Critical federal legacy systems in GAO's review still use outdated languages.
ROTATION / CONTINUOUS
GROUND TRUTH FORCODE NOBODY CANREAD.
Rosetta gives AI a way to prove every change against the real program and database state.
UNDERSTAND / CHANGE / PROVEOLD CODE.
LIVE MISSIONS.
Operate with known cybersecurity vulnerabilities while their missions remain live.
Systems flagged in 2019 were complete by February 2025.
THE FAILURE MODE
FLUENT.
CONFIDENT.
WRONG.
AI can explain a patch perfectly while changing behavior nobody intended. In MUMPS, the database can move even when the printed answer looks right.
MAKE CORRECTNESS
EXECUTABLE.
Rosetta turns the real runtime into the reviewer. The agent gets specific evidence and another move.
HOLD STATE
Start every case clean.
RUN BASELINE
Capture known behavior.
RUN CANDIDATE
Use the same inputs.
NAME THE DIFF
Output, errors, and state.
THE HARNESS GETS
THE LAST WORD.
› Guard the next-of-kin lookup without changing behavior.
✓ PATCH COMPLETE
CONFIDENCE / HIGH RUNTIME CASES / 0› Guard the next-of-kin lookup without changing behavior.
01 ISOLATE STATE
02 EXECUTE CASES
03 DIFF OBSERVABLES
ONE HARNESS.
THE WHOLE WORKFLOW.
01 AGENT / PLAN / VERIFY
02 RUNTIME TOOLS
03 HELD-OUT BENCHMARKS
04 PROOF RECEIPTS
05 TRAINING SIGNAL
Open Rosetta inside the codebase. Trace unfamiliar logic, change it, execute it, measure the repair, and keep the proof without exporting customer code.
TRUST BECOMES
MEASURABLE.
The same held-out repairs run under every condition. Rosetta keeps the cases, traces, and verdicts for inspection.
SHIP THE CHANGE.
KEEP THE PROOF.
Rosetta puts an execution harness inside the coding environment. Runtime behavior becomes benchmark evidence and a proof receipt for every proposed change.
› rosetta verify AJETIU2 -c candidate.m
01ISOLATIONclean state held
02EXECUTIONbaseline + candidate / 4 cases
03OBSERVABLESstdout / errors / persistent globals