How wrong is a model about a company when it has no register to check?
registry-mcp puts three national company registers behind one MCP tool — Brønnøysundregistrene / Enhetsregisteret ( brreg ) by organisasjonsnummer (orgnr, org.nr), Companies House by company number, and Bolagsverket by organisationsnummer. The pitch has always rested on a counterfactual: without a register, an agent answers company questions confidently and wrongly. That is an easy thing to…
A recent study examined how well a model performs when given three national company registers - Brønnøysundregistrene, Companies House, and Bolagsverket - to verify information. The pitch behind the model was that without a register, an agent could confidently provide incorrect answers. To test this, the study used the claude-sonnet-5 model and conducted 45 calls with hand-written prompts.
Without using the registers, the model answered 8 correctly, 3 incorrectly (confidently or dangerously), and 34 hedged. It hesitated and provided unsourced answers on 13% of the trials. When the registers were used, the model answered 18 correctly and 4 incorrectly, with 27 hedging and 4 producing unsourced answers.
However, the study found that the register's main benefit is not accuracy, but provenance - the ability to claim that an answer is backed by a reliable source. The model demonstrated this when it correctly stated that Companies House does not publish VAT registration data for a UK company, even though it could not verify the record. The no-tools arm (no access to registers) also made the same mistake, showing that the register only buys provenance, not accuracy.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
