How we reach a finding

1 · What verification means here

When we say a supplier is “verified”, we mean something narrow and checkable: the registered legal entity has been identified, its public records have been reviewed, and its published identity has been compared against those records and against the documents the buyer provided. Verification describes work performed on evidence — it is not an endorsement, a certification or a prediction.

This narrowness is deliberate. Every widening of the word — verified quality, verified capacity, verified reliability — describes something a documentary review cannot see, and using the word that way would make the part we can actually verify less trustworthy.

2 · Source categories

We work from four source categories, in descending order of evidential weight: official registry and government records; the supplier’s formal documents (license copies, quotations, PIs, payment instructions); the supplier’s published identity (marketplace profiles, websites, domains); and statements made by the supplier in correspondence or during a visit.

Weight matters when sources disagree. A registry filing outweighs a storefront claim; a dated record outweighs an undated one; a document the supplier signed outweighs one they merely described. Reports state which category each finding came from.

We do not publish the specific commercial databases and access routes we use, both for contractual reasons and because publishing a checking route in detail also publishes the recipe for defeating it.

3 · Entity matching method

The core operation in every report is matching: does the entity on the storefront, the entity on the quotation, the entity on the PI, and the beneficiary on the payment instructions resolve to one registered company — or to an explainable, evidenced group of related companies?

Matching works from the unified social credit code and the registered Chinese name, never from English trading names, which are unregulated and freely reusable. Where a Hong Kong or offshore entity appears in the chain, we check its registration separately and look for evidenced links — shared directors, disclosed group membership, consistent addresses — rather than assuming a relationship from a similar name.

4 · Online identity comparison

A company’s published identity should agree with its registered identity in mundane ways: the domain registration roughly contemporary with the claimed history, the email domain matching the website, the phone numbers stable across profiles, the claimed address consistent between platforms.

No single mismatch is damning — companies rebrand, migrate platforms and outsource their websites. The report treats online-identity findings as questions with different weights, not verdicts, and says which mismatches most need an explanation.

5 · Document consistency checks

Buyer-provided documents are read three ways: against the registry (does the paper lead to the verified entity?), against each other (does the PI agree with the quotation?), and against common transaction practice (is there an inspection stage, a balance trigger, a packaging spec?).

Omissions are treated as findings with the same seriousness as errors, because omissions are where first-order disputes actually start.

6 · Factory-versus-trader indicators

Documentary indicators include: manufacturing wording in the registered business scope; the registered address type (industrial park versus office tower); registered capital and social-insurance headcount bands where available; and the age and specificity of the company’s product claims.

These indicators are probabilistic. We report them as a weighted picture with an explicit confidence statement, and where the question is decision-critical we say plainly that a live visit to the claimed production address is the direct test.

8 · Risk-rating framework

Reports use five levels and no others. There is deliberately no “safe”, “approved” or “certified” level, and no numeric score — a number would imply a precision the underlying evidence does not have.

The rating is driven by the number and materiality of unresolved inconsistencies across evidence categories. A single unexplained beneficiary-name mismatch weighs more than several immaterial gaps; a refused visit weighs differently from a failed one. The report always states which findings drove the rating, so you can disagree with our weighting using the same facts.

7 · Live-visit evidence framework

A live visit is run against a written checklist agreed with the buyer beforehand, covering the site outside-in: exterior and signage, office or showroom, production floor, machinery, warehouse, packaging, samples, and questions to personnel. The buyer directs the camera in real time.

The representative records three things with equal care: what was seen, what was asked for and refused (with the reason given), and what was said in answer to questions. Refusals are evidence. So is the difference between a confident specific answer and a deflected one — reported as observation, not as verdict.

Visits are never covert. The supplier knows the visit is happening, has consented to filming, and may restrict areas. A visit under these constraints still answers the questions it is designed for: existence, identity, location, visible activity and visible equipment at one moment.

9 · Human review

Every report is reviewed by a person before delivery, and the visible-risk rating is set by that reviewer — not by a scoring formula. Automated retrieval is used for speed; judgement is not automated, because the most important findings are usually about the relationship between facts rather than any single fact.

10 · Conflicting information

Where two sources conflict, the report shows both, with dates and categories, and states which we weight more and why. We do not silently prefer the version that makes the report tidier. A conflict that cannot be resolved becomes an explicit question for the supplier, and unresolved conflicts raise the visible-risk level.

11 · Missing information

Absence is reported as absence. Where a record could not be found, the report says whether that is unusual for a company of the claimed type and age, and what it would take to resolve. Where the total evidence is too thin for a reliable assessment, the rating is “Insufficient information” — we do not average thin evidence into a middling score.

12 · Report limitations

Every report carries a limitations section stating: the review date; that registry data reflects filings which may lag reality; that documentary review cannot see capacity, quality systems, solvency or intent; that a live visit shows one site at one moment; and that nothing in the report is legal advice, certification or a guarantee.

These limitations are printed in the report itself, not linked from it, so they travel with the document wherever it is forwarded.

13 · Corrections and supplier appeals

A supplier who believes a factual statement about them is wrong can write to us with the correction and supporting evidence. Where we got a fact wrong, we correct the report and notify the buyer who commissioned it. We do not remove accurate findings because a supplier objects to them, and we do not accept payment from suppliers in connection with reports about them — see the independence page.