There is a quiet asymmetry in vendor procurement that decides most of what goes wrong later.
A vendor who publishes the specific observation that would prove their claim wrong can be evaluated. A vendor who does not, cannot. That is the entire asymmetry, and it is doing more damage to buyer outcomes than any single technical decision in the contract. (Class E) Karl Popper drew the same line for science in 1934: a claim is only testable if some observation could force you to give it up. The procurement version is the same shape — a proposal you cannot test before you sign is a proposal you cannot evaluate after you sign either.
We publish a falsifier next to every non-trivial claim on this site. (Class C) The falsifier for "our receipts page shows the same numbers our pipeline produced" is a mismatch between the page and the upstream logs. The falsifier for "a workshop lead reaches a human within one business day" is a timestamp gap longer than that. The falsifier for "we will tell you what we are not sure about" is a post on this site that makes a strong claim with no evidence class and no failure mode. The pages where the claims live are the same pages where the tests live, so the reader does not have to ask. The mechanism is described at /falsifier-on-every-claim and the things we explicitly will not say sit at /standard/what-we-do-not-claim.
The reason vendors do not publish falsifiers is rarely incompetence. It is that an unfalsifiable proposal is a better commercial instrument than a falsifiable one. "Most advanced platform in its category" cannot be disproven, so it cannot be the basis for a refund, a credit, an out clause, or a renegotiation. "Indistinguishable from a human analyst" has no test condition, so the buyer has no anchor for the conversation when the output turns out to be distinguishable from a human analyst in the third week. The language pattern is engineered to keep the relationship inside the vendor's control after money has changed hands. That is the extraction posture, and the absence of falsifiers is its surface sign.
What we ask buyers to do, including buyers who never become our clients, is small and changes the game. Before you sign anything, write the sentence: "If, after ninety days, [specific observable] is true, this contract was mis-sold and the following remedy applies." Then ask the vendor to fill in the observable and sign next to it. Vendors who can do this on a phone call are vendors you can work with. Vendors who cannot, regardless of brand or price, are selling you a story. The receipt is the willingness to write the sentence, not the content of the answer.
Dr. Alianna J. Maren has written from an adjacent angle in her tummy-churning post, which in our one-line frame argues that the field has overspent on one architectural bet while underinvesting in the human verification step that would catch the errors that bet produces. (Class E) We read it as the same asymmetry seen from the other side: when verification is absent, the failure mode is not detected, so the claim is not falsified, so the vendor is not held to it, so the spend continues. The remedy in both directions is the same — the human, with the test, on the same page as the claim, before the money moves.
This is the operating discipline of the agency. It is also the reason our engagement model looks the way it does, which is described at /anti-extraction-engagement-model. We do not believe a project is real until a falsifier is on the page next to the claim. We do not believe a vendor is partner-grade until they will write one in front of you. We do not believe a buyer is protected by a contract that does not contain one.
If you have a real procurement decision in front of you and you want help writing the falsifier before you sign, that is exactly what the /workshop is for. Bring the proposal. We will write the sentence with you, on the page, before the money moves.
- /workshop — bring a proposal, leave with a falsifier.
- /anti-extraction-engagement-model — why our contracts are shaped this way.
- /falsifier-on-every-claim — how we apply the rule to our own work.
- /standard/what-we-do-not-claim — the sentences we refuse to write.