although this is the only game that's surfaced with RPG aspirations. It's not quite an RPG by my definitions, but at the same time, it offers an interesting approach that had me thinking a lot about optimal strategies.
The setup is that you're trying to join the Wise Council, which requires you to prove yourself in a contest. Asking the player for nothing more than a name and a sobriquet, the game dumps you on Level 1 of a five-level dungeon. (You can technically choose to start on a higher level, but I'm not sure why you would.)  Each level is 30 x 13 squares. Scattered across these 390 squares are 18 monsters, which do not move, and 13 treasures. Collecting treasures and defeating monsters increases your score. If you defeat all the monsters and collect all the treasure on one level, you move to the next level. If you clear all the levels, or die, the game ends and you learn from the Wise Council whether you earned enough points to join and, if so, at what rank.
I was "confused" because I thought the game was asking for my class and the instructions hadn't said anything about that.
While simple in concept, plot, and mechanics, Adventure Dungeon introduces one devilish twist: When you encounter a monster, you have to wager a certain number of your hit points against its hit points. If you wager exactly the same number as the monster has, your odds of beating it are 50%. If you wager double the number the monster has, your odds are 100%. In between 1x and 2x the monster's hit points is a linearly increasing probability of success.
Fighting an orc on Level 2. Wagering 340 hit points against his 271 gives me a 63% chance of success.
So why wouldn't you always wager two times the monster's hit points? Because character growth depends on how much risk you take. Both your maximum hit points and your score increase in proportion to your risk. If you play it safe and wager 2x, you get no increase at all (and values above 2x will actually drop your values). 

For instance, on Level 1, you meet a spider with 51 hit points. You have 300 hit points. You can attack it with any number of points between 1 (about a 1% chance of success) and 102 (100% chance). If you wager 101 points, the largest number that would give you anything in return, you would earn 1 hit point and 10 points to your score. But if you were successful when betting only 1 point, you would earn 101 hit points and 990 points to your score. Earning points is important because monster hit points increase depending on the number of monsters you've already killed. Enemies start as weak as 2 hit points (dwarves, trolls, and harpies at the beginning of Level 1) and as strong as 22 hit points (dragons at the beginning of Level 1). As you slay their fellows, each monster experiences a change in its hit points, but not in a predictable way. (These changes are solely dependent on the number of monsters remaining and not on your score or hit point total. I checked.) For instance, here's how the orc's hit points change depending on how many monsters remain:
Do you think anyone has ever made a chart with "orc hit points" in the title before?
I don't have a TRS-80 emulator with save states, by the way, so creating that chart meant starting enough new games in a row to encounter the orc after each number of enemies while surviving all those previous enemies. Yeah, it took a while. I couldn't possibly do it for every enemy on any level, but the highest hit point value I saw for any creature on Level 1 was the dragon at 519.
Obviously, 519 is well above the character's starting hit points of 300. The character needs to have earned 738 hit points on the level to be sure of defeating the dragon, if he encounters it last, and he can't possibly earn that many if he plays it safe with all the other enemies.
Every level has all of these monsters, randomly scattered. They are roughly in order of most difficult to least difficult.
On Level 1, you have the advantage that you can uncover the entire level before picking up any treasures or fighting any enemies. There are no barriers. Every step reveals the eight squares around you. You just have to walk carefully. The instructions tell you what symbols match up with what monsters. You could attack them in a strategic order. But as we've seen, the number of hit points that each monster has at each stage is unpredictably variable. You'd have to learn the "right" order through a lot of trial and error, perhaps creating a matrix of each monster's hit points at each stage.
And each level has all of these treasures.
There are a couple of other confounding considerations. If you wager less than 1.5x an enemy's hit points, the enemy automatically kills you if you lose the combat. If you wager more than 1.5x, the enemy "lets you leave" but you lose what you wagered. This is hard to recover from some combats. For instance, in the only session where I made it to Level 3, I met a balrog with 634 hit points while I had 2083. I wagered 1110, or about 1.75x the balrog's total, which gave me an 87.5% chance of victory. This was the standard I had been following. In this case, I lost the odds, and my hit points dropped to 973. The balrog was still on the map, but now I could only wager 973 hit points to his 634, giving me only a 72% chance of winning. Fortunately, I won the second time; otherwise, I would have ended up with 0 hit points. As it was, the number I'd lost hurt me when I next met a sphinx with over 1,000.
Without enough hit points to wager, I was "torned to bits."
If you're able to make it to Level 2, you'll note that every enemy has a couple hundred more hit points than on Level 1 (to start; again, it changes and generally increases as you kill their comrades). The terrain is also different, with walls blocking access at some point, making it more difficult to reveal the level before having to fight anything. That strategy becomes impossible on Level 3, where all enemies and treasures are nestled in a crosshatch pattern through which you have to move diagonally.
The weird Level 3.
Level 4 is like Level 2, but without the dots that mark the tiles. Level 5 has a proper maze, albeit a small one. The wall squares are chaotic and unpatterned on Levels 2 and 4, but they create corridors and rooms on Level 5. After you leave each level, whether by clearing it or by defeat, you have the option to have the game reveal the rest of the level before you move on.
Viewing all of Level 5.
If you choose to start on a Level higher than 1, the monsters have the same hit points as if you had started on Level 1. You still have to clear five levels to completely win the game, so the game just adds as many re-iterations of Level 5 as is necessary depending on where you started. For instance, if you started on Level 4, you'll face four iterations of Level 5 after you complete it.
In six hours of playing--admittedly, a lot of that was recording data for the preceding chart--the best score I could accomplish was 8,693 points. According to the instructions, that should have been enough for me to achieve the fifth rank, "Eldar of the Outer Council," but the game screen said I had only achieved the fourth rank, "Member of the Outer Council." The highest rank, "Lord of the Wise Council," requires at least 27,000 points. 
What I received . . .
Versus what I was supposed to receive.
What consumes me now is what strategy you would have to adopt to get that high of a score. The obvious thing is to first carefully map the levels, avoiding monsters and picking up unguarded treasure. Treasure gives you points with no risk. Then, I guess you want to take on the monsters in roughly descending order of difficulty. Although the hit points are variable, in general, dragons, balrogs, wyverns, and hydras are the strongest monsters and dwarves, trolls, harpies, orcs, and black unicorns are the weakest. The trouble is determining how much to bet. Even betting enough to give you an 85% chance of victory gives you only a 50% chance of winning 4 combats in a row and a measly 5% chance of winning all 18 combats on a level. A 95% chance of victory gives you only a 40% chance of winning all 18 combats on the level. 
Here, I've uncovered most of Level 1 before attacking anything.
I actually set up an Excel sheet to start modeling different strategies and likely outcomes. Ultimately, because your score increases in a linear manner with risk, your expected payout from battle alone is identical no matter what strategy you adopt. But the presence of treasure encourages you to play conservatively to increase your chances of staying alive for the next level, where the treasure can add to your score for no risk. But as we've seen, playing too conservatively means that you don't have enough hit points to take on some of the tougher monsters. The whole thing is tying me in knots. I need Ahab to help tackle this one. (Ahab, by the way, is racking up game totals scarily fast. He'll hit 500 before I do at this rate.)
In some ways, Adventure Dungeon really isn't different from any other RPG; it just has fewer variables with more transparent odds. If you were to analyze Ultima III from the same year, you'd have a much more difficult time deriving a single probability of success from any one combat. You'd have to consider enemy and party hit points, attributes, accuracy and damage statistics from weapons wielded, armor class, and the dice rolled for each attack--and that's without any consideration of spells. But if you had all the numbers, the expected probability of success would be determinable.
How do developers properly balance games with so many considerations?
This makes me wonder, in turn, if game designers do create master spreadsheets for individual battles (or combat overall) and tweak the variables to aim for a particular chance of victory, perhaps even accounting for the average player making x errors. I was thinking about this recently in relation to Gloomhaven, to which Irene is addicted. Every battle must have a few thousand variables to consider. Every character card is different, and there are something like 15 character classes, which might appear in any combination of two, three, or four in any battle. Each character enters a battle with 9 or 10 cards out of a deck of maybe twice that many, each card offering a variety of offensive and defensive options, including spells (which depend on various elements having been "infused" during previous turns). The characters' levels determine enemy levels, but characters can have equipment and augments to which enemies have no analogues. Every attack is accompanied by the equivalent of a die roll, but complicated with curses and blessings and various special cards achieved through success in battle objectives.

This is all a lot for the players to keep track of, but it must have been even more difficult for the author, Isaac Childres. I find the game generally well balanced. We started playing at normal difficulty and then moved up to "hard" and "very hard" as we got more experience, but no matter what, we've always had a success rate of about 85%. Either we've been extremely lucky in this consistency or Childres had some way of distilling every advantage and disadvantage, every card, token, and pip, into some kind of standardized value that resulted in roughly the same odds for every battle no matter what combination of characters, levels, and other factors. Which do you think is more likely? And whether Childres had such a formula for Gloomhaven, do you imagine other developers do when writing CRPGs? Or do they just wing it?
When and how often you allow a player to save makes a huge difference, of course. With no ability to save in Adventure Dungeon, even extremely favorable per-battle odds translate poorly into long-term success. A game like Elden Ring can offer much worse odds, at least for boss battles, and still engage players simply by not making defeat the end of the game.
GIMLET? Bah. It gets an 11. The probabilistic strategy element is worth a couple of points, and it's short and replayable. It's not really the sort of game I started my blog to play, but it does pose some interesting questions about the types of probabilities that underlie many other games, and I'm thus grateful to David Lo for prompting an interesting topic of discussion.
Ed. from 06/08/2023: A few days after publishing this entry, I heard from author Dave Lo. He says that his primary influences were Dungeon (1979) for the Commodore PET (which I reviewed 10 years ago), which Lo played at his high school, and Monster Combat, a type-in game published in the February 1981 Creative Computing, which offered the same kind of risk/reward system.

Lo verified that he thinks the best strategy is to collect treasure first, then attack the monsters in descending order of difficulty.

I asked him about the business of writing for diskmags in the early 1980s, and he said it was "very casual--just submit something when it's done." The pay varied from $100 to $200 per program. Lo continued to work on Adventure Dungeon, changing some of the math and adding a spell system, but he never published an updated version.


  1. David Lo, local author of one of my favorite early text adventures, Beyond the Tesseract! I had no idea he ever dabbled in dungeon adventuring. If you had any interest in reaching him, you could find him via his homepage at http://davelo.net/ ... but probably he doesn't have all that much insight to lend to this effort.

    This is an interesting specimen of statistics, gamified... which I appreciate is in effect for many RPGs, but extraordinarily laid bare here.

    1. I wrote to him before I posted the entry, but I didn't give him much time to respond before I published it. He did write back, and I may have some supplemental info soon.

  2. Gloomhaven may be a special case, but I think what is mostly done with boardgames is to test them extensively during the entire development.

    1. I'm sure that's done, but there are so many potential combinations of scenarios, classes, cards, equipment, and other factors that even thousands of playtesters couldn't possibly hit every combination. That's why I wonder if there's some master way of quantifying every benefit and liability.

      I should mention that although the game is well-balanced, there are some notable moments when various character and item powers create unintended loopholes. Any scenario that doesn't require you to kill every enemy is almost trivially easy with a character who has quick movement and invisibility, whether by ability or equipment.

    2. These days, what comes to mind is basically something like Monte Carlo methods - run the game engine with different combinations umpteen times and just measure the distribution of results. Historically, when computers were slower, I doubt this would have been possible for all but the most trivial games. There is no reason to display the UI, update save game files, etc. - all of which slow the "dice rolling" down.

    3. I am a boardgame player, and I can assure you that most of them receive little or no playtesting before release. They're in a rush to get the game on shelves and playtesting is annoying and makes them spend work on revising their game. Instead, you do something called "the living rules" which means you let your paying customers do the playtesting for you after release, and if anyone wants to play a non-broken game they have to know to go to the game's page on BGG (and even know what BGG is) to download v4.21 and discard the printed booklet that came in the game box.

    4. Maybe, but that's because there's a mass of board games released each year, and there's bound to be a ton of crap (especially with Kickstarter games). Buy a game from a well-known publisher and/or designer and you get something that has been playtested for years.

    5. "The Campaign for North Africa" comes from a well-known publisher (SPI) and a well-known designer (Richard Berg), and yet...

      It has a pretty fun wikipedia article explaining how arbsurd it was.

    6. What I mean is that in my experience (in the VG industry), even professional projets are not very well playtested. Usually, the playtests happen on a not-final version, and there is no time to playtest the change done after the final playtest. Good playtests take a LOT of time.

    7. Modern video games suffer a playtesting problem because the commercial need to make market deadlines provides a perverse incentive against testing. Better to get to market early and release a patch. This can happen with board games but the economics of "patching" a board game are radically different, so they're more similar to console games of the 80s and 90s, where a game-breaking bug would be ruinous, but minor bugs just have to be lived with. Board games can handle a certain amount of "errata" - last minute addenda to tweak minor problems in the form of an insert you stick in the box for the second print run. And, of course, the bigger the company, the more they can get away with "The whole first run of this game is broken".

      (Somewhere in the middle, I bought my son a low-end Transformers kiddie console whose gimmick was that you could plug in one of four action figures to change the play mode. They discovered at the last minute that one of the minigames was broken, so they simply didn't release the fourth action figure. My wife and I kitbashed an approximation of the misisng figure so my son would have a complete set. The associated minigame was still unplayable, but he could at least unlock the character for use in other minigames.)

    8. In my opinion, "market" deadline are less strigent now than they were 20 years ago when you could only sell through stores you did not own. While you would do everything you need to kill game-breaking bugs (though I am curious to know whether Chet will play the pre-patch version of U8... I am not sure whether he likes Nintendo-like platformers), fine-tuning the balancing would, presumably, take the back-seat.

    9. I think board games are different from video games, because they are so easily prototyped (just take some cardboard), and they play quickly so you can easily play the entire game several times a day. And at least for euro style games, which often have few mechanics interacting in complex ways, testing seems like the only viable strategy for me to see if they work. Take something as simple as Pandemic - it would probably be a nightmare to model this mathematically.

    10. CNA has come up before, but I always like being reminded of it. Irene and I have talked seriously about designating a room in our house the "Gloomhaven Room" because we want to leave the board set up between sessions, but it takes over our dining room table and the cats jump up and knock the pieces everywhere. I can imagine that a game with a nine-foot map would similarly require a dedicated playing space.

    11. You have heard of CNA, but have you tried to open the rule at a random page and see what's there ? It's fun, and with 100+ pages of rules you are unlikely to land twice on the same page.

      You can find the rules there :

      Last time I tried, I landed on the rule for transfering supplies from one disabled truck to a truck in working condition. I just tried again, and I got the cost of supplies of PoWs. Another column on the same page reminds you that before using water (52.4) you need to check the rules in 49.3 for water evaporation and spillage, and of course the rules in 29.3 about the impact of weather, weather which (I checked) also impacts "water evaporation".
      I find it fun, but maybe I am different.

      About board and pests, I tried to play the "computer assisted" wargame Fighter Command by SSI (1983), but my then 18 months old daughter ate the 141st Fighter Squadron (or to be exact its counter), so coverage of the game has been postponed.

    12. Apologies. Here :


    13. I'm kind of surprised boardgames receive so little testing, because I remember reading about how prospective board game developers should playtes the hell out of their games and check other games on the market, so they don't duplicate someone else's by mistake. Guess that's a case of do as I say, not as I do.

    14. From what I've heard studying board game design, admittedly from afar, most board games go thru rigorous play tests. Dungeon Crawlers and other RPG-lite board games probably not as much because of how many more variables are involved. I'm going to guess that Harland is referring to the glut of Kickstart or crowdfunded games that are largely only playtested by a limited and budget-crunched staff.

    15. Things might also work differently in the wargaming scene. Although looking at CNA I don't think lack of playtesting is the problem here. I'd consider starting an actual war before playing this.

      Reading rules can be fun, though.

  3. I guess Dungy and you found the precursor to Desktop Dungeon, proving once again that for any [turn-based] contemporary game with a unique design, someone already thought about it and did it on eiher mainframe or TRS-80..

  4. I believe a common way is to make a "construction set" of abilities for monsters and players, assigning some kind of cost to each, along with more linear parameters like HP (or level alone), and then create encounters with "model player party" in mind. E.g. "a normal, non-gimped party of 4 characters all at level 8 have a rating of 1000, so a normal encounter for it should have monsters with total rating of 1000, too, and a hard encounter should have a rating of 1200 (e.g. a boss encounter, or optional hard side-quest)".

    Since number of opponents is often very important, it usually scales the rating of encounter in addition to base monster rating. Such system described in detail in DM guide for Dungeons and Dragons, and there are a lot of "encounter calculators" online that let you input levels of players' characters and HD (Hit Dice) of monsters.

    And then you do a lot of play-testing, because nothing beats the real thing (in tabletop game, the master can cheat e.g. by not having monsters use all abilities if he sees players are not having fun, but in a CRPG playtesting is the only way for now, though it would be interesting to try to make a DM AI at least for combat - like Director in Left 4 Dead).

    Of course, combinations of features and abilities often lead to unpredictable effects on difficulty, which is why CRPGs with more-or-less complex progression are almost impossible to rid of "broken" classes (or combinations of classes in D&D) which become too powerful at some point. But "Build Porn" is its own draw for some people, so it's not even much of a flaw, if a dedicated player with a spreadsheet can completely break your single-player game (MMOs are a different thing) - you just have to make sure that there is no One Simple Trick to win all encounters in your game that a casual gamer can discover without much thought (admittedly, a lot of developers fail even at that - see for example a completely broken infiltration step in Phantom Doctrine, which allows you to take out 90% of enemies in every mission even on Hard difficulty without much effort).

    1. This is exactly what I was going for. I would love to see the actual statistics that an author put together for a particular game.

  5. Nice to see this blog returning to the usual fast-paced schedule – I hope that your health, timetable and equipment are all back to a good state!

    A small question that I maybe overlooked: has the “Index of Games Played by Series” been dropped? I will miss it since I used it to choose articles when I wanted to go back to old series, but I can fully understand if you want to reduce the amount of upkeep this blog needs.

    1. Something got messed up with the hyperlinks in that index, so I took it offline until I could get it fixed, which is probably going to be a long, manual process.

  6. "Do you think anyone has ever made a chart with 'orc hit points' in the title before?"

    Yeah, flip open any Bestiary for AD&D or Pathfinder and you'll find plenty of those charts. The question about balancing extraordinarily complex gaming systems is very interesting, but boils down to 'let the players win' in the end, I believe.

    All those perks and feats are basically there to make the players look cool or heroic, and have very little impact on the overall balance. A tough encounter, which is barely winnable, should be the exception rather than the rule (in tabletop rpg's, I should add).

  7. Even if they aren't proper CRPGs as you note, I still love your review and analysis of these small games. I will be sad when the pre-1990 era is mopped up!

    1. Little risk of that happening. There are dozens of early to mid 80s text-based RPGs not currently on Mobygames. I discover them all the time. Most recently: https://www.mobygames.com/game/199558/cells-serpents/

    2. Yes, these little obscure ones are really the best. Especially when they're played according to the designer's intent and real effort is put into them. Usually they just get spray-painted with "LOLZ dumb" and thrown out of the car window into the ditch.

  8. Isaac does in fact have a mathematical model for purposes of balancing *haven scenarios.
    I don't know how he arrived at it; I would guess that it's the result of a lot of playtesting.

    I think he shares it with guest designers. Fun fact: in the bigger *haven boxes, a good amount of the side scenarios are created by non-Cephalophair people. They get edited and adjusted, of course.

    There's even been some critique of the model, and you allude to it yourself. It doesn't handle power creep quite correctly. Your experience of needing to turn up the difficulty in later game isn't uncommon.

  9. I would guess that the graph of orc hit points versus enemies remaining is actually a linear function multiplied by a random factor.

  10. Your win rate is significantly higher than mine in Gloomhaven. For about 85 completed scenarios I am at just over 50% win rate (it would be significantly lower if you factor in the solo scenario attempts).

    1. Irene and I cheat a bit, which is why we play on "very hard." We each play two characters and we don't really make any effort not to collude. The instructions say that you're only supposed to tell other players what you're going to do in general terms, but we have more fun making tactical plans.

  11. Ah, Gloomhaven. I was also a bit addicted, having played through the campaign once "for real" and twice in the digital version.

    No matter party composition or difficulty, we also had an astounding number of games where we JUST were able to squeak out a win with some last-minute actions. Far more often than cruising to a victory, or being beaten down.

    However, power creep is a real issue. It's not just learning to play better, but I can nowadays beat almost any scenario on +3 difficulty with characters above level 5, but would have trouble on +2 difficulty for the first scenarios. Items, Enhancements and overall card synergy does a lot here.

    Still an amazingly good game, though.

    Another difference to computer games is house rules. Game is unbalanced? Share some gold in the party, or whatever makes it more fun! Impossible to do in computer games (before mods).

    1. My "loss rate tolerance" is also much lower for a game where I have to get 4 people who don't live in the same house together to play.
      If we play every 2 weeks, an 85% win rate is pushing it. Took us a few years to finish Gloomhaven, didn't really want more losses to add to that.

      (Very different in solo digital where I am willing to try harder things and try again after a loss)

  12. For future reference, re: emulator with savestates, trs80gp now has save states (as of a few months ago)

    1. Thanks. I'll try that on my next TRS-80 game.

  13. AlphabeticalAnonymousJune 6, 2023 at 11:17 AM

    I've been playing a lot of 'Brogue' recently, and the Level 5 map you posted seems vaguely reminiscent of Rogue dungeons. Does Level 5 (or any of the levels, really) use the same map every time, or does its layout vary on each playthrough?

  14. "I was "confused" because I thought the game was asking for my class and the instructions hadn't said anything about that."

    I chuckled. Somehow reminds me of one of my favorite Simpson jokes:

    "Billy Corgan, Smashing Pumpkins.
    (shaking hands)
    Homer Simpson, smiling politely."

    1. "Please give your name and position."
      "My name is Ted Stryker, I'm sitting down, facing forward, but that's not important right now."

  15. Optimal strategy may also depend on what you're trying to optimize: Probability of completing the game? Expected score at game completion? Probability of achieving score above some threshold at game completion? Expected amount of time taken to achieve score above some threshold?

    1. Getting a score in that top bracket. I don't really see a way that it could be done. You've got to lower your odds to at least 90%, probably more, to earn enough hit points to survive the tougher combats. But you also have to win at least c. 75 out of 90 battles, and you probably can't lose 2 in a row. No matter what, I think you're looking at odds of significantly less than 1%.

    2. Would restarting the game again and again, always trying to bet 1HP against the first monster(s) help? Could enough restarts give you a good starting position, if you just tried often enough?

    3. So I disassembled the code and the monster HP is:
      1.8 * (monster "number" minus a random int between 0 and 6) * (number of monsters killed + 1) + 1.
      If you check this for monster number 8 (orc), you can see it agrees with your empirical study. The 2hp is hard to explain, but I think it's probably due to numbers getting rounded down during intermediate steps in the calculation.

      When you defeat a monster, you gain twice its HP to your HP, and you gain 1000*(probability of loss) to your score. Each treasure is worth (treasure "number" * floor level * 10); so sweeping the first floor should give you 910 to start.

      Since converting score to HP loses efficiency each time you do it, you want your first conversion to have as high a score as possible. I wonder if optimal strategy is to wager 1.5x HP against weak monsters and 2x HP against strong monsters, since your score increase doesn't depend on monster strength. The risk of substantial HP loss isn't paid back by improved rewards; you get the HP bonus either way.

    4. So my idea to run 2x is bad, because there's no net gain of HP. It's surprisingly viable, because you can boost HP by converting score, but it's not great.

      I've been running some simulations, and it looks like optimal may just be 1.5x all the time. About 8% of the time, it takes off and you get up to 27k.

    5. I wouldn't have thought the odds would be that favorable. Thanks for verifying the mathematics in the code!

  16. Have to wonder how reliable a Wise Council can be if its leadership is comprised of the most daring gamblers. Feels like any advice they give you will be a coin flip whether it actually works out or not.

    1. AlphabeticalAnonymousJune 6, 2023 at 2:15 PM

      > the most daring gamblers

      They're just a council of consummate Monte-Carlo modelers... which perhaps amounts to the same thing, in the end.

  17. > [balancing spreadsheets] do you imagine other developers do when writing CRPGs? Or do they just wing it?

    Just for curiosity, speaking for the stumbles in balancing that an inexperienced dev goes through, we did a spreadsheet and robust DPS calculations for progress in Aeon of Sands.

    When we started framing the problem, we knew that, as we designed the system, characters had the possibility to grow in "power" roughly to twice their starting overall capability in the game, from let's say an untrained to fit / experienced.

    We designed the weapons and defenses in kind of tiers, templating them with a number of variations, with an early range of stats that then go through a long iteration process.

    We also had a bot playing dumbly the game over and over, helping us to analyze the data we extrapolated in our playthrough.

    That way, we managed to keep a fairly sensible balancing for melee and (a bit less) range and magic combat.

    There were though two big obstacles that we did not manage to overcome though:
    - because the number of companions in the party was not predictable, the players could go through the game with one or three characters, making the game hard or easier depending on dialogue choices that did not clarify that at all;
    - because there's no organically conveyed reason to use an even balance of melee, range, and, most of all, magic, (that may or may not be available, depending on the path taken), the same party could output a wildly different DPS and absorb a wildly different amount of damage, between playthroughs.

    While that was a design choice, it was maybe a bad or at least a careless design choice, in relation to the game accessibility.

    In the end, we made available the difficulty setting at any point (making monsters slower or faster, with a worse or better party-hunting system, and with other tweaks, if I remember correctly, morale?), but still, very few players ever used that, so, many found it too easy, many found it too hard.

    1. Awesome feedback, thanks. This is exactly the type of story that I was looking for. I really need to explore the development process in more detail at some point. 13 years into this blog, I still remain extraordinarily ignorant as to how RPGs are made.

    2. Thanks to you Chester, your blog manages to remain a source of fun and interest through the same 13 years!

    3. An analogous situation I ran into when designing Hero's Quest was how to handle lockpicking from a Thief character. My first thought was to make some locks easy (90% chance to pick), and others hard (20% to pick). A classic case of, "It seemed like a good idea at the time." :-)

      In practice, most players would only try picking a lock once. If they succeeded, there might as well not have been a lock. If they failed, the door might as well have been a wall. :-) For the 10% of players who failed at the "easy" lock, it was actually an impossible one.

      Eventually, we got around this with two mechanisms. One was giving appropriate text feedback to the player - "You fail to pick the lock, but you should be able to get it with a little more practice." Or "The lock fails to open. This one seems incredibly difficult."

      The second mechanism was a hidden skill point development system. Each failed attempt resulted in the player gaining a variable number of "tick marks." When they reached the player's current skill level, the skill increased by 1-3 points, and we notified the player of the improvement.

      The number of tick marks was partly determined by the player's relevant prime stat - for lockpicking, that would have been a weighted average of Agility and Intelligence - and by how close the player's ability was to the difficulty check. A skill of 35% on a lock that required 40% gave a lot of marks, while trying an 80% lock at that skill level only gave a few.

      Players gained ticks even when they succeeded on a check - With 45% skill and a 40% difficulty lock, the player still "learned quite a bit." Above 30% differential, there were no more gains.

      We applied the same system to combat. Each attack or defensive move gave the player credit towards improving that combat skill. By fighting multiple combats against appropriate level enemies, eventually the player became powerful enough to take on more dangerous ones.

      We called this a "practice makes perfect" system. Rather than gaining experience levels as in D&D, the player improved at the skills they actually used. We also had a feedback system for the prime stats - Doing a lot of climbing increased that skill, but also eventually led to gaining points in Agility and Strength, the controlling stats for the Climbing skill.

    4. That was very smart! I'm curious, how did you establish the range for the character's progression?

      Did you have an idea from the start about what the skill of a character would compare at the end of its progression to its skill at the start, (like, "I want the character be X times more skilled"), or you let the progress inform itself based on the length of the game?

    5. (BTW, the first Quest For Glory was a big inspiration for me in developing, not the progress of the characters, but the idea of the progress in our game: I had to scrap actually a first version of AoS that was built around three totally different solo paths based on character type, because it was out of my ability to make it. Quest for Glory was just brilliant.)

    6. No discussion of Quest for Glory lockpicking skill development is complete without someone noting the optional avenue for unlimited practice at any location through the "pick nose" command.

  18. An interesting idea which has some potential to teach how probability works, but the typos and the faux-Elizabethan verbs really threw me off.

  19. If you write a game properly, having separate code for "business logic" and actual rendering/drawing, with good abstraction layers, it becomes rather easy to program a simulator with player and monsters, for testing even a million combat encounters if necessary. Sky is the limit. For example, one could imagine a command line utility when you give parameters that you want to test player level 1 with a set of equipment against all types of Frogs in your game (so all enemies with a type/tag "frog"). Then, because the rendering is separate code, the simulation of each combat takes <1ms, because the program does not need to wait until something is drawn, sword is swung, textures are loaded etc. If you have a game with many enemy types (and skills) this kind of simulator becomes a necessity.. unless you prefer players to become testers (as it is often the case) ;) Then you can get the stats from the simulator run and see if something needs to be balanced.

  20. Delta has spent years digging through the math behind the oldest versions of Dungeons and Dragons, running millions of simulated combats and posting plenty of charts and statistics of the Orcish Hit Points flavor. http://deltasdnd.blogspot.com

  21. Before the new edition came out in 2016, GW's "Blood Bowl" board game/RPG of fantasy races playing a hyperviolent version of American football went trough a golden age of having a rules council involving the creators and fans that would gently tweak the rules on yearly basis in what they called a "living rulebook" by the time the the project's plug was pulled it had lead to probably their finest tuned game ever.

    As an aside, Cyanide's PC games of Blood Bowl probably belong on your master list, though the earliest didn't come out 'till 2009. The older PC game from the 90s didn't have character progression.

    1. I played the old Blood Bowl computer game a fair bit. I remember the Dwarf team being so overpowered that I killed a complete Goblin team during one match!

      Lots of fun, but that one was definitely not a CRPG.

      Ken Brubaker

    2. As fun as it is, Blood Bowl (including the Cyanide editions) is not in any way a RPG according to Chet's definition : there is no equipment, nor "character creation" to speak-off. Sadly, there is no "cSportsGameAddict", it would be a fun entry for such a blogger !

    3. An advantage of PvP games is that they self-balance via the 'metagame'.

      In short, if strategy X proves dominant, enough players will adopt the counter strategy Y, such that strategy X's win rate is pulled back towards the mean.

      Because humans are imperfect players who have preferences beyond winning, a few corrective nudges still need to be applied now and then. Games like Blood Bowl (previously), Hearthstone, League of Legends all analyse match data when determining what nudges to make.

      PvE games like Gloomhaven and CRPGs are harder to balance in some ways, but balance doesn't matter in quite the same way. A person playing by themselves might self-balance if they find a strategy overpowered, or might simply enjoy using an overpowered strategy - imbalance is a significant source of fun! On the other hand, imbalance in PvP games tends to result in player fury.

  22. Most (possibly all) professional CRPG developers use spreadsheets and other tools to run simulations. For Hero's Quest (later Quest for Glory), I created a simple command-line program (using C) that let me try out combinations of key stats to determine the outcome of a typical battle.

    That was based on an influential article in early issues of White Dwarf magazine. The author (apparently Don Turnbull) proposed a quantitative analysis system for tabletop RPGs. The "MonsterMark" was the ratio of the number of turns the player (or party) would take to defeat a given monster to the number of turns the monster would take to defeat them. The higher the resulting number, the more powerful the monster.

    On a computer, this could either be a strict probability calculation, or the result of a Monte Carlo simulation - Run the battle multiple times to get a range of results. Let's say on average the player could defeat an Orc in 3 turns, while it would take 7 turns to defeat the player. That gives it a power rating of 3/7. If it takes 20 turns to beat a Dragon, and it typically obliterates the party in 4 turns, it has a power rating of 5.

    My rule of thumb was that, if the player was in an appropriate area of the wilderness, they should be able to defeat a monster while losing 30-40% of the player's maximum health. Two battles in a row would be fine, but fighting three battles without healing would be dangerous, and possibly fatal.

    As for "appropriate areas," Lori and I designed Spielburg Valley like an archery target (scored in reverse). The closest areas to town were the safest. Farther out from town was more dangerous. In addition, monsters typically encountered during daytime were less dangerous than nighttime encounters. Early in the game, players would be more successful if they explored areas near town during the daytime. As they became more powerful, they could venture farther.

    I'd like to say that I thoroughly tested the system over hundreds, or thousands, of hours. But it really came down to a small amount of testing, and lots of educated guesses as to what would balance well. In hindsight, I consider it a minor miracle that the game balance turned out well. I don't recall whether we got much feedback from the testers about combat balance, but that certainly would have been helpful!

    From talks with other RPG designers, I'd say that most of them used similar calculations and/or simulators to balance their combat.

    1. "From talks with other RPG designers, I'd say that most of them used similar calculations and/or simulators to balance their combat."

      Are you referring to current designers, or those from the eighties as well? I got the impression that at least back in the eighties and early nineties, many developers designed their games with an informal, nonsystematic method. Sometimes quite literally winging it, and sometimes with merely some basics written down (on paper) and otherwise relying on intuition and playtesting (and often not enough of the latter).

      I'd also have speculated that the approach to test outcomes with thorough tables of data and formulas, before they're programmed into the game for real, were more often the work of designers who didn't code, especially designers coming from tabletop games - for example Arnold Hendrick, designer of Darklands. But back then a lot of games were designed by programmers, and I'd speculate that they often had neither the inclination nor the time to go this route.

      But I'm just sharing my assumptions here, I don't have any first-hand knowledge.

    2. Well, this is all fascinating. I'm so glad that Adventure Dungeon led me to think about the subject.

      It's funny that Monte Carlo simulations have come up in this thread. I was just trying to figure out how to program one to run on some social science data. Somehow it didn't occur to me that they'd be used, probably extensively, in video games.

  23. Despite being an old English major, who sometimes uses superfluous big words, I had to look up "sobriquet".

    1. That comes from my days of researching Arthurian literature. In normal speak--to the extent it's ever used in normal speak--sobriquet just means "nickname." But Arthurian scholars have always used the term for the little tag that follows a lot of knights' names: Gringalas "the Strong," Mador "the Black," Ozanna "le coeur hardi," and so forth. I wish I knew a better term for this specific type of tag. "Suffix" doesn't quite do it.

    2. It seems that the most specific and accurate word for this kind of descriptive individual name is "agnomen". Other possible synonyms are cognomen (in its current English meaning, not the original Latin one), epithet, and byname.

  24. I actually set up an Excel sheet to start modeling different strategies and likely outcomes.

    Never change, man. Never change.

  25. I heard back from Dave Lo a few days after I published this and added an addendum to the entry with some additional information.

    1. The game Monster Combat Dave Lo mentions as one of two influences has been covered by El Explorador de RPG .

      It was brought to different platforms and of course someone has recently made a "new and improved" version of it on github (Monster Combat 2).

  26. Holy crap! I almost spit my drink out when I read David Lo's comment referencing the type-in BASIC listing game Monster Combat from the book Big Computer Games by David Ahl.

    Thought I was the only one who recognized the unique game playing characteristics there. I even attempted a recreation of Monster Combat using the modern Twine interactive fiction tool.


I welcome all comments about the material in this blog, and I generally do not censor them. However, please follow these rules:

1. Do not link to any commercial entities, including Kickstarter campaigns, unless they're directly relevant to the material in the associated blog posting. (For instance, that GOG is selling the particular game I'm playing is relevant; that Steam is having a sale this week on other games is not.) This also includes user names that link to advertising.

2. Please avoid profanity and vulgar language. I don't want my blog flagged by too many filters. I will delete comments containing profanity on a case-by-case basis.

3. NO ANONYMOUS COMMENTS. It makes it impossible to tell who's who in a thread. If you don't want to log in to Google to comment, either a) choose the "Name/URL" option, pick a name for yourself, and just leave the URL blank, or b) sign your anonymous comment with a preferred user name in the text of the comment itself.

4. I appreciate if you use ROT13 for explicit spoilers for the current game and upcoming games. Please at least mention "ROT13" in the comment so we don't get a lot of replies saying "what is that gibberish?"

5. Comments on my blog are not a place for slurs against any race, sex, sexual orientation, nationality, religion, or mental or physical disability. I will delete these on a case-by-case basis depending on my interpretation of what constitutes a "slur."

Blogger has a way of "eating" comments, so I highly recommend that you copy your words to the clipboard before submitting, just in case.

I read all comments, no matter how old the entry. So do many of my subscribers. Reader comments on "old" games continue to supplement our understanding of them. As such, all comment threads on this blog are live and active unless I specifically turn them off. There is no such thing as "necro-posting" on this blog, and thus no need to use that term.

I will delete any comments that simply point out typos. If you want to use the commenting system to alert me to them, great, I appreciate it, but there's no reason to leave such comments preserved for posterity.

I'm sorry for any difficulty commenting. I turn moderation on and off and "word verification" on and off frequently depending on the volume of spam I'm receiving. I only use either when spam gets out of control, so I appreciate your patience with both moderation tools.