A research synthesis,measured against the best.

The first signal shows Rhizome can build. This one asks whether the same coordination generalizes past code — whether a field of agents can do open-ended research, and do it well enough to stand next to the strongest single systems.

The task: a structured synthesis of roughly 2,500–3,500 words on a deliberately hard question — how autonomous AI agents verify their own work: the 2026 landscape — error detection, fighting hallucinated progress, and gating real actions. It calls for synthesis, not fact-lookup: a taxonomy of approaches, a comparative analysis, the open gaps, and a list of real, verifiable sources.

What makes the signal trustworthy is the comparison protocol. The same prompt goes to Rhizome, LLM Council and ChatGPT Pro. The rubric is fixed before any run — factual accuracy, source quality and reality, originality, synthesis depth — and every raw output is published so the comparison is something you can audit, not a claim you have to accept.

The mechanism on display is self-verification: the system doesn't trust its own prose, it runs the result through independent checks — source existence and relevance, originality — and a dissent agent that keeps the field from collapsing into early consensus. To be explicit: the originality checks are a quality control, never a way to dodge AI detectors.

And the report will be honest about the result. Where Rhizome wins an axis we show it; where it loses one, we show that too. A partial, verifiable win is more convincing than a clean sweep no one can check.