The listing and every number on this site come from the binary and are reproducible with make. Nothing here was measured by eye.
openmsx -machine Philips_VG_8020 -cart hypersports1.romAn MSX1 machine is enough; the cartridge maps into page 1 and drives everything from the frame interrupt.
To confirm a state or a timing, put a breakpoint and watch the RAM the notes name: 0xE000 is the main state, 0xE052 the event, 0xE054 the attempts left. The per-frame cost of a routine is the difference of machine_info time between a breakpoint at its entry and one at its ret, times 3579545 for Z80 cycles.
Output paths in any Tcl script must be Windows paths, never /tmp.
make vram runs both. Neither needs anyone to play: the attract mode plays all four events on its own from power-on, and it is deterministic —two independent runs give byte-identical dumps.
tools/omsx_escenas.tcl waits for each scene's recorded second and dumps the 16 KB of VRAM, plus the 384 bytes of the BIOS font that 0x43BF copies. That is where the five .vram files in src/vram come from: the tile map behind this site's images.tools/omsx_posturas.tcl puts a read watchpoint over the athlete's frame block and records, per event, which addresses are read and from which PC. That gives the strips each event uses and checks the format: 679,198 reads in 230 emulated seconds, with 11,463 passes through 0x0038 as a control.openMSX's screenshot lags behind the Tcl script, and does not match the VRAM read of the same instant. Checked twice, and with the machine paused too: asking for the trampoline scene produced the previous event's screen, and asking for the horizontal bar produced the start-up logo. VRAM, read synchronously with debug read_block, is from the instant asked for, and that is why this site's images are drawn from it and from the ROM, never from a capture.