The problem with using citizens outside of cities is that they are not a set unit of a quantity of something. A citizen can cost 10 Food or 100, so not all citizens are equal. It would make more sense for units to just cost Food. A big city might be able to produce four units without losing a citizen while a small city might lose a citizen right away when producing just a single unit. But even that might still not scale well, making the cost too prohibitive early and too neglible late. Also, there's the issue of maintenance. Do they still consume 2 Food per turn? If so, from where? Is there a global stockpile of Food for your military, maybe that's what you build Granaries for?
[...]
If military units are produced with food, that could go toward having the system of multiproduction queues, because here, there is then a strategic cost to making military units again even though you can also build infrastructure at the same time.
Expanding on this subject of sourcing and maintaining military units. . .
Just a few comments on a very good statement of the Problem:
I would argue that the "problem with using citizens outside of cities" in Civ at least, is that there is no in-game measurement of citizens or population outside of cities. The only population measure is the number of Pops in each city, and the Civ total is the total of all the cities. All population 'working' tiles outside of the city have always (as far as I am aware) been related directly to the pop of the city: want to work another tile, either move someone or add someone to the city.
I've posted on this long ago, but in a game spanning 6000 years there simply cannot be a fixed ratio of population to unit. The largest organized permanent or semi-permanent (between wars) unit in ancient Egypt and Mesopotamia seems to have been between 600 and 1200 men, and anything larger was indicated by words that all translate as "host" or Whoever Shows Up. By the Industrial Era, 1200 men is barely a Regiment of infantry, and European powers like Britain, France, Prussia, Russia and Austria in the Napoleonic Wars fielded 100s of regiments each. To avoid trying to manipulate units by the hundreds in the mid-game, the unit to population ratio has to be on a sliding scale - and that suggests that the population Point to City population measure also has to be a sliding scale to accommodate the earliest cities at >20,000 and 20th century cities at <10,000,000.
At least in the mid-20th century, the US Army's Maneuver Control manual of the mid-1950s assumed that a unit that lost 10% of its personnel within a few minutes was ineffective for the rest of the day, but that was for units of platoon or company size - less than 100 men. During WWII the German and Soviet armies both kept divisions in the line and fighting when they had lost up to 75% of their combat strength, but the Germans, at least, didn't expect those units to mount a sustained attack successfully. How much and what kind of losses a unit can sustain and continue to function is, frankly, more a matter of the morale and leadership - 'soft' factors - than numbers and equipment, which makes it difficult to put hard numbers to them.
To your summary on gameplay and abstractions required I would simply add that whether or not some kind of citizen/population figure is used, that cannot be the sole criteria. Societies (and in game terms, that means Religions and their tenets, Governments, Civics, Social Policies) may have more effect than 'raw' population numbers on military capacity, no matter how the population numbers are divided into tile, city, or Specialist workers.
I am interested in iterating the "sliding scale" factor of population ratio to unit size. Once again, I'm going to point to Caveman2Cosmos. However, in C2C, this is only a system that is literally just there, it isn't a given that the gameplay is balanced around it at all, but C2C attaches a "size" measure to each unit on the field. It essentially describes the number of warriors actually present in the unit, with a logarithmic scale that has "sole individual" at the bottom, "30-50" in the middle somewhere, and goes up to thousands, and even millions. One idea is that combat outcomes can depend on the numbers of soldiers that are brought to an engagement. (Also, it even factors into some of the noncombatant abilities that C2C units have, like detecting criminals and spies, which I mention just to brainstorm.)
It is, I assume, a given that one thing you mean when you bring up the idea of a player's units having a certain count-of-heads magnitude, is to factor that into the combat result formula, but I think I want to extract from you more about the unit production side of things, and also the progression through eras and technology. You reference the idea that regiments of soldiers have become greater and greater by orders of magnitude in our own history, and I believe you're implying that this is a secondary reason why a less developed army would be utterly ineffective put against an advanced one, (secondary to the difference in the weapons themselves). It makes me think of having a dynamic where a player in the game can have an army, but they do not multiply their numbers and so you run into the situation of representing on the map a totally irrelevant and ineffectual cadre of warriors once you get into the upper turns.
Like, I think of it as strange how, in Civ V and Civ VI, you can even have an archer stand in front of a tank. It feels as though the standards for what can even count as a military asset would evolve globally and automatically obsolete the archer, disbanding it. Put another way, I mean, as a Human being standing there in the mud with a bow, how is it possible that they would even charge the tank, I do not care how well drilled you are. So, while thinking of these reasons, I am curious to inject into the game this headcount statistic for the units we build, which the game's combat rules and possibly other systems on the maintenance side, will consider and calculate.
Maybe if we bring back the Civ2 army support thing (also successfully used in Alpha Centauri) where soldiers are supported with hammers, in combination with a units as population limiter, this can provide a counter effect to the snowballing issue. (Here, "the snowballing issue" is the problem for fun created by the numbers of your output becoming blindingly large and everything becoming too plentiful too fast to even take seriously - in a 4X game where you don't have any increasing costs or diminishing returns.)