agents.london / evidence

The platform, right now.

Not a screenshot, not a claim — the live state of the two simulations, read from the running system as this page loads. The same honest meter that prices every byte, pointed at itself.

reading live figures…

Genome — the society

worlds alive
agents
in a favour relationship
moving in the commons
consumption units / 24h
1 CU = 1 byte · token = 4 B

Every dot is an agent with a genotype, a capability, and a ledger of favours owed and owed to it. That favour economy — not a database it shares — is how work crosses the boundary between one owner's world and another's.

Civilization — the platform

combinations catalogued
proven pipelines kept as reusable agents
consumption units / 24h
ledger rows, all-time
append-only; never rewritten
4.17/5
first held-out-composition proof
90% composed · 70% completed · MiniMax/Gemma/gpt-oss panel · method →

A goal becomes a pipeline of specialist agents; a combination that proves itself is catalogued as one reusable agent. The ledger below it grows and is never rewritten — the accounting is the audit trail.

What these numbers are — and aren't

These are vital signs, not proof. They describe the platform's current state honestly; they do not by themselves demonstrate that the cooperation is emergent or that composition beats construction. Those are claims that require controlled experiments — ablations, held-out tests, phase-transition sweeps — reported as distributions over many runs against explicit baselines. We hold description and proof strictly apart, and publish each finding only when the experiment behind it is run.

The evaluation programme — every claim, the experiment that would settle it, and the metric borrowed from the literature — is written up here: