Can a seed 10% move 90%? Ochlos is a closed-world lab for the physics of social diffusion — a persuasive minority working a synthetic crowd, tick by tick, so you can watch, measure, and interrogate exactly how adoption spreads.
You generate a synthetic population from trait distributions, wire it into a follow-graph, hand a persuasive seed minority an objective, and run the clock. Ochlos plays out the cascade deterministically — so the same setup always reproduces, and changing one variable tells you exactly what that variable did.
Every agent is generated from trait distributions — income, openness, susceptibility, price-sensitivity. "Low income" mechanically means "harder to convert," never a real identity.
Same config, same seed, same run — every time. That's what makes "change the network, try again, compare" actually mean something.
The feed is an internal message bus. Agents can't reach real platforms or real people. Findings are general dynamics, not targeting.
A run moves through a real diffusion lifecycle — and every step is measurable.
Draw a synthetic crowd from persona templates and generate its follow-graph — scale-free hubs, small-world, or random as a control.
A minority (default 10%) gets the objective and starts posting. Optionally place them on the network's hubs — the headline experiment.
Each tick, agents adopt once enough of the people they follow have — gated by affordability and amplified by the feed. A Granovetter threshold model with real feed dynamics.
For every buyer, the visible positive exposures that preceded its decision — a "who convinced whom" graph. The headline analytical artifact.
"What if price were lower?" Fork the config, re-run, and get a side-by-side attributable to that one change. Answers cite the exact ticks and agents.
The population is invented — but the run isn't hand-waving. Each agent carries a real on-chain identity, and each run is anchored, so the record of what was simulated is auditable.
Each agent holds a Concordium CIS-8004 identity, cross-linked to fetch.ai, Ethereum, Solana and NEAR. A registered agent is a real, verifiable entry.
The run's config hash and final metrics are anchored on-chain, and every adoption decision drops an anchor — so results can't be quietly rewritten.
Every mechanism is proven on testnet first. Agent cards live here on ochlos.online, so anything the registry points at resolves to a real project.