Skip to content

Methodology

Citations and evidence standards

Citations and evidence rules define how Stonewake attaches every customer-visible claim to a stable, publicly served source page and a verbatim quote. The product labels findings as Verified or Reported. It does not invent unsupported assertions for export finance, project finance, or commercial real estate desks.

Why citations and evidence matter on credit desks

Bank credit files already treat documentary support as a control, not a nicety. Customer due diligence, country risk reviews, and adverse media screening all turn on whether a statement can be traced to a reliable publisher. The European Banking Authority's ML/TF Risk Factors Guidelines ask firms whether adverse media reports and other relevant sources are reliable and credible, and state that firms should determine that credibility on the basis of the quality and independence of the source and the persistence of reporting of the allegations, among other considerations.

Stonewake applies the same discipline to market and credit intelligence surfaces. A claim without a usable citation is incomplete. A citation that cannot be opened and matched to the quoted text is incomplete. Desks that later reconstruct a credit memo need the same trail an auditor or second-line reviewer would expect.

Stable URL and verbatim quote rule

Under the citations and evidence standard, a citation must be a stable URL whose publicly served text carries the verbatim quote used in the finding. Pagination indexes and homepage landing pages are not acceptable citation targets when they do not contain the quoted passage. If the quote cannot be located on the cited page, the finding is honestly labelled rather than forced into a verified state.

This rule has three practical consequences:

  • Customer-visible claims point to a concrete public page, not to an internal scrapbook or an undocumented summary.
  • Verbatim quotes keep paraphrases from drifting away from the source language.
  • Analysts can re-open the source and confirm the passage without relying on product memory alone.

The standard is capability-level. It does not disclose crawling architecture, provider names, or extraction internals. It states the acceptance test a finding must meet before it is presented as Verified.

Verified versus Reported labels

Stonewake distinguishes evidence status rather than collapsing every item into a single confidence number. Verified means the quote is present on the cited public page under the stable-URL rule. Reported means the item is shown with transparent limits: the desk sees what is known and what remains unconfirmed. Missing data is shown as missing. Absence of a source is never scored as safety.

That posture mirrors institutional practice outside the product. The OECD country risk classification is produced for Arrangement premium-setting among Participants, not as a general-purpose sovereign rating for capital markets, a scope limit the classification states about itself. Public data products that over-claim create false precision. Labels that separate verified text from provisional reports keep the desk in control of judgment.

Scope across product surfaces

Citations and evidence rules apply across Deal Book findings, country composites, organisation and country adverse-media results, and monitoring briefs. Each surface still has its own analytical job:

  • Deal Book items carry explainable scores and source-backed deal facts for origination review. See deal book scoring and export finance market intelligence.
  • Country pages attribute composite inputs. See country risk composite and the export finance hub.
  • Screening findings quote-verify allegations before draft verdicts. See adverse media screening and counterparty adverse media for banks.
  • Monitoring briefs publish Atom-readable updates with citations for EF, PF, and CRE desks.

Sanctions screening against official lists remains a separate control. Stonewake evidence labels do not replace list matching, and they are not a sanctions or PEP determination. LEI and other identity anchors may appear in dossiers where public registers publish them, always with citation.

Human review and auditability

Scores and screening verdicts are drafts for analyst confirmation. The citations and evidence layer supports that human-in-the-loop posture: an analyst can accept, challenge, or discard a finding because the underlying quote is visible. Append-only audit behaviour for material confirmations is part of the compliance posture described in data residency and compliance.

Stonewake output functions as research evidence for credit and financial-crime workflows, not as an automated credit decision. Nothing in the citations standard converts a news allegation into a listing decision or a capital treatment conclusion.

Boundaries and non-goals

The citations and evidence methodology does not teach register-mining workflows, does not name customer banks, and does not publish provider or model-vendor inventories. It does not claim that every public page on the internet is equally reliable. Source quality still requires desk judgment, consistent with EBA emphasis on the credibility and reliability of adverse media and related sources.

For institutional definitions that sit beside product evidence, the glossary and hubs cover project finance, export credit agency, the project finance hub, and the sanctions and AML trade finance hub.

Working with monitoring briefs and Excel exports

Monitoring briefs inherit the same citations and evidence discipline. Each brief item that states a fact about an organisation or market carries a usable source pointer. The Atom feed is a distribution channel for those briefs; it does not relax the quote rule. When a Deal Book workbook is exported, citations and dataset attributions travel with the sheets so an offline reader can still reconstruct the trail.

That export behaviour matters for credit administration teams that archive committee packs outside the product UI. A score without a citation is incomplete in the same way a guarantee claim file without policy wording is incomplete. The methodology treats citation completeness as a first-class quality gate, not as optional metadata.

Operational failure modes the standard is designed to catch

Common failure modes in open-source credit research include homepage citations that never contain the alleged sentence, paraphrases that harden into false precision, and silent drops of negative facts that lack a convenient URL. The citations and evidence standard forces those failures into the open via Reported labels or omission until a stable page is available. A thinner but honest dossier is preferred over a polished memo that cannot be re-opened six months later.

Internal product telemetry and provider contracts are intentionally excluded from this public page. What banks can rely on is the customer-visible contract: stable URL, verbatim quote, honest labelling, and human review before disposition.

Related terms

Sources

  1. [1]EBA Guidelines on revised ML/TF risk factors

← All methodology