A category is a property, not a feature list
Writable into a procurement requirement — that is the test of a category.
A feature list is what a vendor says their product does. A category property is what a buyer can write into a procurement requirement before any vendor is in the room. "Reproducible, tamper-evident, defensible" is writable; "smart AI dashboard" is not. The test of a category is whether the property precedes the product.
There is a simple test for whether something is a category or a feature list: can a buyer write the property into a procurement requirement before any vendor is in the room? "Reproducible, tamper-evident, defensible" can be written into an RFP — a procurement officer can specify "every output carries a per-output hash-chained manifest with a replay anchor" without naming a vendor. "Smart AI dashboard with insights" cannot be written into an RFP without naming a vendor, because it is not a property, it is a pitch. The test is whether the property precedes the product.
The property the auditable oracle owns
The auditable-oracle category is defined by one property: every output is reproducible, tamper-evident, and defensible. Reproducible: a third party can rerun the computation and get the identical bytes. Tamper-evident: the manifest is hash-chained (BR-842), so a retroactive edit breaks the chain. Defensible: the manifest answers the examiner's questions field by field — provenance, not narration. The property is the category; the product is one instance of it. A buyer who writes the property into an RFP is buying the category, not the vendor, and that is what makes it a category.
Why the property, not the feature, is defensible
A feature can be copied by a competitor in a sprint. A property that is writable into an RFP cannot — because the competitor has to actually deliver the property, not the feature, and the property is a constraint on the substrate. "Reproducible by construction" is not a feature you bolt on; it is a property of how the system is built, down to the memory model and the hash chain. A competitor who copies the feature list without the substrate fails the RFP. That is the moat the category provides: the property is harder to deliver than the feature, and the RFP makes the difference visible.
A feature can be copied in a sprint. A property that is writable into an RFP cannot — because the competitor has to deliver the property, not the feature, and the property is a constraint on the substrate.
- Reproducible: a third party reruns the computation and gets identical bytes.
- Tamper-evident: the manifest is hash-chained, so a retroactive edit breaks the chain.
- Defensible: the manifest answers the examiner field by field, not by narration.
- Writable into an RFP without naming a vendor — the test of a category.
What the property refuses
The property does not include "accurate" or "true." Certified ≠ True is the boundary: the category certifies provenance and reproducibility, never truth. The refusal is what makes the property writable — because a buyer who wrote "certifies the output is correct" into an RFP would be asking for a claim no honest system can make, and every honest vendor would fail it. The property that is writable is the property that is honest, and the property that is honest is the one that refuses to certify truth. We are pre-revenue, building the POC; the property is built, the calibration that would add "calibrated" to the property is gated (Gate 1b), and the category claim is gated behind a signed paid pilot.
The test of a category is whether the property precedes the product. "Reproducible, tamper-evident, defensible" can be written into an RFP before any vendor is in the room. "Smart AI dashboard" cannot. Only one of those is a category.
— category note