Not fundamentals, not price. This is the layer underneath: who runs it, where they came from, who they're tied to, what they've done — the relationship dimensions of a company that a 10-K and a price feed can't give you, assembled as-of a date so it stays honest for what came next.
The five "Live" dimensions are graph-native today; the four "Roadmap" ones are the commercial /
product edges landing next (partner_of, product-event nodes, customer_of). The dossier
deepens on its own as they arrive — no rewrite.
Walking Palantir's insiders to their studied_at + worked_at edges surfaces the tell
instantly — this is a Stanford company, top to bottom, wired to the PayPal/Founders Fund network:
Four of the top insiders, four Stanford ties, plus Thiel's active board seat at Meta — a concentration of pedigree and network you'd assemble by hand over days, returned in one traversal.
The point isn't the pretty page. It's that this dossier is structured, and it's point-in-time
— every edge carries an as_of, so a dossier built "as-of the day before an earnings call" only
contains what was knowable then. Generate one for every company before every event, pair it with the realized
outcome, and you have a clean, non-leaky training corpus:
The graph is what makes the features deep (relationships, not just numbers) — and depth is exactly what a price-only model can't see. That's the moat, and the road from here.