Vassoura

Vassoura is a periodic sweep across websites, earnings releases and public data that orbit a business. You describe in plain language what you want to look at; it browses, extracts, versions, builds the series and writes the report — and every number on the screen carries with it the page capture, the PDF or the API call that produced that value. It's where did this come from? with a ready answer.

Three templates cover the common cases: competitors' prices and terms (keeping what changed since the last collection and tracing a per-brand timeline); quarterly releases and presentations read as indicator-by-period series (with values flagged when the company restates a number in a later release); and public APIs and open-data portals collected month over month — including the ones that only answer whoever figures out the right call (Vassoura drafts the extraction plan, then takes over at whatever cadence the source tolerates). Adding a new target is usually a matter of editing a YAML.

It was designed with one use case in mind: someone needing to bring a number to a meeting and prove it wasn't pulled out of thin air. That's why every record carries its evidence attached — screenshot, PDF, raw response — and changes between runs are explicit, line by line, with the previous value preserved next to the new one. Four collections live in production today: 11 toll-tag brands (Sem Parar and competitors) compared every two weeks, 4 insurers tracked through quarterly releases, the entire FIPE vehicle table (more than 11,000 year-models, ~100k queries per run), and traffic across São Paulo's toll roads via ARTESP's open data.