Industries / Industrial equipment and services / Testing, inspection and certification
Reports and certificates
A test report records what a lab did to a sample and what it found, and a certificate or mark says that a certifier vouches for a product; both are worth only what a buyer can check.
5
Milestones achieved, from 1903 to 2021.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1903
A mark for the standard
BSI, the UK's national standards body, registered its British Standard Mark, later known as the Kitemark, as a trademark in June 1903. BSI says a Kitemark now means a product's claim has been "independently and repeatedly tested", and the certificate number printed with the mark can be checked in BSI's online Kitemark directory.
Credit
BSI
1969
A seal carries a duty
A California shopper bought shoes advertised with Good Housekeeping's Consumers' Guaranty Seal, slipped on her vinyl kitchen floor the same day and sued the magazine's publisher, Hearst. On 8 October 1969 the California Court of Appeal let her claim go forward: one who endorses a product for gain may be liable to a buyer injured by relying on it, because the seal implies "reasonable steps to make an independent examination of the product endorsed, with some degree of expertise".
Lessona certifier that invites buyers to rely on its mark can answer to them, not only to the maker who paid.
Credit
California Court of Appeal, in Hanberry v. Hearst Corp
2003
Liability stops at scope
The heirs of a man who repaired small appliances for more than 35 years and died of mesothelioma, a cancer caused by asbestos, sued Underwriters Laboratories (UL), arguing that by testing and certifying the appliances it had undertaken to keep him safe. On 16 April 2003 the California Court of Appeal upheld judgment for UL: it had never undertaken to test the appliances for medical safety or to certify that they would not cause cancer.
Lessona certifier answers for what it said it tested, so a report or certificate is worth no more than the scope written on it.
Credit
California Court of Appeal, in Dekens v. Underwriters Laboratories
2020
Fake certificates for masks
In February 2020 BSI warned that manufacturers were selling medical face masks and other protective equipment for healthcare "on the back of false certificates" made to look as if BSI had issued them, and told buyers to look each certificate up in its online VerifEye directory or send it to BSI to check.
Lessona certificate number is evidence only once someone looks it up with the body said to have issued it.
Credit
BSI
2021
China defines a false report
China's State Administration for Market Regulation (SAMR) put in force on 1 June 2021 rules for the country's inspection and testing bodies, its Order No. 39. They separate an "untrue" report, whose data are wrong or cannot be rechecked because samples or records were mishandled, from a "false" one, which includes a report issued without any testing, one made on a swapped or altered sample, and one with a forged seal, signature or issue date. Labs must keep original records and reports for at least six years and publish a statement on their independence and honesty.
Credit
SAMR
Watch for
One format for signed certificates
UN/CEFACT, a United Nations body for trade facilitation, is writing the Digital Conformity Credential as part of its UN Transparency Protocol: a certificate or test report issued as a digitally signed file that software can verify, naming the conformity assessment body and linking to the accreditation behind it. Its page says version 0.6.0 can be used for pilot testing and that a stable version 1.0 was "due in June 2025"; on 20 November 2026 the project's repository had tagged only version 0.6.1, of June 2025.
Achieved whena stable version is published and an accredited certifier issues routine certificates in the format that a buyer or regulator verifies, as reported by someone outside the project.
Claimed by
UN/CEFACT's UN Transparency Protocol project, June 2025
Open
01
What a report covers, scored
A public test of the AI tools that read test reports and certificates, on the question of what was and was not tested. Take 200 public CB test reports and certificates (CB reports are the clause-by-clause reports labs write under the IECEE scheme), set ten fixed questions for each (which edition of the standard, which clauses were marked not applicable and why, which listed models were tested and which accepted as similar, which national differences were checked, which tests were replaced by component certificates), have two compliance engineers answer them, and score commercial readers and an open model against those answers.
The briefCredit · unclaimed02
Retail units against their reports
For small plug-in power adapters whose makers publish their full CB reports, buy retail units from several sellers, open them, and compare every part in the report's critical-components table with what is inside: maker, model and rating. Publish photos and a part-by-part tally of matches.
The briefCredit · unclaimed
Open 01 · the brief
What a report covers, scored
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of the AI tools that read test reports and certificates, on the question of what was and was not tested. Take 200 public CB test reports and certificates (CB reports are the clause-by-clause reports labs write under the IECEE scheme), set ten fixed questions for each (which edition of the standard, which clauses were marked not applicable and why, which listed models were tested and which accepted as similar, which national differences were checked, which tests were replaced by component certificates), have two compliance engineers answer them, and score commercial readers and an open model against those answers.
Why it is still open
Vendors already sell AI that reads certificates and reports. Certivo says its agent, CORA, extracts certificate data "with certification body and scope lookup" and flags "out-of-scope certificates", but publishes no accuracy test against an outside answer key, and a search found none for any vendor. Makers post full reports, such as Power Integrations' 58-page IEC 62368-1 report from the lab Nemko, in which each clause is marked pass, fail or "test case does not apply".
Built with what exists
Public CB reports and certificates from makers' websites; a language model that reads page images (see the Note "Seven more flagged"); a model that answers fixed-choice questions (see the Note "A thousand questions, one pass"); Python for scoring; two engineers.
First step
Download 20 public CB reports on IEC 62368-1 products, then write and answer the ten questions for five of them.
Done looks like
200 documents from at least 20 laboratories and 2,000 questions with the two engineers' agreed answers; accuracy by question type for at least two commercial tools and one open model; every disagreement listed.
Open 02 · the brief
Retail units against their reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
For small plug-in power adapters whose makers publish their full CB reports, buy retail units from several sellers, open them, and compare every part in the report's critical-components table with what is inside: maker, model and rating. Publish photos and a part-by-part tally of matches.
Why it is still open
Teardowns of cheap chargers are common. In 2023 the YouTuber DiodeGoneWild opened two that met neither the current nor the voltage regulation on their labels. Such teardowns compare a unit with its label or with a genuine one, not with the report that certified the design. IEC 62368-1 reports carry a "Critical components information" table (Table 4.1.2), and a search found no study matching retail units against it.
Built with what exists
Public CB reports; units bought new from at least three channels and never plugged in; hand tools, calipers and a USB microscope; a spreadsheet of each report's table.
First step
Find ten adapters whose full CB reports are public, buy one of each, and match one unit's parts against its table.
Done looks like
30 adapter models, three units each from different sellers; every listed component marked match or mismatch with a photo; each mismatch sent to the maker and the testing laboratory 30 days before publication; a second person rechecks 20 units.