What is not settled, said plainly. None of this has been filled in by guessing.
At 0x57EC there are four bytes — 0x28, 0x50, 0xA0 and 0xF0 — which are the boundaries between the canvas's five depth bands. 0x55B6 indexes them with the band number minus one; but the band goes as high as 5, so index 4 runs off the end of the table and reads the 0x3A that is already the instruction at 0x57F0.
It may be harmless — 0xE4C0 + 0x3A still lands inside the canvas — or it may be a bug in the cartridge. It has not been checked in the emulator, so it stays here as what it is: a measured suspicion, not a finding.
When both players are on the green and the balls land within three pixels of each other, 0x551F does this:
push de ; DE arrives holding 0xE1C0, the tee marker
ld de,07b30h ; sprite 12's attribute
call vuelca_cuatro_bytes ; 0xE0B0, the ball -> sprite 12
pop hl
call vuelca_cuatro_bytes ; 0xE1C0, the marker -> sprite 12 AGAINvuelca_cuatro_bytes does not touch DE: it only advances HL. So the second call reopens VRAM at 0x3B30 and writes over what the first one just put there. The ball is lost under the marker, and the first transfer is wasted work. One more inc de — or a 0x3B34 — and they would be two different sprites.
It may be an oversight in the cartridge, or it may not matter on screen, since the two attributes are within three pixels and whichever sprite survives covers the other anyway. It has not been checked in the emulator: it stays here as what it is, a measured suspicion about the code.
The six bytes at 0x6E70 are valid code — ld a,c / neg / ld c,a / jr — twins of the ones at 0x6E6A that do the same to B. The three jrs around them all point past. No path has been found that reaches them.
The gameplay screen is missing from the pictures on this site: the view of the hole from behind the ball. The pieces are there — the 20 by 16 canvas at 0xE4C0, the 54 cell codes and the blocks they are stacked from at five distances, all drawn in bloques.png. What is missing is the chain that puts the ball on the tee — 0x4850, then donde_esta_la_bola (0x5A2A) and elige_la_rejilla (0x5AA5) with the strip at 0x5B1D — to know which of the four grids is being looked at and from which cell. We stopped rather than invent it.
All 256 tiles and the 54 cell codes are now drawn from the ROM, and can be looked at in tiles.png and bloques.png. What is still undone is smaller: naming each one — which is the tree, which the bunker's edge — across the tiles_girables_1..6 and colores_del_campo_1..10 stretches of the listing. They are placed and they decode cleanly; what they lack is a label of their own.