VerdictPal is live

Why this exists

Every guide to AI research tools reads the same way: an affiliate listicle that goes stale in a month and never says what a tool is bad at. That is a fine business and a useless artefact if you are the one who has to defend a bibliography.

So this is the opposite bet. Finite instead of exhaustive. Dated instead of evergreen. Opinionated instead of neutral. If a tool cannot show you when its pricing was last checked, it has no business telling you what to buy.

What shipped

• 52 tool dossiers, 9 at flagship depth
• 175 models with sourced benchmark rows
• 29 benchmark pages in plain language
• 1789 sourced score rows, each with a source and a check date
• 29 workflow guides — stacks, paths, playbooks, showcases
• An Atlas Report of aggregate figures, free to cite
• Every reader-facing surface in English and German

What it refuses to do

• No leaderboard. Ranking these tools on one axis implies a kind of quality that does not exist for this stuff, so each tool stands on its own dossier.
• No “best AI tool for X”. The desk will tell you which tool it would back for a job, and show the reasoning.
• No hidden affiliate links. Zero programs are live today; if one ever is, it gets labeled on the tool dossier surface and it will not move a score.
• No first-party benchmark scores until grading and two-reviewer sign-off are done. The Citation Fidelity protocol is registered and its captures are complete; the results are not published, because publishing numbers the desk has not verified would undo the entire point.
• No invented figures. Where a number is third-party it says so, and where the terms simply do not state something, the tool says that instead of guessing.