A free 'AI-powered' Crunchbase alternative listed Anthropic at an $18B valuation. The real number was off by 50x

The pitch and the spot-check

StartupWiki bills itself as an AI-powered startup research directory and a free alternative to Crunchbase, assembling profiles from AI agents plus community contributions, complete with VERIFIED badges and an "agent ledger" showing the pipeline's work. As a Show HN it drew a generous but pointed audit. The reliability problem surfaced fast:

"I see quite outdated data. Anthropic listed with valuation 18B and latest round at 4b? Just to compare, their real latest round was 65b with valuation 965b." — a commenter spot-checking the data

Coverage was the next wall. A reviewer with deep startup-ecosystem context came up empty and offered the obvious data path:

"0 for 10 on some startups (large and small, YC and not) that came to mind. It's easy to scrape YC startups from… ycombinator.com/companies. Scrape that and a dozen other investors' portfolio pages and you'll have a useful fraction of startups." — tlb, a commenter from the YC ecosystem

"Verified" has to mean something

The most damaging critique went after the product's core promise — the badge:

"I expected the VERIFIED badges to link to some sort of provenance information. That seems like a must, otherwise (given the 'assume everything's incorrect' disclaimers) I'm not sure why one would take that badge seriously." — a high-karma reviewer on the empty badge

And the constructive note pointed at a way out that doesn't require re-scraping the world:

"you should link your data to wikidata which will get you free connection back to crunchbase and other sources… You could even back some of the data from there" — a commenter suggesting Wikidata federation

The trust problem AI can't paper over

A directory's entire value is that you can trust a number without re-checking it. The moment Anthropic's round is off by 50x and a "VERIFIED" badge links to nothing, every other figure on the site has to be independently confirmed — at which point the directory has saved you nothing. The maker is responding in the thread and adding companies, which is the right instinct. But the lesson the room keeps circling is older than AI agents: in a reference product, provenance isn't a feature you add later, it's the thing you're actually selling. Generating profiles with a model is the easy 80%; making each cell defensible is the 20% that is the whole job.