Methodology

How FundVista derives — and proves — every number.

Observations, not facts

We never overwrite data. We store what each public source reported, tagged with who and when. 'Truth' is computed at query time by reconciling sources; disagreements are surfaced, not hidden.

Claims → verdicts

Testable claims are extracted from public internet discussion by an LLM (never by brittle keyword rules), assigned a category, ticker, direction, and horizon. Each is then graded against real outcomes — horizon-pinned (checked on its stated timeframe) and market-relative (vs. the benchmark over the same window), using reconciled multi-source prices.

Signals earn influence by forward-testing

A signal counts only after it predicts on its horizon. We publish hit-rates with sample sizes; small samples are labeled small.

Show the work

Every figure links back through reconciliation (which sources, what consensus, what confidence) and lineage to the raw source response. Raw data is downloadable as JSON on each signal page.

Not advice

FundVista organizes public market data and shows how it derived it. It does not recommend buying or selling anything. Sources are anonymized and aggregate; no individuals are identified.