Entity intelligence & official-source verification

Is this recruiter licensed? Is this employer flagged?

DueCare checks recruitment agencies, employers, and the companies behind them against official government registries — licensed-agency registers, debarment lists, and forced-labour registries. This is research tooling: it produces proposed findings for curator review, the same one-way public source → vetted pack lifecycle as our context packs. It is not a live worker-facing lookup, and no worker case is ever an input.

01 · Coverage

Official registries, four continents.

Every count is reproducible from the registry it cites — real, not estimated. Coverage grows by curator-reviewed proposal; new registries are config rows, not bespoke code.

Official registries

23 config specs + 11 deterministic resolvers

1,111

Catalogued sources

95 countries · 18 industries

532

Support organisations

helplines · shelters · legal aid · unions

37,566

Entities clustered

→ 27,810 deduped across sources

02 · How it works

Official registry → resolve → link → proposed finding.

Field maps are authored against the live registry response — never guessed. A source that blocks or returns nothing parseable is skipped and logged, never fabricated.

Public source step · 01

Official registry

A government licensed-agency register, debarment list, or forced-labour registry.

e.g. US CBP WRO · UK Modern Slavery

→

resolve

Deterministic step · 02

Resolve & normalise

Config-driven resolver fetches + parses the source into canonical entity records.

registry_spec · csv/json/html/pdf

→

link

Linkage step · 03

Link on the LEI

Probabilistic record-linkage joins the same company across registries on the GLEIF Legal Entity Identifier.

splink · GLEIF LEI · ownership edges

→

review

Human gate step · 04

Proposed finding

A curator reviews matches and sources before anything reaches a worker-facing surface.

propose-only · curator review

▸ Findings stage to a review queue, never the live model or a worker's screen directly. The lifecycle is one-way: public registry → reviewed finding.

03 · Registries

A sample of what's covered.

Forced-labour and debarment registries plus corridor licensed-agency registers and global corporate ownership. Counts are the live entity totals.

US CBP · WRO

Withhold Release Orders

Forced-labour import bans & findings.

forced labour59

US DHS · UFLPA

UFLPA Entity List

Uyghur Forced Labor Prevention Act designations.

forced labour144

UK · Home Office

Modern Slavery Registry

s.54 transparency statements w/ disclosed risks + ILO indicators.

transparency11,718

PH · DMW

Licensed recruitment agencies

Philippine licensed overseas-employment agencies + status.

recruitment3,788

BD · OEP / BMET

Licensed recruiting agencies

Bangladesh licensed overseas-employment recruiters.

recruitment2,834

World Bank

Debarred firms

Cross-debarred suppliers (MDB sanction).

debarment1,241

AfDB

Debarred entities

African Development Bank debarments.

debarment1,334

GLEIF

Legal Entity Identifiers

CC0 global corporate identity + parent/child ownership — the cross-registry join key.

ownershipglobal

+ 26 more

Corridor & corporate registries

TW, SI, BR, CO, SG, HK, AU, CA, NO, OHCHR, US DoD & OFAC, GLEIF, OpenOwnership, and more.

34 totalfull map →

04 · Provenance

Every entity traces to an official source.

Licensing is explicit per source. We consume only public / open-licensed data, respect each site's terms, and embed no secrets.

Source	Publisher	Type	Licence	Status
us_cbp_forced_labor	US Customs & Border Protection	Forced-labour bans	US public domain	Resolved
us_dhs_uflpa	US Dept. of Homeland Security	Entity list	US public domain	Resolved
gb_modern_slavery	UK Home Office	Transparency registry	OGL v3	Resolved
gleif_lei + rr	GLEIF	Corporate ownership	CC0	Resolved
openownership_bods	OpenOwnership	Beneficial ownership	CC0	Resolved
doj_press · dol_whd	US DOJ · DOL	Prosecutions · labour enforcement	US public domain	Feed

The full connector + registry map, with per-source counts and the licence ledger, lives in the repository at docs/entity_intelligence_pipeline.md.

05 · Relationship graph

Entities, linked across sources.

Every connector writes into one canonical edge — subject, predicate, object, source, weight — so ownership, control, registration, and same-entity links from different registries form a single propose-only graph (scripts/entity_edges.py).

Edge type	What it links	Emitted by
parent_of	parent company → subsidiary	gleif_rr · GLEIF Level-2 RR
owns_or_controls	beneficial owner → company	openownership_bods · BODS
registers	registry / authority → listed entity	registry listings · domain_intel
registrar_of	registrar → domain	domain_intel · RDAP
admin_of	admin contact → domain	domain_intel · RDAP
tech_of	tech contact → domain	domain_intel · RDAP
hosted_on	domain → nameserver	domain_intel · DNS
mail_via	domain → MX host	domain_intel · DNS
same_as	entity ↔ same entity (deduped)	entity_link · clustering

Illustrative sub-graph

Subject	Edge	Object	Source
A. Beneficial-Owner	owns_or_controls	Sailwind Holdings FZE	OpenOwnership BODS
Sailwind Holdings FZE	parent_of	Sunrise Overseas Recruitment	GLEIF Level-2 RR (CC0)
PH DMW	registers	Sunrise Overseas Recruitment	PH DMW
Sunrise Overseas Recruitment	same_as	Sunrise Overseas Rec. Agency	entity_link cluster

Illustrative composite example. Real resolved edges stay propose-only — staged for human review, never auto-merged — and the unifier dedups on the triple (higher weight wins, sources unioned) into one edges.jsonl + manifest.