pretium

Method & sources

How the numbers on this site are collected, checked and shown.

Sources

How a number gets here

  1. A collector fetches each source on its own schedule. It identifies itself, reads robots.txt first, waits between requests, and asks conditionally so an unchanged file is not downloaded again. If a site refuses, the run is recorded as blocked and left alone.
  2. The response is kept exactly as received, named by its SHA-256 hash: the evidence.
  3. An extractor reads the observations out of it and publishes them as events on Kafka. Every observation carries the run that produced it and the hash of its evidence.
  4. The catalog checks every price before storing it. The strongest checks are identities the numbers must satisfy: total cost equals fee plus exchange-rate margin; the margin equals the gap between the provider's rate and the interbank rate. A small breach is flagged; a large one holds the price back.
  5. Channels that publish the same rate are reconciled daily, and every value a source later changes is kept as a revision.

Reading the checks