From collision to brief
Eleven steps. Two pipelines: L0 generates the hypotheses; the post-🔥 stage validates the strongest ones in depth.
Collision
~100 collisions/day500 scientific domains. A random draw collides two of them.
Filter
30-60% passA filter rejects trivial or already-explored pairs.
Synthesis
~70% bridgesAn LLM generates a causal bridge between the two domains.
Critique
5 scores producedA devil's advocate and an angel's advocate debate.
Selection
Top 15%Top 15% retained on composite score.
Review
~15-20% 🔥Calibrated verdict: rejected / interesting / to test 🔥.
Literature grounding
Post 🔥5+ references8-12 Semantic Scholar queries. Zero hallucination — every DOI is verified.
Hypothesis sharpening
Post 🔥Rigorous formulation with variables, quantitative predictions, H0, statistical tests.
Protocol
Post 🔥Three phases: in silico (€0-2k) → minimal (€2-15k) → full (€15-200k).
Panel (5 reviewers)
Post 🔥Methodologist, domain expert, devil's advocate, industry reviewer, funding strategist.
Research brief
Post 🔥Publication-ready document of 4-6 pages. Markdown + JSON.
The funnel in numbers (100 typical collisions)
Across SPORE: 4,425 collisions explored → 41 briefs published.
Three non-negotiable principles
SPORE proposes; humans dispose
The system generates and validates; it does not replace human judgement. Every brief is an invitation to think, not a conclusion.
Zero bibliographic hallucination
Every reference is verified through Semantic Scholar. If a DOI does not resolve, it is dropped. Citations are the foundation.
Five reviewers, five perspectives
Methodologist, domain expert, devil's advocate, industry reviewer, funding strategist. No blind spot. Every doubt is voiced.
Ready to explore?
Browse the briefs →