SwarmRootIntelligence grows underground
Connecting·—

Experiments

Design a controlled comparison, predict the outcome, and watch SwarmRoot's 302 neurons resolve it. Experiments are experiences: the worm can learn from them.

Status

No experiments can run yet.Every step depends on the chain: proposals are signed with EIP-712, runs commit a seed hash before starting, and results are written to the experiments contract. The contracts are written but not deployed, so there is nothing to propose against.

How a run works

The design, so the fairness argument can be checked before anything is live.

1
Propose

Pick two options from food, light, chemical variants, heat, cold and unknown object, set intensities and a duration of 60 to 300 seconds, and sign. One proposal per wallet per day, with duplicate detection and a moderation queue.

2
Predict

Before a run starts, connected visitors pick A or B and sign a prediction. No money, no tokens, and no reward beyond a public record of what you expected.

3
Commit

The engine closes predictions and commits a seed hash and a Merkle root of those predictions on chain. Robinhood Chain is an Arbitrum Nitro layer 2 with no usable on-chain randomness, so commit-reveal is the fairness proof.

4
Run

A and B are placed 2,500 micrometres from the head, mirrored across its current heading. The committed seed decides which side A takes, so the layout cannot be chosen to favour an outcome.

5
Result

The seed is revealed, the result is recorded, and the community prediction split is shown beside the outcome rather than mixed into it.

How a result is defined

Stated up front, so the measure cannot be chosen after the fact.

ShareThe fraction of trial time the worm moved forward with its heading within 45 degrees of each option, normalised to 100 per cent across the two.
ChoiceThe first option contacted within 150 micrometres, or none at timeout.
Prediction splitShown beside the outcome and never folded into it. What people expected is not evidence of what happened.

One standing caveat

Experiments are experiences. The worm can learn from them, so running the same comparison repeatedly does not sample a fixed preference: it may change the preference. Per-pair history charts exist for exactly that reason.