Distribution

Four ways to run it.

The method is free to read and free to implement. What you’re choosing here is how the measurement reaches your companies and where the results live. Every route runs the identical published method, so results from any of them are comparable with results from any other.

Level 1 · nothing to build

Cohort link

A dedicated address for your programme. Companies complete the measure, results are tagged to your cohort, and the report comes back to you. No engineering, no procurement of software, usually live the same week.

Right for: accelerators, growth hubs, university programmes, first pilots of any kind.

Level 2 · your brand

Embedded

The same measure inside your own site, carrying your brand, so the company never feels handed off. The method underneath is unchanged and the deployment declares its conformance at the same public path as every other.

Right for: banks and institutions where the programme brand matters.

Level 3 · your systems

API and webhooks

Submit answers, receive the full structured result. Positions and diagnoses land in your CRM, case management or programme system automatically, and a webhook fires when a company rescores.

Right for: banks, larger authorities, anyone whose reporting has to live where their other data lives.

Level 4 · your estate

Certified self-hosted

You run the instrument inside your own infrastructure, pass the published conformance vectors, declare the version you run, and return anonymised aggregate movement. For cases where data genuinely can’t leave.

Right for: a small number of buyers who can’t let data leave, for residency or risk reasons. The reporting obligation is the condition of using the name.

Levels 1 to 3 keep results in the register, which is what makes them verifiable by someone else. Level 4 exists because refusing it would push a buyer into building an unverifiable clone, which is worse for everyone.

Where the line sits

Worth stating plainly, because it’s the whole commercial model.

Free, foreverLicensed
Reading the methodThe cohort and portfolio reporting
Implementing it yourselfThe sixteen cell diagnosis and the routing
A company taking the measureAccess to the hosted instrument at programme scale
A company's own full resultA verifiable result in the register

Anyone may rebuild the measurement from the published specification. Nobody gets the diagnosis library. And a self-run score can't be verified by anyone else, which is the only thing that matters once a funder, an investor or a committee is asked to trust it.

Machine and agent access

The specification is published as data, so software can read it without a human transcribing anything. That includes AI agents, and we would rather design for that deliberately than pretend it won’t happen.

SurfaceWhat it’s forStatus
/method/v1.4/*.jsonThe model, scales and bands as structured dataPublished
/llms.txtA machine-readable index of the specification, so a model reads the method rather than guessing at itpending
/.well-known/growth-score.jsonConformance declaration for any deploymentpending
/verify/{reference}Confirm a result is real, without exposing anything privatepending
MCP serverLets an agent administer the measure, retrieve a result, or check conformance on behalf of a consenting userpending

An agent can already read 37 published questions and ask them. What it can't do is produce a result anyone else will accept.

That’s the point of publishing the method and keeping the register closed. Anything that calls itself a Growth Score can be checked: which version produced it, when it was taken, whether it’s current, and whether the implementation passes the published vectors. A number generated by a model that has read this website isn't a Growth Score, and it won't verify.

The scoring path itself contains no generative model and never will. The same answers produce the same score every time, which is the only reason a second measurement can be compared with a first.