Holding the first signals back — on purpose
Rhizome can already finish real software, end to end. Before we publish the first Early Signals, we're tuning the one trade-off that decides everything — coordination against execution.
We could publish the first Early Signals today. The runs finish — Rhizome takes a specification and carries it through, specified, implemented, tested, and integrated, without a human driving each step.
We're not publishing yet. Not because the runs don't work — because of where the balance between coordination and execution still sits, and what it costs.
The trade-off underneath everything
Every multi-agent system lives between two failure modes. Too little coordination and the swarm fragments — duplicated work, contradictory changes, no shared thread. Too much coordination and it seizes up the other way: everyone aligns, everyone defers, and nobody commits to doing the work. Rhizome's whole design — tensions, attachment by fit, minimum tenure, the policy engine — exists to hold the productive middle between those two.
What we're debugging right now
At the moment the field leans too far toward coordination. We see a diffusion of responsibility: agents fractally delegate — re-scoping a tension, handing it to a peer, negotiating, waiting, handing it back. Each move is locally reasonable; the sum is a kind of elaborate procrastination where a lot happens and little lands. Plenty of motion, not enough committed execution. It's the exact failure the attachment scoring and tenure rules are meant to suppress — and tuning them is what we're on now.
Why we wait
Because “it works” and “it works efficiently” are different claims, and the gap between them is money. The same finished product can cost wildly different amounts of tokens depending on how cleanly the field stops coordinating and starts building. Before we show you a signal, we want the configuration tuned for real token-efficiency — cost per finished product — not a flattering one-off.
What this costs us
This is what evidence over narration actually costs. We could ship an impressive-sounding run tomorrow. But a signal is meant to be receipts, not adjectives — the real process and the real bill, including where it's still rough. Publishing something unoptimized would undersell the system; cherry-picking a lucky run would be dishonest. So we tune first, then publish, and show the numbers either way.
The signals are close. When they land, you'll see exactly what a Rhizome run costs — and exactly how it got there.