Proven on real products#

Note

A real client brief, built by an agent with Specgate. The brief came from a real client; the product was built end to end by a coding agent (Claude) driving Specgate, with business decisions delegated to it and written down as assumptions. It is not a customer deployment, and the code repository is private.

These six pilots are agent-driven adoption, not a customer story. Each one started from a brief a real client wrote — vague, detailed, heavy on architecture, or regulated — and ended as a working product with its specification, tests and CI. The numbers come from each product's generated traceability matrix and its test suite. What these pilots are not: a team choosing Specgate and using it on their own. That reading is still to come.

Every page lists what Specgate got wrong during the pilot and the release that fixed it. Most of the tool's rules exist because a pilot needed them.

Rebar CrewA field app for a rebar crew: 47 of 47 requirements and 599 tests, after a first round that was done and unusable.BookingsAppointments for local businesses: 38 of 38 requirements in five cycles, each closed by using the product.Credit shopA five-line brief became 24 requirements; every test green, and nobody could buy until the smoke test.FinCoreCore banking on Spring Boot: 59 of 60, with GDPR against AML retention left to the client.EventTechEvent operations on Quarkus: 38 of 38, the event rebuilt at any time from its facts.HealthTechClinical cases on Quarkus: 23 of 24; drug interactions wait for an answer instead of a guess.

Side by side#

Project	Brief	Stack	Requirements	Tests	Waiting	Specgate defects found
Rebar Crew	Detailed, hand-written as domain packs	Expo + Next.js, hexagonal TS core	47 of 47	599	0	32 (one open: #21)
Bookings	Detailed, with screenshots	Next.js + hexagonal TS core	38 of 38	212	0	22, plus #60
Credit shop	Vague	Next.js + hexagonal TS core	24 of 24	272	0	2, plus #60
FinCore	Architecture-heavy	Spring Boot 3.3 · Java 21 · PostgreSQL	59 of 60	112	1 (REQ-055)	7, plus #79
EventTech	Architecture-heavy	Quarkus 3.40 · Java 21 · PostgreSQL	38 of 38	71	0	2
HealthTech	Regulated	Quarkus 3.40 · Java 21 · PostgreSQL	23 of 24	35	1 (REQ-020)	1

Requirements include REQ-001, the adoption baseline. Java test counts are the suite's run totals; Bookings and Credit shop are test cases counted in the source; Rebar Crew is its suite's total at the end of the pilot. #60 — init proposing requirements on a fresh skeleton — was hit by three pilots.

What all six found#

The gate verifies what the specification says. Running the product finds what it does not say. In every pilot, using the running product turned up something the green gate could not see: a product with no screens, an API with no button, an action on an id nobody could list, a build that failed in production, an application that did not start. Each one became a short follow-up change — and several became rules in specgate draft --check. The path from a brief to a specification is in From a client brief to a spec.