Methodology
Published metrics carry a date and source. Figure panels also include a plain-language reading line. This page documents where the data comes from, what is computed, and how corrections are handled.
Automated ingestion starts once a day at about 05:07 UTC. A source change cannot appear until the next successful check and deployment. Weekly sources are considered for an update during that same daily job.
Sources
Forms 6-K, 20-F, 424B5, FWP, 13D/G, 3, 4 and 144. Structured 13D/G and insider XML are fetched when the daily check finds a new filing.
Annual 20-F facts only. RUB-era facts before 2023 are excluded and restatements are deduplicated.
Used for the hiring radar. Free-text locations are normalized with a hand-maintained map.
Availability zones and incident history.
Active model catalog changes and launch lag. The endpoint is public and needs no account or API key.
Attributed public list prices. A failed parser leaves the last value visible with its date; history gets a point only when the published range changes.
Official twice-monthly settlement snapshots from the public query endpoint. Daily short-sale volume is not used as a substitute.
CUSIP-matched rows from the SEC quarterly data sets. The figures are delayed and do not represent total or real-time institutional ownership.
Append-only reviewed ledger. Each target needs a public source and independent corroboration; shown as a tracked sample, never complete consensus.
The browser cache lasts five minutes. The edge cache lasts 15 minutes during US market hours and one hour outside them. The displayed quote always includes its market timestamp.
Reviewed figures are stored as versioned JSON. Their source links and dates remain attached.
Derived metrics
Derived values are computed when the site snapshot is built. A missing input renders as —. A stale input stays visible with its original date and a stale marker.
Adapter health
edgar-filings
edgar-xbrl
finra-short-interest
greenhouse
pricing
sec-13f
statuspage
tokenfactory
Corrections
Manual data lives as versioned JSON. Published factual corrections are appended to the correction history instead of silently replacing the old claim. Parsed automated data and run logs are versioned too. Raw fetch bodies are retained as CI artifacts for 14 days so parser failures can be investigated.
Spotted an error? Send the stronger source through the corrections page.