Global Witness: Intelligence Source Guide

Global Witness publishes long-form investigations into the money and ownership behind resource extraction and corruption, plus the sector standard annual count of killings of land and environmental defenders. A documentary source, not a dataset, and it names entities no register connects.

At a glance

Source	Global Witness
Category	Conflict, Crime & Human Security › Environmental & Wildlife Crime
Homepage	https://www.globalwitness.org/
Format	`HTML`
Access	Open — no account required
Disciplines	Environmental Intelligence, Corporate Intelligence
Mission domains	Mining & Resource Crime, Environmental Crime

Resource-corruption & environmental crime investigations. — as catalogued in the platform’s own source registry.

Global Witness is a London-headquartered investigative non-governmental organisation, active since the early 1990s, that documents the link between natural resource exploitation, corruption, conflict and human rights abuse. Its output is not a feed. It is a body of investigative reports – each typically the product of months or years of document work, corporate structure analysis, field research and sometimes undercover reporting – published as PDFs and web articles, with supporting material that varies from a methodology annexe to a full dataset. The subject matter has moved with the organisation's campaigns: illicit timber and conflict diamonds in the early years, oil and mining revenue capture, anonymous company ownership and the professional enablers who supply it, land and agribusiness conversion, and more recently the financing of fossil fuel expansion and the online environment around climate. Alongside the investigations sits its best-known recurring product, an annual report counting and describing killings of land and environmental defenders worldwide, compiled to a documented methodology and now the reference figure cited across the sector and by UN bodies. There is no API. Access is the website, the report PDFs, and whatever data annexes accompany a given publication. Treating it as an intelligence source means treating it as a library of dossiers, each with its own evidentiary standard, scope and vintage.

The analytical job it does that structured sources cannot is to close the gap between a legal record and a real relationship. Registries tell you a company exists and who its nominal director is. Transparency disclosures tell you what a licence holder paid. Neither tells you that the licence was awarded to a shell whose real principal is the minister's business partner, that a timber shipment cleared customs on documents obtained through a specific broker, or that a bank's compliance function waved through a transaction it had internally flagged. Those are findings that require documents nobody publishes, sources who will not go on record, and sometimes an investigator sitting in a room being offered the service on camera. Global Witness produces that category of finding, and it produces it in a form designed to survive legal challenge, which is why its reports are cited in litigation, regulatory proceedings and parliamentary inquiries rather than only in advocacy. The second distinctive job is agenda-setting with consequences: several of the compliance regimes that other sources in this catalogue depend on – conflict diamond certification, extractive payment disclosure, beneficial ownership registers – exist in part because this organisation and its coalition partners campaigned them into being. For CORPINT and ENVINT work the practical consequence is that a Global Witness report is often the earliest public documentation of a network you will later find in registries and sanctions lists, and reading its back catalogue for your sector is a cheap way to acquire several years of accumulated case knowledge.

Who publishes it, and why that matters

This is a campaigning investigative organisation, funded by philanthropic foundations, some government development donors and individual giving, with a stated mission to change the systems it investigates rather than merely to describe them. Every property of the source follows from that. The investigations are evidence-led and legally reviewed, because an organisation that publishes allegations against well-resourced companies and states cannot afford to be wrong and has been sued; the standard of documentation in a typical report is high and the legal review is visible in the careful phrasing. But the selection of subjects is driven by campaign strategy and funding, not by incidence, so coverage is deliberately non-representative: it goes where a finding can change a policy, a regulation or a company's behaviour. Publication timing is likewise strategic, frequently aligned to a legislative moment, a summit or a company's annual meeting, which means the date of publication tells you about the campaign calendar and not about when the conduct occurred. The organisation is also a party in the disputes it documents, has been the target of legal threats and hostile pressure, and operates in an environment where its researchers and partners face real risk. Read its output as high-quality investigative work with an explicit position, corroborate the load-bearing facts independently, and do not mistake the absence of a subject from its catalogue for the absence of a problem.

Provenance is the first question to ask of any dataset and the one most often skipped. Who collects it, what their incentive is, whether they publish a methodology, and whether they correct the record when they get something wrong all bear directly on how much weight a finding drawn from it can carry.

What a record actually contains

The fields you will be working with, what each one means, and whether it is something you can pivot on. Read the meanings carefully — more analysis is wrecked by misreading a field than by failing to find one, and a field that looks like an observation is often an inference.

Field	Type	What it means	Pivot value
`report_title`	string	The publication title. Global Witness reports are frequently known by title alone in subsequent citation, and the title is the practical identifier because there is no report numbering scheme.	Citation tracking, follow-up coverage, related reports in the same campaign series.
`publication_date`	timestamp	When the report was published, which is a campaign decision and not an indicator of when the conduct occurred. The investigative window is usually stated inside the report and can precede publication by years.	Timeline construction; alignment with legislative and regulatory events.
`campaign_theme`	enum	The programme the report belongs to – forests, land, minerals, oil and gas, corruption and illicit finance, defenders, digital threats. Theme determines the methodology and the type of evidence presented.	Related publications, coalition partners, the policy instrument being targeted.
`country / region`	string	Geographic focus. Reflects where the campaign has leverage, not where the problem is worst, and the distinction matters when using report coverage as any kind of indicator.	Country risk assessment, national civil society partners, regional regulatory context.
`named_entities`	array	Companies, banks, government bodies and individuals identified in the report. Naming is a deliberate, legally reviewed act; entities described but not named have usually been withheld for legal or safety reasons rather than absence of evidence.	Corporate registry resolution, sanctions screening, ownership graph, subsequent enforcement actions.
`allegation_type`	enum	The conduct described – bribery, illegal logging, land grabbing, sanctions evasion, money laundering, human rights abuse. This is the organisation's characterisation, not a finding of a tribunal.	Applicable legal regime; regulators with jurisdiction; parallel proceedings.
`evidence_basis`	enum	What the finding rests on: leaked or obtained documents, corporate records, satellite imagery, field interviews, undercover recording, trade data. The basis determines how much independent corroboration you need.	Verification strategy; whether the underlying material is obtainable elsewhere.
`company_response`	string	Responses from named parties, which reputable reports include. The response is frequently as informative as the allegation – a non-denial, a jurisdictional deflection or a claim of subsequent divestment each tell you something.	Company statements, subsequent filings, follow-up correspondence.
`data_annexe`	enum	Whether the report is accompanied by a structured dataset, a methodology note, or neither. Only a minority of reports carry machine-readable annexes, and their formats are inconsistent.	Direct ingestion where present; manual extraction otherwise.
`defender_case_record`	array	In the annual defenders reporting, per-case entries describing a killing: country, date, sector of the dispute, and identifying detail where publication is safe and consented. Compiled from documented cases, not from a registry.	Country and sector aggregation; corroboration against partner organisations; case-level follow-up.
`defender_sector`	enum	The extractive or land-use sector associated with the dispute in which a defender was killed – mining, agribusiness, logging, hydropower, fisheries. Attribution to a sector is an analytical judgement made by the compilers.	Sector-level risk assessment; correlation with concession and forest change data.
`policy_target`	string	The regulation, standard or institution the report is aimed at. Frequently the clearest statement of why this investigation exists and what evidence was assembled to support it.	Legislative record, consultation submissions, subsequent regulatory change.
`source_documents`	enum	Whether underlying documents are published alongside the report. Some investigations publish contracts, filings or correspondence; many cannot, for source protection reasons.	Direct examination of primary material where available; otherwise a request or a records application.

Coverage — and what is not in it

Coverage is thematic and campaign-shaped, extending back to the early 1990s and forming, over that period, one of the deepest continuous documentary records of resource-linked corruption available anywhere. Geographically it concentrates where the campaigns have concentrated: the Congo Basin and West Africa for timber and minerals, Central and Southern Africa for oil and mining revenue, Southeast Asia for timber, palm oil and jade, Latin America for land conflict and agribusiness, and the financial centres – London, and to a lesser extent other secrecy and professional-services jurisdictions – where the money and the corporate structures are arranged. That last emphasis is distinctive and worth noting: a substantial part of the output investigates the enabling side in wealthy countries rather than the extraction side in producing ones. The defenders reporting is the only genuinely global annual series, covering all regions, though its own methodology acknowledges that documentation quality varies enormously by country. Update rhythm is irregular by design. Major investigations appear when they are ready and when the campaign moment is right, which can mean several reports on one theme in a quarter and nothing for a year. The defenders report is annual. Entity coverage is companies, financial institutions, government bodies and named individuals, with the depth on any given entity determined entirely by whether it happened to fall inside an investigation. There is no systematic entity coverage of any sector, and there was never intended to be.

Known blind spots

Absence of evidence here is not evidence of absence. These are the conditions under which Global Witness will not show you something that is nevertheless real:

Subject selection follows campaign strategy and funding rather than incidence, so the absence of a country, company or sector from the catalogue carries no information about whether a problem exists there.
Legal risk shapes what is published. Findings that cannot be defended against a well-resourced claimant in a claimant-friendly jurisdiction are softened, anonymised or withheld entirely, and the report will not tell you which.
Source protection means the strongest evidence is frequently the least visible: documents that cannot be published, interviews that cannot be attributed, and material held back to protect people who would be identifiable from it.
Publication timing is a campaign variable, so a report's date tells you about a legislative or corporate calendar rather than about the currency of the conduct described.
The defenders count is a documented-cases count, not a census. It systematically under-records killings in countries where media and civil society are suppressed, which are frequently the countries with the highest actual risk.
Non-lethal harm to defenders – criminalisation, strategic litigation, threats, forced displacement, sexual violence – is far more common than killing and far less consistently documented in any comparable series.
There is no API, no consistent schema and no stable identifier for reports or entities, so any structured use requires manual or semi-automated extraction that will not be complete.
Retrospective coverage is not maintained. A report is a snapshot; the companies in it merge, rename, divest and sometimes remediate, and the published document is not updated to reflect that.
The organisation is a participant in the policy disputes it documents, so its characterisation of an actor's conduct is an advocate's characterisation, however well evidenced, and should be handled as such in any assessment that must be seen to be neutral.

Write the blind spot into the product. A statement that something “was not observed in Global Witness” is defensible; a statement that it “did not happen” is not, and the difference is what survives cross-examination.

Access, licensing and what you may do with it

Access model: Open — no account required

Everything is free and open on the website. Reports are published as web articles and PDFs, organised by campaign theme and by date, with a search function that is adequate for known-item retrieval and weak for systematic discovery. Practical collection has three parts. First, a one-off harvest of the back catalogue for the themes and countries relevant to your mission, stored locally as documents with their publication dates – this is a few hours of work and it gives you years of case knowledge you would not otherwise have. Second, a subscription to new publications, which for a source with irregular cadence is best handled by monitoring the site rather than expecting a feed. Third, targeted retrieval of any data annexes, which exist for a minority of reports and are the only machine-readable material available. There is no bulk export, no identifier scheme and no licence for automated redistribution of report text. For serious use, extract the named entities into your own entity store and keep the report PDF as the provenance record, because the assertion is only usable if you can point at the paragraph it came from. Where a report cites an underlying document you need, the correct route is usually a records request to the relevant authority rather than an approach to the organisation, whose obligations to its sources come first.

Licence

Report content is copyright of the organisation and published for public reading, campaigning and citation rather than under an open data licence. Quoting with attribution is normal practice and expected; wholesale republication, redistribution of PDFs as a dataset, or scraping the site to build a derived commercial product is not covered by any general permission and should be cleared directly. Data annexes accompanying specific reports sometimes carry their own more permissive terms; check each one, because there is no site-wide answer. Photographs and film in reports frequently carry third-party rights and separate restrictions, and images of individuals in these contexts can carry consent conditions that are not visible in the file. Where a report contains information about identifiable individuals – defenders, victims, witnesses – reproduction raises safety and data protection issues that a copyright licence does not address, and those are the binding constraint rather than the licence.

Rate limits and fair use

There are no published limits because this is a website rather than a service, and the correct etiquette is simply not to behave like a scraper. Harvest the back catalogue once, at a polite rate, with a descriptive user agent and a contact address, and cache everything permanently. Poll for new publications at daily or weekly cadence, not continuously; the publication rhythm is measured in weeks. Retrieve PDFs once and store them; re-fetching a document you already have is pure cost to an organisation running on grant funding. If your requirement is genuinely systematic – a research programme covering the full corpus, for example – contact them rather than crawling, because a research relationship is more likely to yield the methodology and annexe material that makes the corpus usable in the first place.

Licensing changes, and it changes without warning. A dataset that was free for research this year may not be free for commercial or evidential use next year. Confirm the current terms before you build a dependency on it, and record the terms you relied on alongside the data — the licence in force at the time of collection is part of the provenance.

Collecting it

How Global Witness is actually pulled, in the order you would set it up. Prefer the bulk or export interface over per-item lookups wherever one exists: it is kinder to the publisher, faster for you, and gives a reproducible snapshot rather than a series of point-in-time answers you cannot reconstruct later.

Method	Format	Cadence	Notes
Back catalogue harvest	bulk	once, then top up	Retrieve report PDFs and article text for your themes and countries. Store with publication date, campaign theme and a content hash; this is your provenance layer.
New publication monitoring	HTML	daily or weekly	Watch the publications listing. Cadence is irregular, so a schedule is more reliable than any expectation of a feed.
Data annexe capture	CSV	per report where present	A minority of reports ship structured data, most importantly the annual defenders material. Formats are inconsistent and require per-report handling.
Named entity extraction	JSON	per report at ingest	Extract companies, individuals and institutions into your own entity store with a citation back to the specific report and page. Manual review is required; automated extraction over legally careful prose produces attribution errors.
Response and follow-up tracking	HTML	ongoing	Company responses, subsequent corrections, regulatory action and litigation arising from a report are part of the record and change how much weight the original finding deserves.

Ingesting it into the platform

Every step below is idempotent and cursor-based: interrupt one and it resumes from where it stopped rather than duplicating rows or losing progress. Collection is recorded per source, so a feed that quietly stops publishing shows up as a stale timestamp instead of silently thinning your coverage.

Register as a documentary source, not a feed — Add Global Witness in sources.php with an explicit note that records are analyst-extracted assertions from published investigations rather than machine-collected observations. Every downstream analyst needs to see that distinction on the record itself.
Schedule discovery, not collection — Configure collect.php to check the publications listing daily under cron.php. The job is to notice a new report, not to pull a dataset, and a failure should surface as a monitoring gap rather than as an apparently quiet period.
Store the document as the primary artefact — Ingest the PDF or article with its hash, publication date and campaign theme, and treat every subsequent extracted fact as a child of that artefact. If you cannot point at the page, you cannot use the fact.
Extract entities under human review — In ingest.php, create organisation and person entities for named parties, with the allegation recorded as a claim attributed to the report rather than as a property of the entity. The distinction between the entity is X and this report alleges X is the whole of your legal exposure.
Resolve entities against structured sources — Run resolve-everything.php to link extracted companies to corporate registries, sanctions data and, where relevant, extractive licence disclosures. Report-derived entities that do not resolve are the ones worth a closer look, not the ones to discard.
Attach geography and sector — Enrich with concession, mining site or forest area locations where the report identifies them, so investigative narrative can be placed on the same map as satellite and mapping data rather than compared in prose.
Correlate with the observational sources — Use correlate.php to test report assertions against forest change alerts, mining site mapping, vessel activity, trade statistics and sanctions actions. Convergence between an investigative claim and an independent observation is the strongest product this source participates in.
Surface to the mission areas with provenance intact — Publish entities and claims into human-rights.php, organized-crime.php and financial-crime.php with the attribution visible in the record, and configure alerts on new publications naming entities already in your watchlist.

Registered sources and their last-collected state are listed in sources.php, and the scheduled chain that keeps them current is in automation.php.

How it is wrong, and how to tell

Every dataset is wrong in characteristic ways. Knowing which ways is the difference between using a source and being used by one, and it is the part of source evaluation most often skipped because it is the part that takes work.

By the standards of NGO investigative output this is at the top end, and the reasons are structural rather than reputational. The organisation publishes allegations against parties with the resources and the appetite to sue, has been sued and threatened, and consequently operates a legal review process that filters out anything it cannot stand behind. That produces a characteristic style – careful attribution, hedged formulations, named parties given a right of reply that is printed – which reads as cautious and is in fact a reliability signal. The documentary basis is generally strong: corporate filings, contracts, leaked material, trade records and field research, frequently combined so that a single source's failure does not collapse the finding. Where you should apply more scrutiny is at the interpretive layer. The characterisation of conduct is an advocate's characterisation; the framing selects the facts that support a policy argument; and quantitative claims, particularly aggregate valuations of losses or flows, are frequently estimates built on stated assumptions that deserve to be read rather than quoted. The defenders series has a documented methodology and is the best available, but it is a count of documented cases and its compilers say so. The correct posture is to treat a Global Witness finding as a strong, well-sourced allegation that shifts your prior substantially, to corroborate the specific facts you intend to rely on through registries, filings, court records or independent observation, and to keep the organisation's characterisation and the underlying facts separate in your own product.

Characteristic false positives

Advocacy framing selects and orders facts to support a policy conclusion, so a report can be entirely accurate in every particular and still leave a reader with a distribution of blame that a neutral assessment would not reach.
Aggregate figures – hectares affected, revenue lost, volumes trafficked – are frequently estimates from stated assumptions, and they get quoted downstream stripped of the assumptions until they harden into apparent facts.
Corporate identification is difficult in exactly the structures these reports investigate, and a named entity may be a similarly named affiliate, a former parent, or a company that had already divested by the time of publication.
Publication date is regularly mistaken for the date of the conduct, producing timelines that place a decade-old transaction in the year the report appeared.
Entities are named as at the time of investigation and the report is never updated, so a company that subsequently remediated, was acquired, or was cleared carries the allegation forward unqualified.
The defenders count is under-inclusive in repressive states and its year-on-year movement can reflect changes in documentation capacity or partner networks rather than changes in violence.
Sector attribution in defender cases is an analytical judgement about a dispute that may have several drivers, and treating it as a categorical fact overstates the precision of the underlying record.
Findings withheld for legal reasons are invisible, which means the published version of an investigation can systematically understate the role of the best-lawyered party in it.

None of these make the source unusable. They make it a source that requires corroboration before an assertion built on it goes into a product, which is true of every source and admitted by few.

Ageing

Investigations age unevenly. The documentary core – a contract, a filing, a shipment record, a recorded conversation – does not age at all; it is evidence about a moment and remains so. The entity layer ages quickly, because the companies described in these reports are precisely the kind that restructure, rename, redomicile and divest, frequently in direct response to publication. Within two years of a significant report, a meaningful share of the named corporate entities will have changed in some material way, and the report will not say so. The legal and regulatory context ages too: a practice documented as lawful-but-abusive may since have been prohibited, or a sanctions designation may have been imposed or lifted, either of which changes what the finding means today. The defenders series ages in a specific and useful way – old years remain valid as documented counts, but the methodology has been refined over time, so a long series should be read with its methodology notes rather than as a clean trend. A stale record here looks like a confidently cited allegation against a company that no longer exists under that name, in a jurisdiction whose law has since changed, based on a report whose subsequent litigation history you did not check. The mitigation is to re-resolve every named entity at the time of use and to check what happened after publication before you rely on what the report said at publication.

What this source feeds

A source is only worth what it lets you conclude. These are the disciplines that collect through it, the mission domains it serves and the data points it yields — every one is a tag, so you can follow any thread from here into the rest of the library.

Collected by these intelligence disciplines

Serves these mission domains

Yields these data points

How each sector uses Global Witness

The same dataset is worked very differently depending on who you are, what authority you hold, and what you are ultimately producing. A military analyst is supporting a commander’s decision; a journalist is meeting a publication standard; an NGO caseworker is protecting a person. The records are shared — the constraints, thresholds and outputs are not.

🎖 Military and defence

The relevance is to the political economy of conflict rather than to targeting or force protection. These investigations document how armed actors, connected elites and commercial partners convert resources into revenue, and that is directly useful for understanding which economic actors sustain a conflict, which international companies and banks sit in the chain, and what leverage exists outside the military instrument. For stabilisation and security sector work, the reporting on land conversion and mining is also a reliable indicator of where communities and armed actors will come into conflict next. Two limits. The reporting is retrospective and campaign-timed, so it informs assessment rather than current operations. And it is an advocacy product, which means anything used in a formal assessment should be corroborated and attributed, not folded in as background fact.

🕵 National intelligence

Treat this as high-grade open-source reporting from a collector with unusual access, and handle it accordingly. Its distinctive contribution to CORPINT and ECONINT is documentation of relationships that appear nowhere in structured data: the true principal behind a licence holder, the intermediary who arranged a payment, the professional firm that built the structure. Those are exactly the links that turn a scattered set of registry entities into a network. Its second contribution is early warning about compliance regimes, since the campaigns it runs today frequently become the regulations that constrain targets in five years. The analytical discipline is provenance: record the claim as attributed to the report, keep the advocate's characterisation separate from the documentary facts, and corroborate anything load-bearing. Cross-referencing report-named entities against sanctions actions taken subsequently is a productive routine, because the sequence is often report first, designation later.

👮 Law enforcement

For financial investigators, anti-corruption units and environmental crime prosecutors, these reports are lead generation of unusually high quality, and in several documented instances they have preceded formal investigations into the same conduct. The practical use is to identify entities, relationships and jurisdictions worth pursuing through legal process, then to obtain the underlying evidence yourself. The report itself is generally not evidence: it is hearsay in most systems, its sources are protected and will not be produced, and building a case on the publication rather than on the primary material is a route to failure at the first evidential challenge. Two further cautions. Approaching the organisation for source material puts it in a position where its duties to its sources conflict with your request, and should be done through appropriate channels if at all. And publication may already have alerted the subject, so assume the evidential trail has had time to be tidied.

🔍 Private investigation and corporate security

In due diligence, corporate intelligence and litigation support, a Global Witness report naming a counterparty is a material finding that most clients will want to know about, and its absence from a due diligence report will be difficult to justify later. The correct handling is precise: report the allegation as an allegation, with its date, its evidential basis, the response the named party gave, and what has happened since – regulatory action, litigation, divestment, or nothing. Adverse media screening tools will usually surface the headline and miss all four of those, which is where a competent analyst adds value. The ageing problem is acute in this context, because a report from six years ago about an entity that has since been restructured needs to be characterised carefully rather than repeated. And the legal risk of republishing an allegation in a client report is yours, not the publisher's.

📰 Journalism and OSINT media

For investigative journalists the relationship is usually collaborative rather than extractive: the organisation partners with newsrooms, publishes alongside them, and its reports are structured to be reportable. Used as a source, the value is a well-documented starting point with entities already identified and a right of reply already exercised, which saves weeks. The professional obligations are the ordinary ones plus two specific to this source. First, verify independently before publishing, because your legal position rests on your own reporting rather than on someone else's; a report is a lead, not a substitute for verification. Second, attribute clearly and check what happened after publication, since companies frequently respond, remediate or sue and a story that ignores the aftermath will be corrected in public. For defenders reporting, follow the safety practices the sector uses when naming or locating people who are still at risk.

🌍 NGO, humanitarian and human rights

For NGOs this is a peer source and often a coalition partner, and the practical use is threefold: methodology, precedent and leverage. The methodology is worth studying – how a corporate structure is unpicked, how an allegation is documented to a standard that survives legal review, how a right of reply is managed – because it is transferable to your own work. The precedent matters because a documented case in your sector gives your own advocacy a citable anchor. The leverage is the defenders series, which has become the reference figure in UN and regulatory processes and gives national organisations an internationally recognised number to argue from. The reciprocal obligation is real: much of the underlying documentation comes from national partners taking risk, so cite properly, do not re-expose case detail that was published in a controlled form, and coordinate before publishing anything that identifies people still at risk.

🎓 University and research

The corpus supports research in political ecology, corporate accountability, environmental governance, illicit finance and the study of transnational advocacy itself. Its properties as a research object are specific: it is a purposive rather than a representative sample, produced by an organisation with a declared position, in a documented investigative tradition, over three decades. That makes it excellent for case-based and process-tracing research and unsuitable for incidence estimation without heavy caveats. The defenders dataset is the most quantitatively used component and requires careful handling – it is a documented-case count with acknowledged reporting bias, its methodology has evolved, and it should never be presented as a population estimate. The organisation is also a legitimate subject of study in its own right, as an actor whose campaigns produced several of the international transparency regimes that other researchers now treat as institutional background. Ethics review is warranted for any work that engages with case-level information about defenders or witnesses.

Playbook: working Global Witness end to end

A repeatable sequence from first pull to finished product. Each phase states what you are trying to establish, not merely what to click — the objective is a defensible chain of reasoning, not a completed checklist.

Phase 1 — Read the back catalogue for your sector before you start

Harvest and read the reports covering your commodity, region or financial centre. This is the cheapest available way to acquire several years of accumulated case knowledge about how the specific illicit structures in your field are actually built, and it will change what you look for in structured data.

Phase 2 — Separate the documentary core from the argument

For each report, identify what is documented – a contract, a filing, a recording, an image – and what is inference or characterisation. Only the first is portable into your own product without attribution. Doing this systematically is the difference between using the source and repeating it.

Phase 3 — Extract entities with page-level citation

Pull every named company, individual and institution into your entity store, each with a pointer to the specific page. An assertion you cannot locate in the document is an assertion you cannot defend, and legal review of your own product will start here.

Phase 4 — Resolve every named entity independently

Take each company to a corporate registry, each individual to whatever identity sources you have, and establish current status: existing, renamed, merged, dissolved, divested. The report is a snapshot and this step is what makes it usable today.

Phase 5 — Check what happened after publication

Search for regulatory action, prosecutions, litigation, corrections, retractions and company statements. A finding that was followed by a sanctions designation is a different proposition from one that was followed by a successful libel claim, and neither is visible in the report.

Phase 6 — Corroborate the load-bearing facts

Take the two or three facts your own conclusion actually depends on and verify each through an independent route – filings, court records, trade data, satellite observation, or a second reporting organisation. Everything else can stay as attributed background.

Phase 7 — Test the assertions against observational sources

Where the report describes clearing, mining or vessel activity at a location and time, check the satellite and tracking record. Convergence between investigative narrative and independent physical observation is the strongest evidentiary combination available in this field.

Phase 8 — Map the network the report exposes

Build the graph: principals, intermediaries, corporate vehicles, financial institutions, jurisdictions. The value of these investigations is usually a relationship structure rather than a single fact, and the structure frequently persists after the specific transactions have ended.

Phase 9 — Handle the defenders data as a documented-case count

If you use the annual series, read the methodology, state the reporting-bias limitation explicitly, and resist year-on-year trend claims that the compilers themselves qualify. Aggregate by sector and country for risk assessment; do not present it as a measure of total violence.

Phase 10 — Apply the safety filter before anything leaves your organisation

Identify every individual in your output who is still at risk – defenders, witnesses, community members, local researchers – and decide the granularity of identification and location with them where possible. This is a physical safety decision with a documented casualty record behind it, not an editorial preference.

Phase 11 — Get the legal review done on your own product

Republishing an allegation transfers the risk to you, in your jurisdiction, without the publisher's legal file behind it. Have the characterisations reviewed, keep the attribution explicit, and record the response any named party gave in the original.

Phase 12 — Set the watch on entities and themes

Register the named entities in watchlist.php and monitor for new publications touching them, alongside sanctions, registry and litigation changes. The follow-on events after a major investigation – designations, resignations, divestments, prosecutions – are frequently the actionable part.

The platform ships this as a step-checked workflow in playbooks.php, so progress is recorded against a case rather than held in someone’s head.

What to pair it with

No single source carries a finding. These are the datasets that corroborate, extend or contradict this one — and a source that contradicts is worth more than one that agrees, because it is the only thing that will tell you when you are wrong.

Source	Relationship	What it adds
OCCRP	corroborates	Cross-border investigative reporting on organised crime and corruption, frequently covering the same networks with a different sourcing base and often with searchable underlying data.
OpenCorporates	prerequisite	Company registration data for resolving named entities to legal persons and establishing what has changed since publication.
OpenSanctions	extends	Sanctions and PEP data for screening named parties and for detecting the common sequence in which investigative reporting precedes designation.
EITI	corroborates	Official licence registers and payment disclosures against which an investigative allegation about a concession or a payment can be tested.
Global Forest Watch	corroborates	Independent satellite observation of land conversion at the locations an investigation describes, converting narrative into physically verifiable fact.
IPIS Research	extends	Field-mapped mining site and armed actor data in Central Africa, providing systematic coverage where investigations provide depth on individual cases.
Front Line Defenders	corroborates	Independent documentation of attacks on human rights defenders, giving a second count and considerably more detail on non-lethal harm.
Transparency International	extends	Corruption research, national chapters and indices that supply the governance context around a specific investigative finding.
UN Comtrade	corroborates	Bilateral trade statistics for testing claims about commodity volumes and routes described in an investigation.

Legal, ethical and operational constraints

The dominant constraint is defamation and it applies to you, not to the publisher. Repeating an allegation republishes it, and in most common-law jurisdictions the repetition rule means you cannot defend yourself simply by pointing at the original. The named parties in these reports are frequently well resourced, several of the relevant jurisdictions are claimant-friendly, and strategic litigation against public participation is a documented feature of this field rather than a hypothetical – the organisation and its peers have been on the receiving end of it repeatedly. Practical protections are to attribute explicitly, to characterise allegations as allegations, to record the response the named party gave, to check what has happened since, and to have your own output reviewed before it is circulated outside your organisation. Copyright is a lesser but real constraint: report text and images are not open-licensed and bulk reproduction is not covered by any general permission. Where reports contain information about identifiable individuals at risk, data protection obligations and a direct duty of care both apply, and the duty of care is the more important of the two. Finally, if your work is regulated – due diligence, financial crime compliance, journalism under a code – the standards of that regime apply to how you use an allegation, and typically require you to record what verification you actually did.

Operational security

Reading a public website is low exposure in itself, but the pattern of reading is not. Sustained retrieval of every report touching one company or one country, from an attributable network range, is a legible signal to anyone with access to the logs, and this is an organisation whose infrastructure is of interest to parties who would like to know who is researching them. Collect broadly and filter locally, or take a full back-catalogue copy once and work offline. The larger consideration is not your exposure but the exposure you can create for others. Investigations in this field depend on national partners, community researchers and sources who are identifiable to a determined adversary from surprisingly little detail. Any approach you make to the organisation for underlying material, any onward sharing of case-level defender information, and any republication that adds specificity to a location or a role can shorten the distance between a published report and a person. Assume that adversaries in these cases monitor derivative coverage as well as the original, and that additional detail you add is the detail they did not have.

Two rules that hold regardless of jurisdiction. Collection that is lawful is not automatically proportionate, and a dataset assembled for one purpose does not carry consent for another. Where the records concern identifiable people, the question is not only whether you may hold the data but whether holding it serves the purpose you are accountable for.

Is it earning its place?

Sources accumulate. Feeds get added during an incident and are never reviewed again, and a decade later the pipeline is carrying dead weight that nobody dares remove. These are the measures that show whether Global Witness is contributing anything, and they are worth baselining now so the answer is available later.

Number of entities in your investigative graph that originated from a Global Witness report and were subsequently corroborated in a structured source, which measures whether the corpus is feeding your pipeline or just decorating it.
Proportion of extracted assertions that carry a page-level citation to a stored document, which should be one hundred percent and is the basis of your legal defensibility.
Rate at which report-named entities later appear in sanctions designations, prosecutions or regulatory action, which calibrates how much predictive weight the source deserves in your sector.
Currency of your entity resolution: the share of report-derived companies whose registry status has been re-checked within your review period.
Latency between publication and the report appearing in your platform with entities extracted, which for a source with irregular cadence is a monitoring quality measure.
Coverage overlap between this source and your other investigative sources, which tells you whether you are getting independent corroboration or the same finding recycled through several publishers.
Number of cases in cases.php where a report supplied the initiating hypothesis, as distinct from being cited as background.
Count of your own outputs that required correction following a post-publication check you should have done first, which is the honest measure of whether your ageing discipline works.

Beware of volume. Indicator counts rise easily and say almost nothing. Unique contribution — findings this source produced that no other source in your stack would have — is the measure that matters, and it is usually far lower than anyone expects.

Tradecraft notes

The distinctions that separate a competent analyst from a fast one:

Distinguish the documentary core from the campaign argument in every report you read. The documents travel into your product; the characterisation does not, unless attributed.
Publication date is not the date of the conduct. Find the investigative window inside the report, and build your timeline on that.
Absence from the catalogue means nothing. Subject selection follows campaign leverage and funding, so treat coverage as a spotlight rather than a survey.
Read the company response section closely. A carefully drafted non-denial, an assertion of prior divestment, or a jurisdictional objection each tell you something the allegation does not.
Always check what happened after publication before citing. Designations, prosecutions, corrections and successful legal challenges all change the weight of a finding and none of them appear in the original document.
Handle the defenders series as documented cases with acknowledged under-reporting. Aggregate for risk assessment, qualify every trend claim, and never present it as a total.
Look for the withheld material. Careful, hedged phrasing around one actor in an otherwise direct report often marks the point where legal review bit, and that is frequently where the most significant relationship sits.
When you can pair an investigative claim with an independent physical observation – forest change, vessel tracking, imagery – do it. That combination is the strongest thing this source participates in and it is what survives challenge.
Treat the security of named individuals and local researchers as a constraint on your output, not as a consideration to be balanced against publication value. The sector's own casualty documentation is the argument.

Questions analysts actually ask

Can I cite a Global Witness report as evidence?

You can cite it as a published allegation with a documented basis, which is a meaningful thing to cite. It is not evidence in most legal or evidential senses: sources are protected, underlying documents are frequently unpublished, and the report is hearsay. For anything load-bearing, obtain the primary material yourself.

Why is there no API or dataset?

Because the output is investigative reporting, not data collection. A minority of reports carry structured annexes, most notably the annual defenders material. Everything else requires document harvesting and analyst extraction, and any structured use you build will be your own.

How reliable is the annual defenders count?

It is the best available series and it is a count of documented cases, not a census. Documentation depends on functioning media and civil society, so the countries with the worst conditions are systematically under-counted. Use it for sector and country risk patterns, and qualify any year-on-year comparison.

A company was named in a report five years ago. How should I treat that in due diligence?

As a material finding requiring context. Report the allegation, its date, its basis and the response given, then establish what happened since – regulatory action, litigation, divestment, restructuring or nothing. The context is the value you add; the headline alone is misleading in both directions.

Is the organisation biased?

It has a declared position and campaigns for policy change, which shapes what it investigates and how it frames findings. That is different from being unreliable: the evidentiary standard is high because it publishes against parties who sue. Handle the framing as advocacy and the documentation as strong evidence, and keep them separate in your own output.

Can I republish a report or its data?

Report content is copyrighted and published for reading and citation, not under an open licence. Quote with attribution; clear anything more with the organisation. Some data annexes carry their own terms. Images frequently carry third-party rights and consent conditions that are not visible in the file.

How do I know what was left out for legal reasons?

You cannot know directly, but you can read for it. Sudden shifts to passive construction, unnamed parties described in otherwise specific passages, and hedged phrasing around a single well-resourced actor are the usual markers. Treat those passages as pointers to where more exists.

Should I contact them about a case I am working?

Consider carefully. The organisation's obligations to its sources come first, and a law enforcement or commercial approach can place it in an impossible position. Where contact is appropriate it should go through proper channels, and you should not expect source material. For most purposes, records requests to the relevant authority are the better route.

Does the platform generate any of these entities or claims with a model?

No. Entities and claims from these reports are extracted under human review and stored as attributed assertions with page-level citation. The only language model involvement in the platform is the Summarise skill in copilot.php, which writes prose over records that already exist and originates no indicator, relationship or attribution.

Standards, formats and interoperability

What this source speaks natively, and what it has to be translated into before a partner can consume it. Work that arrives in a recognised format is easier to defend, easier to hand over and easier to automate against:

UN Guiding Principles on Business and Human Rights, the framework against which most corporate conduct findings in this corpus are assessed.
OECD Guidelines for Multinational Enterprises and the associated due diligence guidance, including the minerals guidance that several investigations directly target.
Beneficial ownership transparency standards and registers, a regime this organisation and its coalition partners campaigned into existence and continue to test.
Extractive payment disclosure regimes, including EITI and mandatory reporting rules in several major markets, which are the policy targets of much of the extractives reporting.
Anti-money-laundering standards from the international standard-setter, particularly on professional enablers and beneficial ownership, which frame the illicit finance investigations.
Emerging corporate sustainability and deforestation-free due diligence law in major consumer markets, which converts many of these findings into compliance obligations.
The Escazu Agreement and regional instruments protecting environmental defenders, which give the defenders reporting a legal frame in Latin America.
STIX 2.1 and MISP as platform export formats, with report-derived entities exported as identity objects carrying explicit attribution to the source document.

References

Primary documentation and authoritative references for this source. Publishers revise and retire material, so treat the retrieval date as part of the citation and re-check before relying on any of it in a formal product.

Global Witness — Global Witness. The investigations, campaign pages and annual defenders reporting. Start with the campaign theme relevant to your sector and read the back catalogue in date order.
OCCRP — Organized Crime and Corruption Reporting Project. Cross-border investigative reporting covering overlapping networks, frequently with searchable underlying data and a different sourcing base.
Front Line Defenders — Front Line Defenders. Independent documentation of attacks on human rights defenders, including the non-lethal harm that the annual killings count does not capture.
Transparency International — Transparency International. Corruption research, indices and national chapters providing governance context for specific investigative findings.
OpenCorporates — OpenCorporates. Company registration data, the first step in resolving any named entity and in establishing what has changed since publication.
OpenSanctions — OpenSanctions. Consolidated sanctions and PEP data, useful both for screening named parties and for tracking the report-then-designation sequence.
Publish What You Pay — PWYP. The extractives transparency coalition this organisation co-founded, and the route to national civil society perspectives on the same cases.
Global Forest Watch — World Resources Institute. Independent satellite observation for physically corroborating land conversion described in an investigation.
IPIS Research — International Peace Information Service. Systematic field mapping of mining sites and armed actors, complementing case-depth investigations with area coverage.
OHCHR — UN Office of the High Commissioner for Human Rights. The UN human rights framework, special procedures and defender protection mandates that this reporting feeds into and is cited by.

Link integrity: every reference above was verified with a live request when this page was generated. Where a publisher had moved or withdrawn a document, the link was repointed at a preserved copy in the Internet Archive and marked as archived. Anything with no reachable copy anywhere had its link removed rather than left to rot — the source is still credited, it simply cannot be linked.

Put it into practice

The Quantus Intel threat intelligence platform operationalises this source: Global Witness investigations are ingested as stored documents with analyst-extracted entities and page-level citation, resolved against registries and sanctions data, and tested against the satellite, mapping and trade sources that can corroborate them physically.. Browse the full source catalogue, or follow any tag above into the rest of the library.