Vault

ROTK2 — Sangokushi II (SNES)

The first KOEI SNES title reversed, and the title that proved the portable VM from both sides. Its NES twin rot3k2 was the oracle — a port comparison, not a cold read.

Links: system → SNES · toolchain → KOEI tools · NES twin → rot3k2 (NES) · theses → KOEI’s portable VM, The Rosetta Stone Method, DREAM — goto-free structuring

Identity

Field	Value
Logical name	rot3k2-snes-decompiler (public) · github.com/chrisaacson69/rot3k2-snes-decompiler
Sibling path	../rot3k2-snes-decompiler (resolved via .claude/local-paths.md)
Cart	ROTK2 (USA), LoROM, 1 MiB, 32 KiB battery SRAM, RESET $800C
Status	✅ re-walked & complete — ~1,316 bytecode routines named across 26 code modules, 4 data blobs documented
Depends on	koei-snes → snes-decompiler

⚠️ Not cloned on this machine (2026-07-21).

What it contributed

• The portable-VM proof. VM opcodes byte-for-byte identical to the NES twin (NES 219 / SNES 218), same resident-library + command-app architecture, province record stride 25 on both. The bytecode targets a virtual address space behind a stable syscall ABI — architecture-agnostic by design. The SNES is a faithful port + facelift (69 vs ~31 syscalls: DMA/Mode 7/SPC700); the lone data change is the officer record 21→34.
• A found-and-fixed walker bug. The Gemfire pass exposed an offset-0 defect — each routine’s 5-byte JSR $23E6 trampoline was being decoded as bytecode, mis-shifting and mis-keying names. The trampoline profile was backported and all 26 modules re-walked; damage ranged from severe (comroot/AI: 110 corrections) to none. Old files kept as *.offset0-backup.toml. A clean specimen of why the second target is also a check on the first.
• Coverage now includes the hex tactical engine and its 13 sub-overlays, never decompiled before.
• Two decompiler experiments came out of it: a C lift and a DREAM-style structurer (65–75% goto-less).

Tags

snes · 65816 · reverse-engineering · games