Kortex, in two pages

For the person who approves the engagement. The rest of this site is written for the analyst who will use the platform; this page states what they will be buying, why it is different, and who stands behind it. It prints to two pages.

The problem

Siting a large energy load, a data centre, an electrolyser, an industrial plant, is decided on evidence that is scattered across grid operators, land registries, environmental agencies, planning systems and corporate registers. Teams pay consultants to assemble it slowly, or trust screening tools whose reasoning cannot be inspected. When the decision is later challenged, by an investment committee, a lender, or a counterparty, the evidence trail is usually a spreadsheet nobody can reproduce.

What Kortex does

Kortex screens candidate sites and counterparties against 248 primary sources joined into one queryable evidence graph, and answers with a verdict per candidate: proceed, investigate, reject, or insufficient evidence. Every answer ships as a receipt: a frozen, exportable document carrying the verdict, each finding, the evidence behind it, the source and licence of every fact, and identifiers that pin the exact data state it was computed from. The same screen then runs as monitoring: when the underlying record changes, a connection register updates, a queue moves, a designation lands, the screen re-runs and reports what changed and why.

Why it is different

• It abstains rather than guesses. No composite scores, no black-box weighting. Where the evidence does not support a verdict, the answer is "insufficient evidence" with the missing fact named. Buyers use it to decide where to spend diligence money, not to replace diligence.
• Every answer is auditable to source. The receipt can be handed to a lender, a board, or a regulator, and checked line by line. Exports render from the frozen receipt, never a recomputation.
• There is no model to validate. The answer path is deterministic: database reads and disclosed arithmetic, versioned publicly. No LLM, no machine-learned scoring. For buyers with model-risk frameworks, Kortex classifies as a data tool, which materially shortens review.
• Honesty is structural. The platform publishes its own misses (13 validation studies, including where coverage was wrong), its per-source licence position including sources still under review (/licensing), its coverage gaps (/coverage), and its measured uptime (/status). A reviewer does not have to take our word for anything on this page.

The evidence, briefly

12,708,136 graph nodes and 3.88B relational rows across power, grid, land, water, hazards, markets and corporate ownership. Depth is deliberately uneven and disclosed: Great Britain and Europe at screening-decisive grain (operator-published grid capacity, connection queues, planning pipeline, land parcels), the United States congestion record complete across all seven organized markets, corporate ownership global. Validation is published, including blind adjudication of the hardest judgment calls and comparisons against official federal registries, with the misses left visible.

Who is behind it

Governance surfaces are published, not promised: /trust (security posture, stated plainly including what does not exist yet), /data-protection (lawful basis and the complete named subprocessor list), /status (recorded per-minute uptime). Continuity is contractual: everything exportable in open formats, receipts readable after contract end.

Engagement

Kortex is sold through fixed-term pilot programmes: scoped screens of your actual candidates, delivered as receipts, with monitoring through the pilot term. Pricing is metered and machine-readable; pilot terms including confidentiality and data processing are agreed per customer. The fastest way to evaluate it is the worked example: a completed screen, its receipt, and the re-run that shows two verdicts flipping when new register data landed, which is the monitoring product demonstrating itself.

Contact: kortex@orkora.com · kortex.orkora.com