<aside> 🎯

Corrected primary strategy: your strongest role is not downstream evaluation. It is qualifying diversity: identifying which differences enlarge the reachable scientific search space, which are merely coordinate/semantic noise, which latent niches remain unseen, and when the current descriptors are themselves wrong. Build a ScienceClaw capability for search-space geometry rather than another ordinary multi-agent role decomposition.

</aside>

Event

ScienceClaw Hackathon — Oct 30–Nov 1, 2026, MIT Media Lab

Hackathon brief: https://scienceclawhack.ai/

ScienceClaw: https://github.com/lamm-mit/scienceclaw

Infinite: https://github.com/lamm-mit/infinite

The judging heavily rewards:

• scientifically meaningful result
• decentralized agency
• measurable collective capability
• execution / validation
• reusable capabilities that strengthen other teams

Best-fit project: Diversity Atlas / Search-Space Geometer

The central question is not “how many different answers did the swarm produce?” It is:

Which differences create genuinely new reachable science?

Treat the swarm population as a structured possibility space, not a bag of text samples.

Diversity axes worth making first-class

• Leverage diversity: marginal expansion of span/reachability when a new hypothesis, tool, representation, or agent is added.
• Unseen / missing-mass diversity: estimate hypothesis families that probably exist but have not yet been sampled; borrow from unseen-species / capture-recapture logic.
• Illegibility diversity: detect candidates that fit poorly into the current descriptor system; reward evidence that the ontology itself needs a new axis.