It's not really predictable; you'd need a camera watching it to continuously train and compensate and improve. Which would be neat, honestly.
With the lever-arms on these things, you're going to get into all sorts of details with things like the pitch-circle variation of the plastic gears, which may have been nearly-ideal when they were produced but will become uneven with wear.
I would go the other way and grab any broken 3d printer, rip off the hot-end and stick a pen in its place. It's slower by default but a thousand times more precise, and you can tune up the speed now that you're no longer whipping around a heavy extruder or waiting for plastic to flow.
Hysteresis is, in general, predictable, though sometimes compensating for it is impossible. Are you speaking from experience in measuring SG90 hysteresis and gear wear or are you theorizing?
(I'm theorizing, because I think theorizing is valuable, but I want to be careful to distinguish theoretical predictions from reports from experience.)
The broken 3-D printer idea is brilliant. Unfortunately, right now there's only one broken 3-D printer for sale near me on MercadoLibre, and it's an SLA printer, so it only has one degree of freedom of positioning. However, it won't be a thousand times more precise; typically RepRap FDM printers like the Prusa or Creality have positioning errors on the order of 0.1 mm, and a thousand times less precise than that would be an error of 100 mm. The BrachioGraph drawings seem to have an error of about 2 mm, only about 20 times worse than a bog-standard RepRap, and actually, 20 is less than 1000.
Cost is relative; as I noted on here the other day, the minimum legal wage here is US$207 per month https://news.ycombinator.com/item?id=31830007, so even these three toy servos cost a day's wages, perhaps the equivalent of US$1000 for you. And lots of people here don't make even that much, and most people live in poorer countries than Argentina.
The fact that a better product is available at a higher cost is not relevant to the question I was asking, which is what the particular problem is with the cheap servos.
None of those seem like they should be a problem for BrachioGraph except the too-vague "crappy"? It doesn't have the servos trying to lift weight (except for the lifter servo, which has to lift the pen and popsicle sticks) or fighting resistance that can strip their little plastic gears. It mounts them with hot glue, which seems to work okay (surely it introduces uncontrolled compliance that you could reduce by gusseting them with epoxy) and doesn't require them to be quiet.
I'm trying to understand what the problem is with cheap toy servos that leads to problems for BrachioGraph, in enough detail that I could make plans to mitigate the problem.
With the lever-arms on these things, you're going to get into all sorts of details with things like the pitch-circle variation of the plastic gears, which may have been nearly-ideal when they were produced but will become uneven with wear.
I would go the other way and grab any broken 3d printer, rip off the hot-end and stick a pen in its place. It's slower by default but a thousand times more precise, and you can tune up the speed now that you're no longer whipping around a heavy extruder or waiting for plastic to flow.