The reasoning was narrow, and none of it was visionary. Our programs were fine. Our data
files were fine. Our people were fine. Exactly one component in that stack was the problem —
the interpreter underneath all of it — and it was the only component we did not own.
Set against a rewrite, writing an interpreter is the smaller job. It has a definition, it
has a finish line, and you can test it against the thing it replaces. A rewrite has none of
those properties. So we measured the incumbent runtime instead of guessing at it, and we
built against the measurements: reads, writes, decimals, screens, keys.
It took a long time. It runs our company today.
5,079
Programs parsed, 100%
A real production corpus, not a benchmark suite — 536,000 lines of Business Basic source, read without a single exception.
3.63M
Records verified byte-exact
3,631,827 records across ten production files, read on both engines and compared record for record. Not sampled.
245/245
Arithmetic cases matching
Penny-exact decimals, including the PRECISION rounding model. This is financial software. The pennies have to match.