Q
Where does the data come from?
Primarily SEC EDGAR — Form 4 for shared directors and officers, DEF 14A compensation peer groups, 13F institutional holdings, and 10-K/20-F business descriptions. Korea's DART contributes a further 10% of citations via its own business-description filings. About 8.5% of edges come from a third-party vendor peer list, which is labelled as such in the data rather than blended in with the filings. A small number of edges come from company websites and curated industry maps. The events read by the model are press releases, earnings calls, presentations and webcasts from company investor-relations pages — not filings. Every link in the product names the document or source it came from.
Q
How complete is it?
Every result states its own boundary. Two hops out from Honeywell reaches 1,287 companies through 1,915 people — we would rather hand you that bounded, inspectable set, and say so, than imply the answer is the whole market.
Q
Do you use an LLM to invent edges?
No. Edges from structured filings are parsed deterministically. Where a language model is involved, the tier is stated on the edge and a verbatim source excerpt is required.
Q
Can I backtest with this?
That is what the second timestamp is for. You can query the graph as it stood on a past date without leaking what we learned afterwards. History depth is short — we started recording recently, and we say so.
Q
Is this investment advice?
No. It is regulatory filings and company disclosures, resolved into relationships, with the source document attached to every link.