Method
How a standing is recorded
Published ahead of the first index, so that the method cannot be adjusted afterwards to suit a result.
Recording
A category is defined narrowly enough that a buyer would recognise it as a single decision. A set of buying questions is constructed for that category in the form a person would actually ask. Each question is put to every engine in the engine set, and the providers named in each answer are recorded as structured data at the time of asking.
A standing is then a count: how many engines in the set named a given provider, out of how many were asked, in a named month. It is generated from the recorded data and never written by hand.
The floors a category must clear
A thin sample produces a confident number, which is the most dangerous output an instrument like this can have. These floors exist to make silence the default.
Sample depth
A category is not published until enough recorded answers exist across the engine set for the counts to mean something. Below that floor, no index is published for that category at all.
Provider threshold
A provider earns a standing only when named by at least three of five engines. Below the threshold, the provider is absent from the index rather than ranked low.
Provenance on the face
Every published standing carries its category, its engine set, the month it was recorded, and the fact that it was recorded rather than paid for. A figure that travels without its provenance is not a NuLogic figure.
Frame, never claim
"Named by three of five engines in the recorded sample" is checkable. "The best tool in the category" is not, and is never published. Every statement is written so that a reader could in principle contradict it.
Silence over fabrication
Where a category has no recorded data, nothing is published for it. There is no modelled standing, no estimate and no placeholder.
Known limits
These are stated because they are true, not because they are flattering.
- Engines differ from one another, and a standing is specific to the set that was asked.
- Engine behaviour changes without notice, so a standing describes a month and not a permanent state.
- Being named by an engine is not evidence that a provider is better, and NuLogic does not assert that it is.
- Question construction affects which providers appear, which is why the questions are published alongside the standing rather than summarised.
A register that hides these would produce more confident copy and a worse instrument.