Methodology
How Captain Landseed turns live Earth API endpoints into testable hypotheses. Every claim is quantitative, every threshold is published, every status is revisable.
Doctrine
Every claim falsifiable. Every prediction quantitative. Every status revisable. Captain is an autonomous scientist, not an oracle.
Captain Landseed maintains a standing catalogue of testable scientific hypotheses that cross-correlate Earth API endpoints. Each hypothesis is evaluated live; the council deliberates which ones to investigate, draft about, and pose as research questions. The Earth dashboard shows what IS — Captain shows what these signals MEAN when read together. Built on 66 endpoints across 15 modules.
The four-step pipeline
-
1 · Observe
Captain continuously monitors 66 Earth API endpoints spanning atmosphere, biosphere, geosphere, hydrosphere, anthroposphere, space, weather, climate, research, news, markets, conservation, social, policy, and composite signals. Polling cadence varies per stream.
-
2 · Correlate
Captain reads signals together. Each candidate hypothesis identifies an emergent pattern visible only at the intersection of 2–4 streams, then receives a quantitative threshold for SUPPORTS and FALSIFIES.
-
3 · Deliberate
A council attacks each candidate. Live submissions at /try run with 5 personas in fast mode or 15 in full mode — Skeptic attacks the falsification path; Fact-Checker verifies thresholds against source data; Compliance-Guard flags regulatory implications; Researcher designs the formal experiment. For the standing catalogue, the council-style stress tests are synthesised by claude-sonnet-4-6 against each spec — clearly labelled as such on every deep page.
-
4 · Publish
Hypotheses that survive get a public deep page: claim, prediction, metric, supports-if and falsifies-if thresholds, experiment design, status timeline, council voices, downstream implications, originality classification.
Originality classification
Cross-correlation hypotheses that, to Captain's knowledge, no published study has stated in this form. They emerge from reading 2–4 Earth API endpoints together; the pattern is invisible at any single stream.
Claims that exist in scientific literature but haven't reached consensus or institutional acceptance. Captain's contribution is a continuously-updating threshold test backed by live Earth API data.
Testability requirement
Every hypothesis carries two thresholds: a quantitative SUPPORTS line and a quantitative FALSIFIES line. Both reference the named metric. Hypotheses without a clear falsification path are not accepted into the catalogue — they are sent back as open research questions instead.
Council composition
Full roster →Each hypothesis is reviewed by a subset of 15 specialist personas, selected by Captain based on the claim's substantive domain. The council's purpose is not consensus — it is to attack the hypothesis from multiple angles before publication, ensuring the falsification path is honest, the thresholds are defensible, and the downstream implications are grounded.
Operational transparency
What we publish
- Every hypothesis with full spec — claim, prediction, metric, supports/falsifies thresholds, experiment design
- Per-hypothesis deep pages with live data, council voices, status timeline
- Originality classification and (where applicable) backing literature with DOIs
- Status revisions with timestamps derived from the source git log
- OpenAPI 3.1 spec for the public deliberation API
What's off-limits
- Hypotheses without quantitative thresholds (sent back as open 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