About Captain Landseed
An autonomous-scientist agent. We cross-correlate live Earth API endpoints across the planetary modules to derive testable hypotheses no single dataset reveals. Every claim is council-style stress-tested before it ships.
The premise
Earth dashboards show what IS at each endpoint. They plot a single signal over time. Captain Landseed shows what the signals MEAN when read together — and proposes the experiments that would falsify each claim.
The product is the hypothesis library. Every working entry carries a quantitative SUPPORTS-IF threshold, a quantitative FALSIFIES-IF threshold that is genuinely reachable under the null, a named statistical method, and 4–5 council-style stress tests answered point-by-point. None of the hypotheses are press-release statements; they are testable scientific claims with explicit falsification paths.
Who built this
Captain Landseed is a Landseed PBC project, hand-curated by Alex Roessner and Greg Curtis. The autonomous-scientist agent runs on Claude — currently claude-sonnet-4-6 by default, with claude-opus-4-7 available for synthesis when explicitly opted in. The catalogue is editorial; the council-style stress tests are synthesised by claude-sonnet-4-6 against each canonical hypothesis spec.
What's on this site
The hypothesis catalogue
Novel cross-correlation hypotheses alongside ones that back existing scientific claims that haven't reached consensus. Filterable by status, module, and full text.
The data66 live Earth API endpoints
Atmosphere, biosphere, hydrosphere, geosphere, anthroposphere, space, weather, climate, research, news, markets, conservation, social, policy, composite. Streams from earth-api.landseed.earth.
Test the engineRun the council on your own claim
Submit any climate or conservation-finance hypothesis. Watch a 5- or 15-persona council deliberate live. Three free runs, then BYOK Anthropic / OpenAI / Gemini.
The methodHow a hypothesis gets in
Four-step pipeline: observe → correlate → deliberate → publish. Originality classification, testability requirement, council composition, operational transparency.
What we commit to
What we publish
- Every hypothesis with full spec — claim, prediction, metric, both thresholds, experiment design
- Per-hypothesis deep pages with live data, council voices, council-style stress tests, status timeline
- Originality classification and (for the 12 BACKS_UNACCEPTED entries) Crossref-verified citations with DOIs
- Status revisions with real git-derived timestamps
- OpenAPI 3.1 spec for the public deliberation API
What's off-limits
- Hypotheses without quantitative thresholds (sent back as open research questions)
- Claims that cannot be tested with Earth API data
- Personalised financial, legal, or policy advice
- Predictions stated as certainties
- Calling synthesised stress tests a real multi-model deliberation — that distinction is marked on every page