The method
Not magic.
Machinery.
Five stations, in order, every time. The reason Palm’s output can face an auditor is that nothing in the line ever guesses.
The line
Your code enters whole. Verified rules come out — with addresses.
stations 01–05 · one pass · in order, every time
- 01 Ingest every artifact, read in place
- 02 Parse deterministic syntax, no guessing
- 03 Graph the dependency web, complete
- 04 Extract rules with line-level lineage
- 05 Verify same input, same output
run 1 · 5b1c·e7 = run 2 · 5b1c·e7 = run 3 · 5b1c·e7
Ingest · parse · graph · extract · verify
Five stations, zero guesses
Ingest reads every artifact — programs, copybooks, JCL, screens — inside your perimeter. Parse builds a true syntax model of each dialect; there is no “probably” in a parser. Graph assembles the full dependency web Palm Key exposes.
Extract walks that structure to lift each business rule with its exact source span — 50,000+ lines held in a single deterministic pass. Verify replays the line: identical input, identical output, run after run. In benchmark this method lands around 95% extraction accuracy, with 100% lineage by construction.
AI assists where judgment helps — naming, summarizing, classifying — but it is bolted to the structure, never substituted for it. That’s the difference between an answer and an opinion.