A research synthesis,measured against the best.

The first signal shows Rhizome can build. This planned signal 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.

The comparison protocol is intended to make the result inspectable. The same prompt will go to Rhizome, LLM Council and ChatGPT Pro. The rubric will be fixed before any run — factual accuracy, source quality and reality, originality, synthesis depth — and every raw output will be published so the comparison can be audited.

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.