NEO Intelligence

NEO Intelligence is designed and specified. It has not launched [D] (neo-intelligence, S4). Every capability described on this page is design intent at that status. None of it is a description of a running service, and nothing here is a measured result.

The name is reserved: NEO Intelligence denotes the investigation vertical and nothing else. The platform is NEO System Intelligence, and its market-facing name is the NEO Intelligence Operating System [D].

Why the investigation vertical is the one that is also a commercial form

Not because it is more important than the others — several are important. It is the only one whose shape differs from the platform's default in all three of the ways that matter [A]:

• Its unit of value is the case, not the mission: a subject, an evidence graph, a graded finding set and a defensible report — a bounded set of missions with a shared subject and a single acceptance event.
• Its cost structure is dominated by premium data and human review, not by inference. Premium data is passed through at cost plus a stated orchestration margin and reported separately, because blending it makes gross margin a mix artefact.
• Its liability profile is different in kind, because the output is a finding about a real person or entity.

It is commercially separate and technically identical: the same 17 components, the same execution contract, the same stores. What differs is the price object, the cost mix and the contract [A].

The operating perimeter, which is contractual rather than a feature flag

• Lawful access only.
• Human review on high-impact conclusions.
• Confidence is never presented as certainty.
• No final legal or sanctions determinations. The system produces evidence and a graded finding. The determination belongs to a person with the authority to make it.

The designed capability set

Each item is design intent at S4 unless marked otherwise. None is a launched feature.

Open-source collection. Scan-issuer sources are genuine primary sources — the issuer performed the scan. Web-crawl corpora are not a licence: it is in a public crawl confers nothing about the crawled pages, and this is the single most commonly mishandled source class in the industry [E]. Leaked-material aggregators raise a question of whether a commercial platform may process them at all, which is recorded as a legal decision rather than as a licence question [O].

Blockchain and wallet intelligence. Chain state is the cheapest primary source in any vertical: a node returns the actual record. Everything above it — labels, attribution, risk scores — is inference sold as fact, and the major forensics vendors are professional aggregators producing criminal-adjacent inferences about identifiable persons [E]. The design consequence is that a label is carried as an aggregator assertion with its provenance attached, never merged into chain state as though it were chain state [A].

Anti-money-laundering, sanctions and politically-exposed-person screening. The statutory lists are free and are the authoritative sources; commercial vendors sell coverage breadth, adverse media and matching, not authority [E]. The whole vertical is special-category data, and the sanctions and PEP sources carry criminal-offence-adjacent data. A false positive is a consequential harm to a real person, so match evidence is retained with the screening decision — a retention requirement that runs against the usual minimisation instinct and is modelled explicitly in the Provenance Passport [D] (data-plane.provenance-passport, S4).

Corporate intelligence and due diligence. Anchored on free statutory registers. The commercial sources in this space define the lane a research product must stay inside: analysis and aggregate statistics permitted, underlying data not redistributable, everything expunged within days of termination [E]. Deal-intelligence and rumour content carries a market-abuse exposure rather than a licensing one, and is flagged as such [E].

Entity resolution. Deciding whether two records are the same company or the same person is owned by the Knowledge Graph, not by the Data Plane: the Data Plane hands over the source, the record and its Passport [D] (brain.knowledge-graph, S4). The vocabulary is explicit that this component is the Knowledge Graph; the earlier name for it is retired.

Evidence graph. Each assertion, the evidence for it, and the path back to its origins [D] (brain.knowledge-graph, S4), served and graded by the Evidence Layer [D] (brain.evidence-layer, S4) over an append-only, hash-addressed Evidence Ledger [D] (brain.evidence-ledger, S4).

Circular corroboration detection. Three sources that agree because they are the same source are one source. Independence is a property that must be established, not assumed from count [A]. A price aggregator owned by a large exchange is not independent of that exchange for corroboration purposes [E].

Timeline reconstruction. Point-in-time is the field that decides whether a legal or financial reconstruction is honest: what the rule said on the date of the decision, not what it says today [E]. Sources that cannot answer as-of a date are recorded as unable to, rather than used as though they could [A].

Monitoring. Re-running a case against changed sources, and computing in proportion to what changed rather than from scratch [A].

Reporting. A defensible report is the deliverable: the finding, its grade, the evidence behind it, the provenance of that evidence, and the record of who accepted it [D] (mission-control, S4).

What is not claimed

No source is integrated. No data contract is signed. No count of cases, customers or findings is published. The vertical stays counted at zero in the Program's projections — not for conservatism theatre, but because its connector supply is roadmap and its data licensing is a legal decision, and a revenue line that depends on unsigned data contracts is a forecast of someone else's decision [A].