6. Eligibility, Selection, and Inclusion Logic
How prices are grouped, cleaned, and selected for scoring.
6.1 Grouping prices into product categories
Everything a provider sells is grouped into product categories at the granularity of a single purchasable item, not a service family. One virtual machine type is its own product category, as is one storage tier. Where a region publishes more than one price for the same product category, the median of those prices represents the region, so that one atypical entry cannot carry the region's position on that category.
6.2 Removing erroneous prices
For each product category, the price published by every region is compared against the consensus across all regions of that provider. A price that differs from the consensus by orders of magnitude is removed, because observed genuine variation in the price of one product category between regions of the same provider stays below a factor of three, and a difference several thousand times larger indicates that the value holds some other quantity than the one the field claims. Where a product category's prices remain mutually inconsistent after that removal, the whole product category is excluded from the pricing score, on the grounds that no consensus exists to compare a region against. The count of prices removed and product categories excluded is recorded for every run, as described in Section 9.
6.3 Selecting the comparison basket
A product category enters the comparison basket only where it is priced across a sufficient share of the provider's regions for a comparison to hold. A product category sold in a handful of places is left out, because too few regions price it for a comparison to mean anything.
Regions with very small catalogues are excluded from the test of how widely a product category is sold. Newly launched regions open close to empty, and if they counted towards that test, every launch would raise the number of regions a product category has to appear in and would push established product categories out of the basket. Those regions are still scored on the basket like every other region.
6.4 Product categories counted for coverage
The coverage score counts a product category only where the provider prices it in more than one region. A product category priced in exactly one region worldwide says nothing about the completeness of any region, since no other region could have stocked it and none gains or loses position because of it.
A price removed under Section 6.2 still counts towards coverage. An erroneous price is still evidence that the provider sells the product category in that region, and the coverage score asks what is sold and not what it costs.