looking at interrupts, and I think I found a bug.
they set handlers for various CPU errors, but they accidentally set 10 (COPROCESSOR ERROR) twice, instead of the 05 (BOUND check)/10 (COPRPOCESSOR) interrupts they save

someone copy-pasted and missed a bit

I finally found the two helper functions they use to get and set vectors!

all the 30 other places I've seen them set/get vectors, they do it manually, but hey, maybe they use the helpers too

could also be that this is a compiler-provided bit of code, which is left in because the runtime needs it, or they just didn't eliminate dead code

okay I've figured out there's a shared format they're using here. it chunks the file into chunks, which have a 16-bit ID (unique per file, but not globally), an offset, and 16-bit length

so like, midisnd.dat will have 12 entries, and the first 11 are 200-500 bytes each, and then the last is 3k.
presumably it's each song and then some config info?

cities.dat is very interesting. There's 30 cities in total, but 491 entries in it!

So they must be doing something odd there, that doesn't divide equally. Maybe one city-chunk gives IDs of the others?

idea for a test: it's easy to spot which chunk in a city is the image, because it's the biggest. Here's a way to determine if it's looking up by IDs or offsets/indices: swap the IDs of two images

darn. turns out you can't just renumber the chunks, because they have to be in increasing order.

so maybe I just need to leave the chunk indexes as is, and instead of moving the entries around, I move where they're pointing?

Bingo! I'm in Athens, but I'm seeing the image for Baghdad, and apparently with the Baghdad palette?

So one of these other chunks must be the palette for a city. Or it selects from a selection of palettes? Maybe they've just got a couple defined.

okay I figured out the cities.dat IDs:

They're all 1XXYY (in decimal):
XX is the city number (0-29), YY is the sub-chunk-id.

So like:
YY=0: City name
YY=2: City image.

They go between 00 and 22, and not all numbers need to be present.

hmm, reading a buffer and then summing all the values of the bytes in it.

suspicious behavior.

okay I think it has a very simple 1-byte CRC check on the chunks, which are optionally not run.
I can't make the math work but I'm reasonably sure that's what it is

okay they're using a blit that's UI-aware, so it starts the coordinate system at (1,13). Fun!

looking into the blitting code I managed to steal the world map out of RAM

ugh. TODO for my eventual Good DOS Debugger:
Instant Video display.
I don't know exactly how DOSBox-X is doing it, but while single-stepping the debugger, the display never updates. I can dump the ram at A000:0000 and see what updated, but not on the screen in DOSBox

found a suspicious array, which goes:
[
(-1,0),
(-1,1),
(0,1),
(1,1),
(1,0),
(1,-1),
(0, -1),
(-1,-1),
(0,0)
]

POP QUIZ: why does the font renderer need this array? how are they being "lazy" with this array?

there's also this code in the for-loop that steps through this array:

if index==8:
color=white
else:
color=black

@dividuum got it:

they're drawing the font 9 times, offset in each of the 8 directions, and in black. then they draw it in white, with no offset.

It's a pixel-outliner! By drawing their pixel font offset in each direction, they get a black outline on their font.

The Answer to the DRM questions for Where in the world is Carmen Sandiego? Enhanced (DOS, 1990) are, in no particular order:

23
Kent
dragon
calcium
1796
Warren
revenue
1792
Willard
1937
Crater
Tanzania
Hartford
Duluth
London
Gem
Silent
squeaker

if ((0x80 >> ((byte)local_4 & 7) &
(int)(char)*(byte *)((int)((int *)param_1 + 1) + (local_4 >> 3))) != 0) {

COULD YOU USE SOME MORE CASTS MAYBE?