Raw dataset

First drop: September 2026

Nothing is published here yet. The first run hasn't happened. Once it has, every logged response (which engine, which question, which repetition, which companies were named, the accuracy coding, and the cited sources) publishes here as both CSV and JSON, unfiltered.

What will be in it

• One row per (question, engine, repetition), for all 21 questions and the brand-direct loop
• Engine, model/version where disclosed, run date, and session identifier
• Every company named in that response, in the order it appeared
• Accuracy coding for each named company, with the specific claim and what's wrong, if anything
• Every cited or linked source URL, where the engine exposes one

Format and cadence

A new CSV and JSON file per run, published alongside that run's essays, plus a combined file once there's more than one run to compare. Files are named by run date and are never edited retroactively. Corrections are published as a new dated file with a note on what changed and why, so anyone who already downloaded a run can diff it.

No signup or paywall. Licensing terms will be posted alongside the first file drop.