6144 bytes, a byte for each eight pixels of the play area, 32 to a line, the bottom line first: the room is drawn here (CALC_PIXEL_XY_AND_RENDER) and copied to the screen, all of it on the first turn in a room (SHOW_BUFFER) and only what changed after that (RENDER_DYNAMIC_OBJECTS). CALC_VIDBUF_ADDR finds a place in it. The buffer is cleared (CLEAR_SCRN_BUFFER) before anything is drawn in it.

What the tape loads here is not the game's: the block runs on to the top of memory, and the machine it was saved from had these bytes in it. Every one is overwritten before it is read (measured: the loaded game run in the simulator from ENTRY into its first room, noting the first access to each address). They are, in order: zeros; stack debris -- words that look like addresses in the ROM and the display; the Spectrum ROM's capital letters A to U; the first 1464 bytes of the Interface 1's ROM; and zeros.

The Interface 1 code starts 328 bytes in. Disassembled as if at address 0 it is that ROM's restart table: RST 0 at offset 0 (POP HL, then LD (IY+124),0 -- the Interface 1's FLAGS3 -- and JP $0700), RST 8 at 8, RST 16 at 16 (which stores HL into its SBRT routine among the system variables), EI and RET at $38, RETN at $66; and read from address 0 its absolute jumps and calls into itself land on the starts of its instructions, which they do not when it is read from 1 or 2. The system variables it uses beyond the Spectrum's own exist only with an Interface 1 attached, and the error reports at offset $02B8 are the Interface 1's own, each between two numbers counting from 0 to 24. The ROM itself was not compared byte for byte.