5. Inputs and Evidence Sources
The price data the scores are built from, and the errors it carries.
The only inputs to either score are the providers' own published price lists. For each provider, the methodology team reviews the structure of the published list, establishes which fields carry the on-demand price and the unit it is quoted in, and maintains the translation of that provider's list into the common shape the scoring operates on. That translation work is provider-specific and is revised when a provider changes the structure or the field semantics of its published list.
No other evidence is used. The chain contains no customer billing data, no telemetry, no survey input, and no third-party price aggregator. A reader who wants to challenge a score can in principle reach the same inputs, since the source is public for every provider in scope.
The price data is refreshed and rescored on a recurring cycle, and each run is identified by its data version. A region's scores therefore describe the provider's catalogue as published at the time of that run, and not a long-run average. Providers change list prices, launch product categories, and extend catalogues into newer regions between runs, so two data versions are not directly comparable, for the further reason given in Section 8.
The published lists carry occasional errors that survive into the raw input. The most consequential is an entry whose value is not the per-unit price for the unit stated on the entry, for example a multi-year commitment total appearing in a field the list defines as an hourly rate. An entry of that kind holds a different quantity from the one its field claims, and it would distort any statistic computed over the product category it belongs to. Section 6 describes how such entries are identified and removed. Their existence is the reason the method uses medians throughout in place of means, since a median is insensitive to a small number of extreme values in a way a mean is not.