# FORTH-79 Standard-Dictionary Cross-ISA Exerciser

`std79-exerciser.fth` is a 24-case, FORTH-79-standard-words-only program (no kernel-specific
extensions) covering `/MOD`, negate/abs boundary values, `RSHIFT`/`LSHIFT` on negatives,
`M*`/`M/MOD` mixed precision, `*/`/`*/MOD`, `S>D`/`D+`/`DNEGATE` double-cell ops, `MIN`/`MAX`/
`WITHIN`/`AND`/`OR`/`XOR`. It is not a capsule — feed it as raw text to a running REPL (serial
console, e.g. via `socat - UNIX-CONNECT:<serial_sock> < std79-exerciser.fth`), not through `EXEC`.

Written to test whether the same FORTH-79 word set behaves identically across every minted
identity (`zuse`, `rajames`/`bob`, `00`–`06`) and every supported architecture (amd64, aarch64,
riscv64) — see FABRIC-3.md §XII/§XII.4 for the full campaign writeup, including the two real
`D.`/double-precision bugs this found (one universal across all three architectures, one
aarch64-specific), and the WIREBIND heap-corruption bug (§XIII) found and fixed along the way
while getting the campaign's automation working reliably.

`results-20260910/` holds the raw captured output for all 27 (identity × architecture)
combinations from that campaign — `<arch>-<identity>.txt`. Cross-identity parity within each
architecture is perfect (0 diffs across all 9 identities, all three architectures).