The Tavern

I strongly disagree with that sentence, especially when level-wide vaults come to mind.

A player seriously benefits from knowing a layout of a level and number/type of monsters. I personally have done hell endings / unique pans multiple times with appropriate des files open, because it allowed me to:
- choose the fastest, safest way
- map a level without spending a single scroll
- determine how many monsters and which ones are behind a door/corner
- preemptively buff myself
- skip "uninteresting" parts of level by teleporting (particuarly in case of long, linear levels)
- assess whether a spot after teleporting is good or not

Besides of spoileriness, big vaults suffer from repetitiveness, which concerns new players to a lesser extent, but eventually causes these vaults to be disappointing and reduces replayability value.

Some vaults do quite good job at randomizing key elements (like boss / rune location), some are terrible (fixed spots, fixed monsters), but I should mention that more recent vaults seem to be better.

I think that Crawl would benefit in a long term from replacing large vaults with (several) smaller ones with preferably more randomized monster sets and locations. The connections between small vaults can be either generated procedurally or use the same idea (idea, not layout) as v:5. (I do not think that we should keep legacy vaults if they cannot be adapted).

The advantage of smaller vaults is that they are easier to hide, harder to recognize from e.g. outline and recognizing one of them does not provide information about the rest of level. Furthermore, I believe that by mixing random sets better replayability would be achieved.

Well, vault definitions do use a lot of SUBST, NSUBST, SHUFFLE etc. Whether that's a lot of randomisation or not, is a subjective assessment. There's no question that more randomisation is better. I also agree that encompassing (full level) vaults should be more rare; this is very easy to achieve. The SUBVAULT syntax was created for the V:$ randomisation, and is rather young. More vaults (especially big ones) could use it for good effect.

Patches/Pull Requests randomising existing vaults will certainly be reviewed favourably. Perhaps we need some guidelines, or just examples of what we consider "reasonably randomised", to make it easier (for vault making and modifying) to achieve some standard?

I would like to request that anyone adding randomized vaults to trunk ensures that if they can contain grossly depth-inappropriate challenges, they will always have runed doors.

I had a vault in my game yesterday that spawned two storm dragons on D:11, and the doors to the vault were open when I saw it. (It was minmay_shrinking_squares.)

This kind of surprise can be dealt with, sure, but if vaults are placing hideously OOD stuff at floorgen and then allowing it to begin wandering, it seems to me that the result discourages using autoexplore.

The last time I submitted vaults, the devs reduced the randomization in most of them. "Patches/Pull Requests randomising existing vaults will certainly be reviewed favourably" is horseshit, though I wouldn't want to discourage people from trying.

is this about that argument you had with gammafunk over the proper random distribution to use for monster placement?

since i'm not convinced that's critical to this discussion

Lowering the variance of monster count makesmore effective. It is relevant that new vaults are required to have less variance in monster count than a binomial distribution.

I asked this in the last thread before MarvinaPA swooped in to lock it, but why are the .des files public in like, not-source-code versions of crawl? It seems like if you just hid them away somewhere inside the guts of crawl after it compiled then this problem is isomorphic with any problem caused by crawl not giving perfect information.

Because the only things "hiding" them that way would accomplish are
1. making a slightly higher number of players download the source
2. making it a huge pain in the ass to make and edit vaults

Yeah, the more I think about it the more I kind of feel like purpose-built algorithmic vault generators might work better than actual human-authored vaults. Some of the vaults kind of are that already, I think.

those are called layouts

I tend to agree that more randomization in vaults would be a good thing, and I while I can't speak for the team, I would personally look favorably on attempts to improve the randomization within our current set of vaults. I think there's probably a lot we could do with new layout types and other ways to generate distinct types of random content. However, all of this can take a lot of work, and I certainly haven't done my share of the lifting on it.