Dmytro tests tools on ordinary jobs that turn up in a small team's week: summarise a long PDF, draft an outline, clean a transcript, complete a function, generate a short clip. Not a benchmark suite, not a leaderboard — a set of tasks a freelancer or a small office actually runs.
Each test watches three things: whether sources are shown, whether the export works without a fight, and whether a teammate could follow the result without asking what happened. What survives that gets written up in plain language.
There are no lab scores on this site, only workflow observations you can repeat on your own machine. When the evidence is thin, the page says so.