How Chiri Atlas reports are made
Chiri Atlas is a directory of independent research reports on AI tools. Here is exactly how a report gets built, checked, and published, and what we do and do not verify.
The pipeline
1. Research agents gather from primary sources
For each tool, research agents pull from the vendor’s own site, official documentation, SEC and Crunchbase filings, press releases, and reputable third-party reporting. Secondary sources (blog posts, aggregator summaries) are used only to point to a primary source, never cited on their own.
2. Every factual claim is extracted and verified
The draft report is broken into individual factual claims. Each claim is checked against its cited source and marked supported or not supported, with a confidence score and a short note explaining the match or the gap. Unsupported claims are flagged rather than deleted, so reviewers can see exactly what did not hold up.
3. Citability scoring
Each report is scored on how well it holds up to citation: source diversity, claim density, how many claims are supported at high confidence, and whether the required sections (pricing, security, leadership, funding) are present. This score feeds the report’s "earned confidence" and determines whether it needs another verification pass before publish.
4. A human reviews and approves before publish
No report goes live without a human reviewer reading it end to end, checking the flagged claims, and signing off. The reviewer can send a report back for another research pass if claims are thin or sources are weak.
What the Chiri Score is
Where a report shows a Chiri Score, it is an average of five dimension scores, each rated 0 to 100 with a written rationale tied back to sourced claims. The Chiri Score is not a review pulled from user ratings; it is our own analysis, and we say so. Not every report has a Chiri Score yet: older reports predate the scoring system and show research findings without a numeric score.
Update policy
Reports are refreshed as the tools they cover change: new funding rounds, pricing changes, security incidents, or major product releases. Each report shows the date it was last verified. That date reflects when a human last reviewed and approved the content, not when the page was last deployed. If a report looks out of date, it is because we have not re-verified it yet, not because the date was quietly rewritten.
Transparent AI disclosure
Chiri Atlas reports are AI-assisted and human-reviewed. Research agents do the first pass: gathering sources, extracting claims, and checking them. A human editor reads every report before it publishes, checks the flagged claims, and can reject or send back anything that does not hold up.
Google's own guidance is that content is judged on its value to readers, not on how it was produced. We publish this page so anyone can see exactly what "AI-assisted" means here: which steps are automated, which step is not, and what a passing report actually had to clear.