The TEA Commons is building the authoritative data layer for evaluating the commercial viability of novel climate and industrial technologies.

THE PROBLEM

Scientific discovery is accelerating but breakthroughs rarely translate to real-world impact

  • Two innovation pipelines that are producing extraordinary science / innovation:
    • ~40,000 early-stage startups, ~500,000 applied researchers
    • AI-for-discovery efforts generating millions of candidate materials, catalysts, microbes
  • Commercial viability is almost always the dealbreaker - bottleneck is manual assessment
    • Not enough coaches to help individual teams; scientists are poorly equipped to do this analysis
    • No team can manually evaluate an effectively infinite stream of AI-generated candidates
  • The bottleneck in agentic AI has shifted from "how sophisticated is this task" to "can you supply the ideal context for the task to succeed"?
    • We can build AI models to evaluate viability but its conclusions are only as reliable as the industrial context and assumptions that anchor them.
    • AI can look up a number, but there's no adjudication of whether that datapoint is 10% off or 10x off, leads to compounding & unassessable errors
THE GAP

The ground truth data layer for commercialization is missing

  • Historical precedents for data projects
    • ImageNet helped accelerate computer vision, neural networks
    • Decades of structural data in the Protein Data Bank enabled breakthroughs in protein-structure prediction (AlphaFold)
    • PubChem made chemical information broadly accessible for computational research.
  • There's still no PubChem for commercial viability.
    • Industrial innovation needs an analogous public-good layer: expert-validated data on processes, equipment, costs, backed by auditable data.
    • Much of this information already exists, but it is fragmented across engineering studies, reports, and practitioner knowledge.
  • Having this data enables:
    • Entrepreneurial scientists to build credible techno-economic analyses, drive decisions
    • AI-for-discovery programs to down-select candidates with a plausible path to deployment.
OUR THESIS

Our thesis: start with most common data needs, for largest industries, and build out from there

  • Our experiences give reason to believe the problem is tractable / the dataset scope is bounded
    • We've seen which assumptions repeatedly block innovators (100+ teams), which sources tend to be trustworthy, and what level of accuracy is sufficient for an early go/no-go decision
    • Roughly 500–1,000 reference classes could cover the major industrial processes and configurations most relevant to climate and industrial innovation.
    • Intentional about prioritizing rough accuracy over precision (+/- 30% target, FEL 2) - early-stage journeys have massive uncertainty, even ballpark analyses can inform discovery, eliminate obvious dead-ends
  • Uniquely enabled by a non-profit, public-goods approach:
    • public / philanthropic funding - development of public goods increases leverage of AI-for-science and R&D efforts around the world
    • Neutral, shared infrastructure facilitates coalition / collaboration between industry, academia, government
    • Individual startups / research programs don't have incentive / resources to collect and validate data