Hey all. Last week I said I was heads-down on a project and would ship it once I had a chance to vulnerability test it. It is live: theknowngood.com
It tracks published evaluation data for AI models. Benchmark results, pricing, performance, arena ratings. Around 900 models, roughly 2,300 pages. Exports are free and ungated as CSV and JSON. No accounts, no tracking, no API to sell you.
@SuMiTa asked about the vulnerability testing approach, so here is the honest version. It was my first real attempt at red teaming my own thing and I would not call it comprehensive. What I leaned on instead was removing categories of problem rather than defending against them. Postgres is the source of truth but nothing touches the database when you load a page, everything renders to flat static HTML. nginx is bound to loopback only and the site is reached through an outbound tunnel, so there are no inbound ports at all. Still very open to specifics on what I should be testing next.
Fair warning that this is self-hosted on hardware in my house. If it drags for you, please say so and I will go look at it.
Feedback of any kind is welcome. There is a newsletter signup on the homepage, fully optional, no account needed.
I also wrote up everything that went wrong building it. Link in a reply below.