What would be the safest, not-cheesy way to level grind in a sand-box like setting?

[Vahnavoi]

Me, I only care about being able to do the math on food production vs food requirements, determining the level (etc) of the level-able population, and such concerns, not on properly assigning the "Female Half-Orc Apprentice Wizard" graphic to this particular data object. And, for these purposes, one data model is clearly more elegant than the other.

It's simpler, yes. It's also a classic case of Spherical Cows in a Vacuum - by lumping everyone without a class into same category to make a problem more tractable, you end up losing details that would be relevant to the end result.

Dwarf Fortress proper, as a game, puts level of detail way ahead of simplicity. Running a massive zero player strategy game on the background to generate a playable world is very processing intensive and clunky. Proposing a Dwarf Fortress- style simulation hence, very like, includes coding you'd deem inelegant.

-----

Given that the OP was talking about "dogs", I feel it's safe to assume that this simulation really ought to be set up such that it's expected that it will run for thousands of years before the player joins, to give time for Wolves to be Domesticated into Dogs. At which point, much of your commentary wrt early game expectations seem incongruous, like we're playing very different games. I'm trying to envision a T=0 world setup that will lead to multiple races still existing thousands of years later when Dogs walk the earth, and asking myself which races with which Societies have the best chance of existing at that point. Any setup that doesn't produce playable results after thousands of years of simulation, I'm saying the fault is in the setup, and therefore something needs to be changed.

It really isn't a safe assumption, for reasons already listed. To approach this from another angle: domesticating an animal is just a Handle Animal check in 3.5 D&D ruleset. The simulation can include both an initial population of canids and initial population of humanoids at T=0, and it will take only take as much time for domestic canids to appear as it takes for the two groups to meet and some humanoid succeeding a Handle Animal check. The problem with your argument is that you flip-flop between assumed realistic simulation of the domestication process and simulation of D&D rules, when the target of discussion is only the latter. Again, an alternative comparison can be made to Civilization 6: yes, dogs have to be domesticated, but this is a simple-to-achieve thing that pretty much every civilization in every game manages by turn 60 out 500. Even if nominally, the time from turn 1 to turn 60 runs from 4,000 BC to 1,000 BC, those are just numbers, with basically no relationship to the amount of actions taken by game characters.

So, sure, I'm expecting some churn, expecting some races to have died out when the simulation kicks out into active play at T=X thousand years later, with there maybe being dogs for the OP to train. But I'd say any initial assumptions, any settings for the initial setup and underlying systems (NPC leveling, population numbers, tech tree, whatever) that don't involve at least the majority of the initial races still being "in the game" as playable choices probably produces a suboptimal play experience, and should be reconsidered.

Which races exist as playable option to a player once the player enters the simulation, doesn't necessarily have anything to do with what has survived world generation. If the simulation is based on 3.5 D&D core rules, it's safe to assume the player will have character generation choices based on 3.5 D&D core rules. It does not follow every character described by the core rules will survive or be present in the world. That is, if the player chooses to be an elf, it is possible they are the ONLY remaining elf. Talking about "suboptimal play experiences" I consider to be completely absurd. It is ordinary for procedurally generated worlds to only contain subset of possible civilizations in them. This extremely obvious in strategy games both old and new. The appeal of that is that you can then use the same procedural generation rules to create and play in relevantly different worlds.

In Dwarf Fortress proper, the number of civilized races is much smaller (5) than the amount of civilizations world generation can handle (160 for Large worlds). Still, since doubles are possible, smaller worlds might not contain a civilization of every possible race (for example, a Smaller world with Low number of civilizations has only 6, so two doubles would mean one race is missing). 3.5 D&D has a significantly larger pool of races to choose from even if we only include core player picks and their common opponents, for example, humans, dwarves, halflings, gnomes, elves, orcs, goblins, kobolds, lizardfolk, trolls and giants. That's 11, which would necessitate a Smaller world with High civilization count, or a Small world with Medium civilization count, to have a civilization of every race even at the start of the game.

The more races you want to be present when play starts, the larger the world has to be, and play has to begin at an earlier era, meaning everyone will be less developed.

-----

I honestly believe that the easiest thing... not the easiest thing to code, but the easiest thing for the poor computer to run cycles on is (modified versions of) the DMG Population Tables.

It is both easier to code and run iterations on DMG population tables, since they were meant to be used by hand, by a human, on pen & paper. If I wanted to automate this, I could do this in Excell, in a day. Dwarf Fortress-style simulation, by contrast, is much more intricate, involving running aforementioned massive zero player strategy game for an extended period and logging the results. Dwarf Fortress-style simulation is borderline impossible to do by hand, or would take a prohibitively large amount of time, and on maximum settings it can be unreasonably slow even on a contemporary commercial computer.

-----

Oh, and XP are certainly a Gamist abstraction in most games. In 3e, however, where they are spent to cast spells and craft items, figure into payment for spellcasting services, and actively figure into characters' planning and strategies, they pretty much have to be an actual, existent thing in the 3e world, something that's truly part of the Simulation, rather than a mere Gamist abstraction. That's why I worded my statement about them the way I did, because 3e expectations all but mandate it being one of the exceptions to them being a Gamist abstraction.

That's all humbug. Experience points existing in a game world does not make them less abstract or less game-like. They don't simulate anything beyond themselves: a person, in the world, explaining experience points to another doesn't have a better choice than "they're a point score you get for doing things on this arbitrary list of things, which you can then expend on things on this other arbitrary list of things". Their existence in the world is a result of simulating game rules, and a person in the world would reasonably conclude "the world works like a game".

Or, to put it simply: nothing is expected from being game-like in a simulation, when what you are simulating is a game.