Version control for training content across sites

Most operators treat training material the way they treat a Word file. There is a current one, and the previous ones are somewhere: an inbox, a shared drive, a folder called old, a cupboard in the site office. That holds until somebody asks a question with a date in it: which version of the procedure was this person actually trained on, in March, at this address. In a network of more than a handful of sites, the honest answer is usually that nobody can say.

This applies well beyond franchises. Contractors with crews on client sites, production groups with several plants, facility companies, retail chains and agencies placing people across dozens of clients all run the same instruction in more than one building, and certified ones get audited on it.

The record that cannot answer the follow-up question

A completion record is a claim: this person completed this instruction on this date. It survives the first question at an audit without difficulty. It rarely survives the second one. If the procedure was revised in February and the record says someone completed "the safety instruction" in January, it tells you they were instructed on something, not on what is on the wall today. If a colleague at another site completed the same named instruction in April, the two records look identical and describe two different pieces of content.

That is the whole argument for version control, and it has nothing to do with software hygiene. Only the version turns a record of attendance into evidence of content, and only evidence of content is worth anything when something has gone wrong.

What the signature and the toolbox list point at

Belgian law makes this concrete. The onthaal of every new worker ends in a document signed by a named member of the line, stating that the necessary information and instructions were given. A named person is signing a statement about content. If that content is untracked, two workers who started six weeks apart can hold identical signed documents describing materially different instruction, and nothing in the file will say so. Codex art. I.2-21 adds why this keeps happening: training must be repeated and adapted as risks evolve, and a change of function, new equipment or new technology triggers it again. The content is supposed to change.

The Netherlands prescribes no form of proof, and anyone claiming the Arbowet requires a signature has invented it. It does require the employer to supervise that instructions are followed (see the instruction law guide), and supervising an instruction you cannot identify precisely is an assertion, not a practice.

Certified companies meet the same weakness in the audit file. VCA question 4.1 asks for a list of toolbox dates and topics plus attendance lists, and names changes to rules and procedures as required content (the toolbox guide covers the rest). A topic is a label, and the scheme quietly assumes it points at one thing. When four sites record "the revised loading procedure" and two of them presented the version withdrawn a month earlier, the file is complete, the lists are signed, and the record is wrong in a way no auditor can see.

Changeovers: where two versions meet on one line

A production operator who runs the same product on the same line for a year needs very little from you. Everything that goes wrong sits in the moments the job stops being the same: the changeover. Different settings and speeds, different guarding because a format part moved an interlock, a different cleaning regime between runs, different rejects, labels and paperwork.

The written procedure usually covers all of that. The gap is arrival. The realistic test is the third shift: the line leader who ran the trial and the technician who moved the guard are at home, and the operator has a format he last ran two months ago and a target on the screen. What he does next depends on what he was given in advance, in the language he reads, for the version that is actually on the line tonight.

The failure most sites recognise is not a missing course. It is two operators on the same line working from different versions of the same procedure, both convinced they are right, because one attended the session in March and the other joined in June.

Where the changeover is also a food safety event, such as an allergen switch, a metal detector check or a seal check, the bar rises. As of September 2026, BRCGS Food Safety Issue 9, clause 7.1.2, asks for training and a documented competency assessment for staff working on critical control points: competence assessed and written down, not attendance.

Drift is a deployment failure, not a preference

Sites do not decide to run old content. Drift arrives quietly, and each cause has its own fix.

None of these is fixed by an email. What they share is that they are only fixable if somebody can see them. The number to watch is not the completion percentage; it is the spread of active versions across the network. One version everywhere means the last change landed. Three versions in fifty sites means it is still in progress, whatever the completion figure says.

When a published version turns out to be wrong

An update goes out on Monday. On Wednesday someone notices step four is ambiguous, or a value was carried over from the old process, or a control got dropped in the edit. From that moment every extra completion adds a person correctly recorded as trained on the wrong thing.

Most bad versions are not authoring mistakes; they are review gaps, where someone with edit rights fixed something reasonable and it went live without a second reader. Split the two rights. Anyone close to the work can propose a change to the source procedure, because they notice first when reality has moved. Publishing to the network is held by few people, with a second reader on the content where an error is expensive: safety instruction, hygiene and allergen procedures, anything a client rulebook governs.

Questions to ask of any training system

Whatever you use, test it against the dated question before you need it.

How Aristotl handles versions

Aristotl builds the course from the procedure you already maintain, so there is one source per procedure rather than a deck per site. Nothing reaches a site until your team reviews and publishes it, which keeps the publish right with the people who hold it.

Completion and knowledge check results are viewable per site and per version in the platform, so a completion on the previous procedure shows as exactly that, and the group that completed a flawed version is a list you can read; the corrected version then goes to the sites or teams it covers. Content is translated per worker from the same master, so a Polish and a Dutch operator get the same changeover instruction at the same time, on their own phone, in three to five minutes.

What it does not replace: the documented competency assessment for CCP work stays an exercise on the line, observed and dated by a named colleague.

Take any training record from twelve months ago and try to establish which version of the procedure it refers to. Bring the result to a demo, and we will show you the same kind of record with a version on it.