BRENNEN

Industries / Industrial equipment and services / Industrial services / Safety training

Certification and records

Certification and records are the cards, certificates and databases that show a worker has had the safety training a job needs, and let an employer check it.

5

Milestones achieved, from 1971 to 2024.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1971

    Ten hours, one card

    OSHA began its Outreach Training Program in 1971. Trainers whom OSHA authorizes, but who are not OSHA staff, teach 10-hour classes on common hazards for workers and 30-hour classes for supervisors, and each student receives a completion card. OSHA counts more than 6.51 million workers trained through the program from fiscal years 2021 to 2025. The card does not meet an employer's duty to train under specific OSHA standards, OSHA keeps no records of the classes, and replacement cards are issued only for training in the last five years. OSHA publishes a watch list of trainers it has suspended or revoked for breaking the program's rules and refers fraud to the Labor Department's inspector general; the list held 273 suspensions and 24 revocations when read in November 2026.

    Lessonwhen the issuer keeps no central record, a card can be checked only through the trainer who issued it.

    osha.govosha.govosha.gov

    Credit

    OSHA

  2. 1995

    A card for every site

    The Construction Skills Certification Scheme (CSCS) issued its first card in 1995. Most of its cards require a pass in the Health, Safety and Environment test run by the Construction Industry Training Board (CITB). By CSCS's count it issued its millionth card in 2007 and its first smartcard, with a chip for electronic checks, in 2010, and has issued more than six million in all. In 2022 it rolled out Smart Check, one service that now checks all 2.1 million cards carrying its logo, including those of allied schemes, as a phone app, a website and a link that site gate systems can call. CSCS now says fake cards are rare and that fraud has moved to the qualification: test centre staff who let one person sit a test for another, or examiners who pass candidates below the standard.

    Lessononce cards can be checked at the gate, fraud moves to the test behind the card.

    cscs.uk.comcscs.uk.com

    Credit

    CSCS; CITB

  3. 2012

    A shared course for wind

    The Global Wind Organisation (GWO), a non-profit based in Denmark and owned by its members, who are renewable energy manufacturers, owners and operators and independent service providers, published its Basic Safety Training standard in 2012 because the industry wanted one recognised standard. Training providers must be certified by GWO-approved certification bodies, and they upload participants' records to WINDA, GWO's training record database, where employers check whether a technician's training is current. GWO's members accept a current WINDA record as proof that the standard's learning objectives were met. By GWO's own figures, 140,544 people took GWO courses in 2025 from 636 training providers, and each held 6.16 training records on average.

    globalwindsafety.orgglobalwindsafety.orgglobalwindsafety.org

    Credit

    GWO and its member companies

  4. 2015

    Approved courses for heights work

    From 1 April 2015, Ontario construction workers who use fall protection, typically those working 3 m or more above the ground, had to take working-at-heights training under a provincial program standard, with a deadline of 1 October 2017. The standard sets learning objectives, a minimum class length, a classroom part and a maximum class size, and requires a knowledge test and a demonstration of skills; a separate standard certifies the providers. Lynda Robson and colleagues at the Institute for Work & Health found test scores rose from 6.8 to 9.5 out of 10 with the training and stood at 7.5 two years later. Lost-time claims for the falls the training targeted were 19% lower in 2017 to 2019 than in 2012 to 2014, against 6% lower in the other nine provinces. An updated standard took effect on 1 April 2024.

    Lessonmost of the gain in test scores had faded within two years, though scores stayed above where they started.

    iwh.on.ca

    Credit

    Ontario Ministry of Labour; Robson and colleagues (Institute for Work & Health)

  5. 2024

    Cards without qualifications end

    CSCS issued some cards under Industry Accreditation, on the strength of an employer's recommendation rather than a recognised qualification. It withdrew the route to meet a recommendation of the Construction Leadership Council that every construction card scheme rest on nationally recognised qualifications for all occupations. All Industry Accreditation cards issued from 1 January 2020 expired by 31 December 2024, and holders now replace them by showing a qualification.

    Lessona card that records an employer's word, not a tested qualification, lost its standing once the industry asked what the card proved.

    cscs.uk.com

    Credit

    CSCS; Construction Leadership Council

Watch for

  1. AI picks the right card

    CSCS says the upgraded My CSCS app lets a worker upload a photo of a qualification certificate, after which an AI-driven process recommends the right card for the worker's occupation. The digital card appears in the app once an application is approved and can be checked through Smart Check by a changing QR code.

    Achieved whenCSCS or an outside auditor publishes how often the AI's recommendation matches a human assessor's decision on a sample of applications that includes wrong and forged documents.

    cscs.uk.comcscs.uk.com

    Claimed by

    CSCS, December 2025

Open

  1. 01

    Refresher intervals against forgetting

    A public table of how often common safety cards and certificates must be renewed, set against published measurements of how fast the matching knowledge or skill fades.

    The brief
    Credit · unclaimed
  2. 02

    Test AI readers on certificates

    A public test set of at least 500 made-up safety training certificates, each marked SPECIMEN and issued in the names of fictional training providers and people, laid out the way real ones are (course title, dates, expiry, provider number, learner name). Each is printed and photographed by phone in poor light, at an angle or partly folded, with a known answer for every field, and some are expired or altered. Score how accurately AI document readers pull out each field and how often they flag the expired and altered ones.

    The brief
    Credit · unclaimed

Open 01 · the brief

Refresher intervals against forgetting

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table of how often common safety cards and certificates must be renewed, set against published measurements of how fast the matching knowledge or skill fades.

Why it is still open

Renewal periods differ widely for training of the same kind. A CCNSG passport lasts three years; Singapore's CSOC must be passed again every two years, or every four after six years in the trade; a Western Australian construction induction card never needs repeating while its holder stays in the industry. Measurements of fading exist but sit in separate studies: Ontario's working-at-heights scores fell from 9.5 to 7.5 out of 10 within two years, Kluge and Frank tested four refresher methods on a simulated process control task (2014), and Lawani, Hare and Cameron studied early refresher practice in height safety and rescue training (2018). Across 457 reports on skills with a physical part, C. E. Tatel and P. L. Ackerman of the Georgia Institute of Technology found that half the accuracy gained in training was lost after about 6.5 months without use (2025). A search found no table that sets renewal rules against such measurements.

eutils.ncbi.nlm.nih.govecitb.org.ukmom.gov.sgworksafe.wa.gov.auiwh.on.caapi.crossref.org

Built with what exists

Scheme rules from regulators' and schemes' own pages; the Europe PMC and Crossref search services to find studies of skill decay; a spreadsheet; an agent that cites the page behind each figure, such as one built with Cursor's agent kit (see the Note "Findings that cite notes"); Zenodo.

  1. First step

    Enter the schemes named in this domain and in Site orientation with their renewal rules, then search Europe PMC for decay studies on first aid, work at height and rescue.

  2. Done looks like

    At least 40 schemes from at least eight countries, each renewal period linked to its rule; for every scheme with a matching decay study, the time after training at which measured scores fall below the scheme's own pass mark, set beside its renewal period; table and sources published.

Open 02 · the brief

Test AI readers on certificates

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test set of at least 500 made-up safety training certificates, each marked SPECIMEN and issued in the names of fictional training providers and people, laid out the way real ones are (course title, dates, expiry, provider number, learner name). Each is printed and photographed by phone in poor light, at an angle or partly folded, with a known answer for every field, and some are expired or altered. Score how accurately AI document readers pull out each field and how often they flag the expired and altered ones.

Why it is still open

CSCS now has an AI-driven process read uploaded qualification certificates and recommend a card, and publishes no accuracy figures. In a neighbouring field, Henrique Lin, Tiago Dias and Miguel Correia of INESC-ID in Lisbon trained extraction models on synthetic property documents and measured them against manual checks (2025). A search found no test set for training certificates.

cscs.uk.comarxiv.org

Built with what exists

Python with the ReportLab library to lay out the certificates; a phone camera; general AI models that read images, such as Claude and Gemini; the open-source OCR program Tesseract as a baseline; a spreadsheet of known answers; Zenodo.

  1. First step

    Design ten certificate layouts for fictional providers and photograph the first 50 certificates.

  2. Done looks like

    500 certificates and 1,500 photos; field-level exact-match accuracy for at least three readers; the share of expired and altered certificates each reader flags; set, code and scores published under an open license.