Applications are what customers buy and operate; the insight pages are published research proving the same graph discovers non-obvious relationships. Kortex leads with one application, large-load siting, and every additional workflow below runs on the identical substrate: the same identities, topology, market series and evidence discipline. A relationship validated for one application strengthens all of them. The substrate underneath is the full estate; 541 governed datasets from 152 registered sources, refreshed continuously, every one licence-tracked and retrievable (coverage register).
Eliminate weak data-centre and large-load candidates before commissioning utility, engineering and property studies. Bring a shortlist; Kortex re-ranks it against installed dispatchable low-carbon capacity, EHV topology, wholesale price, water stress, queue evidence (where it exists, zones without it say so) and counterparty ownership, every score traceable, every rejection reasoned, the whole answer receipted.
"Before we buy this plant, portfolio or utility; what would we wish we had known?"
Fleet composition on the canonical operational universe, revenue-mix exposure, cooling-water dependencies (validated US subset), queue history where the assets sit, and the ownership chain to the ultimate parent; as one screen instead of five datasets.
"Who ultimately owns this infrastructure; and what else depends on the same counterparty?"
GLEIF-anchored entity resolution with blind-validated precision (96.7% family / 73.3% exact-entity on the manual tier, published), parent chains checked against issuers' own Exhibit 21 filings, abstention by default on weak stitches; a confident false identity is worse than none.
"Our assets are diversified on the map. Are they diversified in physics?"
Common-mode exposure a map hides: cooling-water systems that fail in the same drought, weather positions that co-move, capture-rate erosion as penetration rises. Weather concentration is research-preview until measured correlations land; and is labelled so.
"What does this grid actually pay, hour by hour; and what gets built in it?"
The market-timing suite as one product: a grid's day (capture prices per fuel), hourly clean-supply structure, a decade of capture-rate decline, and queue-cohort reality; observed series, deterministically joined, receipted.
"Give our agents ground truth about the physical world; with receipts."
The same graph as an agent surface: MCP server, deterministic answers, per-metric evidence classes, frozen retrievable receipts and fail-closed rights modes. Built for autonomous workflows that must cite what they claim. Early; the contract is real, the tooling around it is still growing.
Worked results below were verified against the linked endpoints as of the 2026-07-18 snapshot. The graph refreshes continuously, so figures drift between snapshots; the endpoints always return current values with source provenance.