Safety is cited. Competency is attested.

Why safety-critical training is different

In general industry, construction, maritime, agriculture, and the regulated trades, a confidently-wrong AI answer is not a content-quality problem — it is a safety incident. A generic assistant will hallucinate a torque spec or a lockout step in the same fluent voice it uses for everything else. VLE is built so that the reasoning engine is constrained hardest exactly where the stakes are highest.

How Arthur handles safety-critical content

• Every safety assertion is cited. Safety-critical guidance points to an authoritative reference — the relevant OSHA standard, the manufacturer's manual, or the governing regulation — rather than free-floating generation. International equivalents (EU OSH directives, UK HSE, national occupational-safety rules) are handled the same way.
• A deterministic override governs dangerous errors. When a learner shows a safety-critical misconception, a hard rule takes over the reasoning engine: Arthur becomes direct and unambiguous, addresses only that one error, refuses to soften or hedge, and asks the trainee to confirm understanding before moving on. This is hard safety, enforced in the architecture — not a tone preference.
• Trainer attestation before guidance reaches a trainee. In safety-critical scope, AI-assisted guidance is reviewed by the instructor — the safety-of-record — before it is presented. The AI supports the trainer; it never becomes the safety authority.
• Every interaction is logged. Safety-critical exchanges are captured in the audit trail for incident review.

Competency, not seat-time

A funder or an employer does not want to know that a trainee spent the hours. They want to know the trainee can do the job. VLE measures competency against the rubric the training authority signed off on, and every competency assessment captures four things:

• The rubric criterion the trainee is being assessed against.
• The AI's observation, with calibrated confidence and the evidence it is based on.
• The instructor's attestation — the explicit human decision that the competency is met.
• The evidence trail — preserved so the attestation can be audited later.

Aligned to the frameworks you're funded against

Learning paths tag each lesson to the competency it advances, and competencies map to the framework the program is accountable to — NAVTTC's National Vocational Qualifications Framework, the EU's EQF, US CTE / Perkins V, and apprenticeship standards (US Registered Apprenticeship, UK Apprenticeship Standards, the EU Apprenticeship Framework). Trainees see mastery progress against the framework, not just course completion; program managers get cohort reporting that maps cleanly to whichever framework their funder operates under.

Recertification, with the evidence preserved

Many vocational credentials require periodic recertification. VLE tracks the cycle and surfaces a structured retest at the right cadence, against the same rubric the original certification used. The original evidence trail — rubric, observation, instructor attestation, and an optional captured demonstration — is preserved so the recertification authority can audit any individual's record end to end.

What the AI never does

• The AI never certifies a competency on its own — the instructor attests.
• The AI never clears a safety-critical gate without explicit human sign-off.
• The AI never presents safety guidance without an authoritative citation.
• The AI never issues or revokes a credential autonomously.
• The AI never represents itself as the instructor or as a safety authority. It carries the educator-decides posture in every disclosure surface.