Standard open stack
A standard open stack: Next.js, Fastify, Prisma, PostgreSQL. No proprietary runtime. The buyer owns the code outright.
01 Evidence
Measured, not asserted
74 execution audits across 39 distinct business domains: installed, built, containerised, booted and tested, never scored by reading. Best score 9.7.
74Execution auditsAs of 31 August 2026.
39Business domainsAs of 31 August 2026.
138Automated gatesAs of 31 August 2026.
9.7Best score out of 10As of 31 August 2026.
Figures as of 31 August 2026. Audits are executed, never scored by reading.
02 The Corpus
Representative audited results
| Domain | Score |
|---|---|
| Marina operations | 9.7 /10 |
| Hazardous waste transfer | 9.6 /10 |
| Commercial diving operations | 9.6 /10 |
| Cold-chain logistics | 9.6 /10 |
| Housing regulator | 9.6 /10 |
| 20-entity distribution network stress testAll twenty entities intact. | 9.4 /10 |
Execution-audited by us, reproducible by you. Independent verification has not yet been commissioned, and every claim is designed so a third party can reproduce it with standard tools.
The corpus keeps its failures. Published scores include a 5.5, a 7.4 and a 7.5, each followed by a root cause, a permanent gate, and a re-run of the same brief on the fixed engine: 5.5 to 9.4, 7.4 to 9.6, 7.5 to 9.7. Every defect on record was found by our own checks, none by a customer.
03 Output Quality
What every generated application ships with.
Seven layers
Full stack across seven layers: database schema, API with typed SDK, frontend, tests, CI/CD with security scanning, containers, observability.
Size
Roughly 290 to 560 files depending on domain size, measured August 2026 across live exports.
Security defaults
Owner isolation, role-based access control, and schema validation on every route. Unauthenticated requests are refused on every verb. A request for another user's record answers as if the record does not exist.
Tested
Each application ships its own test suite, typically 249 to 463 substantive tests, largest on record 808, with a coverage gate enforced in its own CI at a size-verified floor. Measured server-side line coverage is roughly 40 to 58 percent by export size; user-interface code is excluded because no browser test harness ships.
Money handling
Financial amounts are stored as 64-bit integers, so very large values are handled exactly.
Relational capability
From prose alone the engine has derived seven distinct correct relational designs for two differently-named relationships to the same entity, across seven domains.
Generation time
Typically 45 to 75 seconds from brief to downloadable application.
Ready to extend
Roughly the first 80 percent of any enterprise build, with the domain's proprietary business rules left as deliberate extension points.
138 automated gates run on every change to the generator, each one pinning a defect that was found, reproduced as a failing test first, then fixed.
04 Breadth
Domains exercised to date
No per-domain templates and no human modeller; the engine composes domain-neutral primitives around the declared structure.
Commercial lending
Esports
Third-party risk management
KYC compliance
Automotive service
Field service
Laboratory management
Cold-chain logistics
Social housing regulation
Marina operations
Aviation operations
Sustainability reporting
Physiotherapy clinics
Sports venue booking
Commercial diving
Hazardous waste transfer
05 The Demo
We do not ask you to trust the claim. We ask you to reproduce it.
1
You write the briefs
You write the briefs: your domains, your words, sealed until the session.
2
Generation runs live
Generation runs live on the production system in front of your team. Nothing is touched between submission and result; typical run time is 45 to 75 seconds.
3
You verify on your own machines
Your engineers take the archives away and verify everything on your own machines: build it, boot it, test it, hash the specification seal, download twice and compare. No step needs our participation.
What this eliminates
- Staged demos. You watch the run live, and a deployment freeze applies for the session.
- Cherry-picked output. You write the brief.
- Unverifiable claims. Byte-equivalence either reproduces or it does not.
Submit a brief
Describe the application you need and we will be in touch to arrange a live generation.
06 Category Context
Four categories of AI development tooling. None is deterministic end-to-end.
Code completion
Autocompletes inside the IDE. Architecture, integration and consistency remain the engineer's problem.
Prompt-to-app builders
Impressive prototypes, consumer-grade output. Not built for enterprise audit or extension.
Low-code platforms
Generate applications, but into a proprietary runtime: platform lock-in by design.
Agentic builders
Genuinely capable, inherently non-repeatable. The same request produces different software every run.
We found no public system that combines plain-English input with byte-identical, cryptographically verifiable full-stack output the customer owns.
07 Get in Touch
Reproduce the claim
Exploring strategic conversations in enterprise AI. Technical detail available under NDA.
Send an enquiry
We reply to every serious enquiry.