There is not one screenshot here. Every picture is drawn by repeating what the cartridge does: video memory is rebuilt with its own copies and the characters and the map are put back together with its own tables. Everything else —the listing and the numbers— comes from the binary and is reproducible with make.
SCREEN 2 has no scroll register, so Pippols builds one: VRAM holds 192 characters (0x40 to 0xFF) that are eight copies of the same set of 24, each one a pixel lower. Scrolling the screen by a pixel means adding 24 to the character number and writing the whole name table again, 24 rows by 22 columns, every frame; not one pattern is touched.
Of the twelve characters in each half, four are seams (0x56ED): the end of one strip on top and the beginning of another underneath. Without them the drawing would split halfway down as soon as the offset was not zero.
A stage is thirteen stretches of 24 rows; a stretch, six pieces of four columns; and the 44 pieces are built in RAM by gluing groups of three rows from a dictionary of 25 (0x54C0). The index list —eight per piece— is 328 bytes, and that describes the whole road.
The loop at 0x5496, though, reads 352 indices: the last 24 reads land inside the code at 0x5653, and out of that come three junk pieces. It does not matter, because no stretch points at them. Nor at pieces 0 and 1, which are properly built.
In openMSX, on a Philips VG-8020 (PAL, 71,591 cycles per frame) over 1112 frames of the demo: repainting the name table takes 32,557 cycles on average, 45.5 % of the frame, and the whole interrupt 52,533, or 73.4 %. That is 732.7 writes to the VDP port per frame.
Which is why the main program does nothing: it sits in a two-byte jr $ at 0x4074 and the entire game runs inside the interrupt, one step per frame.
The score is three BCD bytes, but the panel paints the six digits and then a fixed zero behind them. What you see is ten times the counter, which is why every score ends in zero. The first extra life comes at 20,000 and then every 60,000.
At 0x717D there is a jr nz,$+4 that ends up at 0x7181, and 0x7181 is not the start of anything: it is the displacement byte of the jr z on the line above. It executes as dec b, which does not matter there because the creature loop keeps BC on the stack, and carries straight on. It is the only place in the cartridge where an instruction is entered halfway through.
The 18 world screens share the background drawings: there are eight stages but only five sets of strips, and what changes from one screen to another is a swap of three pairs of inks (0x577D) over the 96 colour bytes. The same flower comes out red or pink, the same tree green or blue.
And the route is not a line: the table at 0x7B62 gives two destinations per screen, and one or the other is taken depending on whether you are going up or down.
In the table at 0x785A each object carries its (pattern, colour) pair. Object 8 —the boot, which makes you walk one and a half times faster— and object 9 —the shield— carry the same pattern, 0xC0, and differ only in colour. Objects 1 and 4 do the same with pattern 0xA4.
Each of these pictures is the cartridge replayed outside it: the background strips are decompressed the way the ROM decompresses them, the 192 characters are generated with the same loop and the map is assembled from the same tables. Under each caption is the address of the routine it comes from.
