COI Engine | Nyman Intelligence

Seed + resolved director

Seed entity: NYMAN INTELLIGENCE LIMITED (CH 17049046)

Resolved director: Rickard Nyman (merged from 5 CH officer records)

Both shown as red nodes in the graph. The engine starts from the seed company, resolves the director identity, then expands the network through co-directors and their other companies.

Network statistics

…

seed / resolved person company

Headline finding

Entity resolution surfaces 5-into-1 collapse

A naive Companies House search for the target's named director returns 5 separate officer records. All five share DOB 1988-03 and overlapping name tokens.

Without resolution: 5 different people, 1 directorship each.

With resolution: 1 person, 5 directorships across 13 years.

Click the red node in the graph to see the underlying records.

Address kinship signals

Click any name to find them in the graph.

Free-source cross-references

OpenSanctions ▶

17 namesake matches, 0 confirmed. Entity resolution prevents false positives.

Streamed the bulk targets dataset (~460 MB, free for non-commercial) and grepped for our 17 network surnames. Surfaced 17 namesake entities — none were actual network members. Most striking false-positive candidate: a "John Michael Ormerod" (b. 1950-11-03) appears on the Ukraine NSDC sanctions list. Without entity resolution (DOB + first-name + nationality), a naive name match would flag him as Paul Andrew Ormerod. Production: API call €0.10 for live lookup, or weekly bulk-download for offline matching.

ICIJ Offshore Leaks ▶

0 hits across 5 names tested.

Public search at offshoreleaks.icij.org is behind anti-bot protection, so programmatic batch lookup needs browser automation. Tested across the network: zero result rows for Rickard Nyman, Paul Ormerod, Periander, Diphrontis, and Volterra. For production: ICIJ provides bulk CSV downloads of all leak datasets (Panama Papers, Paradise Papers, Pandora Papers, Bahamas Leaks, Offshore Leaks). Loaded into a local fuzzy-lookup index for instant matching against any seed.

Wikidata ▶

5 of 6 level-1 people enriched with role descriptions.

Free SPARQL endpoint, no auth. Queried each named individual; returned descriptions like "English economist" (Paul Ormerod, Q604474), "British psychoanalyst" (David Tuckett, Q57079818), "British politician (born 1967)" (Lorna Fitzsimons, Q545964), "Chairman, Isle Group" (Piers Clark, Q115735404), and "researcher" (Rickard Nyman, Q114431355). Click any of these people in the graph to see the badge. In production this enrichment runs over the whole network and tags each person with their public-record description.

OpenCorporates ▶

Free API tier removed. Now paid-only for commercial use.

Confirmed via direct API call: HTTP 401 Unauthorized without a paid API token. Free tier appears to have been retired since the original Phase 1 source list. Production options: Phase 2 paid tier (£2,250/yr for 500 API calls/month per Artem's anchor), or substitute with NorthData (€9k/yr) for European corporate coverage. UK Companies House continues to cover the UK case fully for free.

Company-website scraping ▶

1 of 16 found via domain-guessing. Phase 2 limitation.

Tested two approaches on a subset of the network. Domain-guessing (companyname.com, companyname.co.uk, hyphenated variants): 1 hit out of 16 (Volterra Partners LLP at volterra-partners.com). DuckDuckGo HTML search: returned mostly UK business-directory aggregator sites (bizseek, checkfree, bizdb) rather than company sites. UK SMEs frequently have no website at all, or use unusual domain patterns. Phase 2 fix: paid search API (Google Custom Search, Bing) or a layered lookup chain (Companies House registered-website field → LinkedIn company page → domain heuristic → web search).

Methodology

Sources queried for this target

UK Companies HouseReturned data

Wikidata SPARQL5 hits

OpenSanctions (bulk)17 namesakes, 0 confirmed

ICIJ Offshore Leaks0 hits

SEC EDGAR (US)No relevant data

DuckDuckGo + domain heuristics1 of 16 hits

Phase 1 — to integrate

Aleph (OCCRP)Planned

192.com (electoral roll)Planned

Phase 2 — paid enrichment

NorthData (Europe)Phase 2

OpenCorporates (paid)Phase 2

Pipeline

IngestPull officers, PSCs, addresses for the seed and recursively expand through co-directors.

In demo

ResolveMerge multiple registry records for the same individual (DOB + name-token + address overlap).

Hand-curated

ExpandBipartite graph traversal to depth K. High-degree nodes (corporate secretaries, large partnerships) are capped to prevent graph explosion.

In demo

Cross-referenceIntersect the network with the client watchlist. Surface flagged conflicts.

Phase 1

OutputCSV (preserving the analyst's existing workflow) and interactive graph view, with intersection-flag column highlighting hits.

In demo

What this prototype is — and isn't

This page is a one-day spike on a single test target to validate the engine shape against real public data. Every data fetch was ad hoc — done once, against live APIs, with no caching, no checkpoint recovery, no retry-on-rate-limit, no schema validation. It is not a working Phase 1 build. The Phase 1 build needs substantive engineering on, in rough order of effort:

Robust data-fetch layer — checkpointing per seed and per source; on-disk caching; retry-on-rate-limit; schema validation; structured logging. The current spike has none of this.

Phase 1

Entity resolution at scale — turn the rule-based DOB+address+name-token heuristic into a calibrated probabilistic matcher with confidence scores.

Manual only

Persistent batch pipeline — handle hundreds-thousands of seeds. Differentiator vs ad-hoc agents.

Phase 1

High-degree stop-rule tuning — caps shown are rule-of-thumb (<10 companies per person, <5 officers per company). Real version learns from corpus-wide degree distributions.

Rule of thumb

Watchlist ingestion + intersection engine — needs to handle the client's actual format, support fuzzy-match thresholds, emit per-conflict provenance.

Phase 1

Source adapters for the planned sources — Aleph, 192.com (and any extras). Each takes a day or two and ongoing maintenance as endpoints change.

Some done

Operational tooling — alerting on source breakage, dashboards for run health, cost monitoring.

Phase 1

The visualisation, source-tooling pattern, and methodology you see here transfer directly into Phase 1. The 8-12 day estimate accounts for all of the above on the agreed scope (UK-focused, free sources, single-target volume).