Labs is genuinely good for a first pass, though generated tables tend to look more settled than the evidence behind them, so I open every cited link before reusing a number anywhere client-facing. Sorting sources by date also cuts a lot of the stale-citation problem. Disclosure: I work with AllyHub, and we keep a comparison of research tools at https://allyhub.com/blog/ai-research-tools. Has any task type worked well enough that you stopped double-checking it?