Each edition carries its method, source references, data dictionary and an errata log.
What every Ernex edition carries
One frozen extraction. Every figure is produced by code against a single extraction taken on a stated date, and nothing is refreshed while an edition is open.
Contract level data. Each edition ships with the underlying data set, one row per contract, so a reader can check any figure and build on it rather than take it on trust.
Automated release gates. The build stops if a figure loses its source, if a printed link does not match the act it cites, or if a natural person appears anywhere. Repeated builds are byte for byte identical.
A published method and an errata policy. The methodology, the limitations and the correction rules are published with the edition. Corrections are versioned and dated, never silent.
Companies are named, people are not. Legal entities are named where there is no legal obstacle, because an anonymised market cannot be acted on. Natural persons and sole proprietorships are not named, and their totals are reported without identity.
Sources and licences stated per data set. Every data set carries its source, its licence and its link, and the cleaning, matching and classification that Ernex adds is marked as ours rather than presented as the source's.
Builds that can be repeated years later. The extraction, the code and the working files are kept for at least five years and each edition is tagged, so a published file can be produced again from its own snapshot.
A named editor behind the text. Where AI is used in production it is declared together with the steps it was used in, and a named person reviews the content, with the authority to change or reject it, before the edition is released.