Plain terms, hard commitments. Including the one where we lose.

The Evaluation Grant is in the licence file, not in a sales deck: thirty days from first use, on your organisation's own data, for its own internal evaluation, with the output shareable inside your organisation. Three things it forbids, in plain words: relying on the output in production, redistributing the software, publishing the output externally. The period is extended only in writing — and anything beyond evaluation is a separate commercial licence.

The public licence is PolyForm Noncommercial: anyone, with no time limit, may read, study, fork and use the tool noncommercially. An evaluating organisation gets the thirty days above. A client gets a negotiated commercial licence — the public licence says of itself that it is not what a client buys.

Two honest notes: it is not an OSI‑approved licence, which some organisations exclude by policy — a brake on spontaneous adoption, not on a negotiated engagement; and like all such licences it ships without warranty, with a 32‑day cure period after written notice of a violation.

The engagement can end with the recommendation to keep what you have — and the machinery for that outcome is in the public source, not in our goodwill. Under 20 observations the tool refuses to recommend anything. The validation dossier states, in its own words, that it does not certify anything — it assembles what a reviewer needs and the decision is the committee's. The recommendation itself prints what would flip it, in figures. And on the corpus we measured, 3 of the 5 fields are already carried by regexes at zero cost: the cheapest possible answer sits in the headline, not in the fine print.

You supply one CSV — an id, a text, one column per field — and, if you wish, a filled questionnaire. Everything in the questionnaire is optional: what stays empty keeps this repository's default, and the report says so — a missing figure never becomes an invented one. Nothing in this path makes a network call; your file never leaves your machine.

You get back a measured report beside your own CSV — rates per field with intervals, refusing to quote a rate under 20 observations — and, under an audit engagement, a signed report: anyone, including a reviewer who does not trust us, verifies it against the public key that ships in the repository. The verification proves origin and integrity; about the correctness of the figures it proves, in its own words, nothing — that burden stays on the measurement.

The remaining terms of a paid engagement — payment, confidentiality, audit rights, change of control, precedence, and the rights in the produced report — are settled in the engagement letter, written before signature, not negotiated after delivery. They are business terms, so this page states their existence rather than quoting them; the letter is the document your counsel receives.

One rule from the public licence already binds the evaluation period: the output of an evaluation run is internal to your organisation — external publication is excluded from day one.