Training

The lessons

1. 01 Collect

   Get the public record onto your own disk

   Scrape a company website and query a parliament’s public API, keeping a URL and a date on every quote.

   You finish withA folder of sources you can cite

2. 02 Check

   Test every claim against its source

   Turn an overlap between two records into a lead, then fact-check it in a fresh session so the checker can’t see your reasoning.

   You finish withA claim ledger with verdicts and confidence

3. 03 Report

   Write only what the evidence supports

   Draft from your checked claims, then have a decision model flag every sentence its source doesn’t support before you sign off.

   You finish withA report where every sentence has a source

4. 04 Remember

   Build a reporting memory you can reuse

   Move only checked claims into an OpenKnowledge workspace, with aliases and a log of what you left out and why.

   You finish withA memory your agent can query

5. 05 Go local

   Run a fact-check on your own machine

   Install Mycroft in Goose, choose a model that runs on your computer, and fact-check your report with it.

   You finish withA local fact-check, compared with your own

By the end you’ll know what Spotlight does at each step and why, so you can run it on your own leads.