Skip to Content
How It WorksFrom Documents to Findings

From Documents to Findings

Everything in Classifyre starts with turning documents you already have into signals worth acting on. This page walks through that first stretch, in plain terms.


Sources: the systems you connect

A source is a connection to a system you already run — SharePoint, Confluence, Jira, a file share, a local folder, and more. Connecting a source doesn’t move your data anywhere; Classifyre reads it in place, on the schedule and scope you choose.

Scans: one pass over a source

A scan is a single crawl of a source. You can scan on demand or on a schedule, and each scan is recorded so you can see its progress and history. Repeat scans are smart about it: unchanged items aren’t reprocessed, changed items are refreshed, and items that disappeared from the source have their findings automatically closed out. Nothing is silently duplicated or lost.

Assets: what a scan finds

Every document, file, page, or record a scan discovers becomes an asset — along with metadata like its owner, location, and last-changed date. Assets are the unit everything else attaches to: findings, evidence, connections.

Detectors: what looks for what

A detector reads an asset’s content and looks for something specific. There are three kinds:

  • Pattern-based — precise rules for structured things like credentials or ID numbers.
  • Machine-learning — models that recognise entities like names or organisations even when the wording varies.
  • Custom, AI-based — detectors you define, including ones backed by an AI model, for anything specific to your investigation.

A source can run any combination of detectors, and you can build your own when the built-in ones don’t cover what you’re looking for.

Findings: what a detector produces

Every time a detector spots something, it raises a finding — one flagged match, on one asset. A finding carries:

  • What matched and the surrounding text, so you can judge it in context without digging through the original document.
  • Severity — how serious it would be if it’s a real issue, from Critical down to Info.
  • Confidence — how sure the detector is that the match is real, separate from how serious it would be.

A high-severity, low-confidence finding deserves a quick look. A high-severity, high-confidence one deserves action. Findings persist across scans rather than piling up as duplicates — the same issue is tracked over time, and your decisions (marking something resolved, ignored, or a false positive) are always respected by later scans.


Where this leads

A flat list of findings is a start, not an answer. The next step is working out which findings actually matter — that’s what Ranking & the Semantic Layer does.

For the full technical reference, see Sources, Detectors, and Findings & Results.

Last updated on