Zero-Regression
Est. 2024 · UK · Remote
File 01 · IntroductionProof of concept · fintx/fintx-accounting
Evidence, not opinion

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.

Proof of concept

91% mutation score / 0 actionable survivors

fintx/fintx-accounting · abandoned Java financial system

§ 02The gap

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.

Source dataMeasured 2026-04-25
JSR 354 Reference Implementation v1.4.5
Line coverage
61.8%
Branch coverage
52.1%
Modules measured
4 of 5
1 module reported no coverage data despite passing build
§ 03The pipeline

Map → Test → Prove.

Three phases, one artefact. Each stage compounds the last. Click a phase, or let it run.

run: zero-regression / fintx-accounting / 2026-04POC SAMPLE
Phase 01 / MapWeeks 1–2

Map

Read, parse, manifest

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.

GET/api/v1/accounts/:id
POST/api/v1/transfers
LegacyAccountService.resolve()
FxRateCache.get()
LedgerWriter.append()
GET/internal/cron/settle-pending
Phase 02 / TestWeeks 3–5

Test

Generate, execute, log

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.

Phase 03 / ProveWeeks 6–8

Prove

Mutate, tighten, sign

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.

targetfintx/fintx-accounting
target_typeabandoned Java financial system
mutation_score0.91
actionable_survivors0
pipeline_runtime_seconds10847
signing_authorityZero-Regression (Mindway Media Ltd)
— SEALED —

Three models. Three labs. Deliberate disagreement is the point — each phase catches what the previous one missed.

Stage 01 / 03 · MAP
§ 04Proof of concept

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.

zr-cert · fintx-accounting.json● sig OK
{
  "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)"
}
Field
mutation_score

Mutation score

The fraction of deliberate code mutations the generated test suite caught. 0.91 means 91%. Industry considers 80%+ solid, 90%+ strong.

number · 0.00–1.00
Try: hover mutation_score or certificate_hash
§ 05The buyer

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.

You, probably

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.
Not this time

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.
File closed · ready to open yours

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.