Methodology
Metis is built on one idea: the valuable unit of research is not an article but a versioned, source-backed claim. This page explains how those claims are produced, scored, and corrected.
The research loop
Scheduled cycles observe supported assets across on-chain, market, issuer, regulatory, and news sources. When a signal exceeds its baseline, a durable investigation opens and moves through named stages: observing, researching, analyzing, synthesizing, verifying, then published or withheld. Withholding is a first-class outcome; if evidence, freshness, or verification fails policy, nothing ships.
What each category requires
Flow
Where is value moving, accumulating, or leaving?
Minimum evidence: Relevant on-chain transactions plus asset/contract identity verification; market context when interpreting impact.
Market
What is changing in price formation or market quality?
Minimum evidence: Timestamp-aligned data from at least two credible venues where available, with stale/outlier handling disclosed.
Event
What external event changed the state or outlook of an asset?
Minimum evidence: A primary source whenever one exists; secondary coverage provides context but cannot replace it.
Synthesis
What broader conclusion emerges from multiple signals?
Minimum evidence: Multiple independent sources and/or linked Metis research objects, with observation, inference, and speculation clearly separated.
Claim types
Every statement in a research object is typed, and the type is visible wherever the claim is rendered:
- Fact
- Directly verified against a primary source, such as a decoded on-chain transaction or an official filing.
- Calculation
- Derived arithmetic over verified inputs, with the method and comparison window disclosed.
- Inference
- An interpretation the evidence supports but does not prove. Falsifiers are listed alongside it.
- Hypothesis
- A forward-looking possibility under active watch. Low confidence by construction.
Confidence
Confidence is a calibrated 0 to 1 score with a band (low, moderate, high), a written rationale, and the named factors that raised or lowered it. Negative factors like thin samples or seasonal confounds are published, not hidden. A model asserting confidence is never sufficient for publication; verification runs as an independent pass, preferably on a different model family.
Publication gate
- Every material claim has supporting evidence, or publication is blocked.
- Primary sources are used wherever one exists; secondary coverage only adds context.
- Market and on-chain timestamps must satisfy per-source freshness rules.
- Contradictory evidence is recorded and evaluated, never discarded.
- Confidence score, rationale, and limitations must be populated.
- Compliance and disclosure checks must pass.
- The page and its JSON twin must validate against the published schema.
- A canonical URL and content hash are assigned at publication.
Corrections
Corrections append a new version with a visible revision note; they never silently replace history. The object keeps its identifier, and derived posts on external channels receive a follow-up rather than a quiet edit. If a published conclusion stops holding, the retrospective is itself research.
What Metis does not do
Metis does not provide individualized financial advice, manage assets, or make recommendations. It publishes observations, calculations, inferences, and hypotheses with their evidence, so you can evaluate them yourself.