You don’t know
what your legacy
codebase does.
The last engineer who knew how it worked left years ago. Before your team modernises it, Zero-Regression produces a signed, machine-verifiable record of what the system actuallydoes — so you can rebuild against truth, not folklore.
91% mutation score / 0 actionable survivors
fintx/fintx-accounting · abandoned Java financial system
The engineers still in the building know maybe 20% of what the system does.
The other 80% is tribal knowledge that left with the people who wrote it. Modernising on top of that gap is how rewrites quietly ship regressions for the next eighteen months.
Tests are not proof of understanding. Coverage measures whether code ran under test, not whether anything was checked. The gap between those two ideas is wide enough to ship production bugs through.
We measured it on the JSR 354 reference implementation — the Java standard for monetary values — in April 2026. moneta-core ships at 61.8% line coverage, with one module reporting no coverage data at all despite a green build. Full methodology and the surrounding pattern in the linked piece.
Before a board signs off on seven figures of modernisation spend, someone needs to prove what the system actually does. Not describe. Not test. Prove.
Map → Test → Prove.
Three phases, one artefact. Each stage compounds the last. Click a phase, or let it run.
Map
A frontier reasoning model reads your source code inside a VPC-hosted container and produces a BlastRadiusManifest — every branch, every dependency, every cyclomatic complexity hot-spot, every mock seam. Source never leaves your environment.
Test
A second model, from a different frontier lab, takes the manifest and generates a complete test suite targeting 85%+ line coverage. Output: the test suite itself, plus an ExecutionTelemetry file capturing what passed, what failed, what timed out.
Prove
A third model, from a third frontier lab, runs mutation testing against the generated suite, tightens assertions where mutations survived, and emits a CryptographicParityCertificate — a signed JSON document recording the mutation score, actionable survivors, runtime, and a verifiable hash.
Three models. Three labs. Deliberate disagreement is the point — each phase catches what the previous one missed.
One engagement. One certificate.
The only deliverable. A JSON document your team can verify, your board can cite, and your next rewrite can be graded against. Hover a field to see what it means.
{
"target": "fintx/fintx-accounting",
"target_type": "abandoned Java financial system",
"mutation_score": 0.91,
"actionable_survivors": 0,
"pipeline_runtime_seconds": 10847,
"certificate_hash": "330b84f429dfbcd4b9b743aa3569d461c81131d441ebceae181dd8796261217e",
"signed_at": "2026-04-12T14:22:09Z",
"signing_authority": "Zero-Regression (Mindway Media Ltd)"
}Mutation score
The fraction of deliberate code mutations the generated test suite caught. 0.91 means 91%. Industry considers 80%+ solid, 90%+ strong.
If you’ve been having this conversation in your own head for months, this is built for you.
Zero-Regression is for engineering leaders who inherited a system they didn't build. Typically Java or Python monoliths, ten years old or older, where the original technical team has moved on and the board is asking how modernisation will be funded and justified.
Built for you
You recognise one or more of these:
- [+]You inherited the codebase. The team who built it is gone or going.
- [+]Modernisation is a board-level line item this year or next.
- [+]Your CI is green but nobody can tell you what it covers.
- [+]The last incident was caused by behaviour nobody knew was there.
- [+]You're about to fund a rewrite and can't defend it yet.
Probably not for you
Zero-Regression isn’t a fit when:
- [ ]The system is actively and cleanly documented by the people who built it.
- [ ]You're looking for a generic audit or compliance report.
- [ ]You want someone to do the rewrite. We only produce the evidence.
- [ ]Greenfield projects under two years old.
Find out what your
codebase actually does.
Twenty minutes. One call with Ross. By the end you’ll know if Zero-Regression is the right instrument for your situation — or whether you need something else entirely.