We publish only what was instrumented. If a deployment wasn't measured against an agreed baseline, we describe it honestly and leave the figures out. No estimates, no "up to", no numbers borrowed from someone else's slide.
The task: matching supplier items to estimate positions for an industrial supply operation. The deployment runs inside a closed perimeter on GX10-class hardware with open-weight models. Nothing leaves the building.
Why there are no numbers: the deployment predates our measurement protocol. Hours saved and match accuracy were never instrumented against a baseline — so we won't quote them. Inventing a figure would cost more than it earns.
What changed because of it: every deployment since carries the protocol — a hand-matched baseline on the customer's own sample, an acceptance gate agreed in writing, and numbers measured before any licence exists.
We have exactly two founding-reference seats. Each comes with a setup discount — in exchange for a measured, public case: your numbers, verified on your corpus, published here.
Baseline fixed on your sample, gate agreed in writing, numbers published with your sign-off. Setup discount applies.
Ask about the founding seat Case 03 · reservedSame terms, same protocol. When both seats are taken, the discount ends and cases publish at the usual pace.
Ask about the founding seatPublished in advance, so you can hold us to it. If a section is missing from a future case, that case doesn't go up.
Who the customer is, what they do, what perimeter the box sits in.
The specific job — in the customer's words, not ours.
The reference set, and the hours it took a human to do the same work once.
Accuracy against the gate, hours before and after — measured, not claimed.
A named person, with permission. No name, no quote.
A 20-minute call, no slides. Bring real documents — we show the system live and explain the founding-reference terms: setup discount for a measured, public case.