Apprenticeship in which accepted output helps pay for supervision.

Economists have a well-known explanation for why firms underinvest in skills that transfer. If a trained worker can leave, the firm that paid for the training captures only part of the return, so it trains less than would be worthwhile overall ^[2]. The sector programs with the largest effects appear to work partly by supplying exactly this kind of transferable, certifiable training ^[2]. Small Ontario electrical contractors described the same problem: after bearing the cost of training apprentices, they lose them to larger employers that can pay more ^[3].

That creates a tension at the centre of this architecture. Portable evidence makes workers more mobile, which is good for them, and it thins the individual trainer’s return. The design has to pay for both. Four mechanisms can help. Pooling the cost across several employers helps, as the results for group programs suggest ^[1]. Public funding for the transferable parts of training helps too; grantees that helped employers pay for classroom instruction made more progress registering apprentices, though that association isn’t causal ^[1]. Paid output during training covers part of the cost directly. And a trainer’s reputation for producing capable people can become an asset in its own right (Shift 2).

Portability helps the worker and thins the trainer’s return. The design has to pay for both.

A trades apprentice produces output that is easy to see. Knowledge work is less forgiving. A junior developer’s code needs review and may introduce security risk. A junior analyst’s research may be useful but hard to separate from the supervisor’s guidance. Outputs are ambiguous, data is confidential, feedback is delayed and productivity depends on the team, so the strongest evidence on returns doesn’t transfer automatically to distributed knowledge work.

The right question is therefore not whether apprenticeship works for knowledge workers. It is which task families have output visible enough for learning inside work to pay.

Illustrative: in an apprentice’s first weeks, accepted output may be worth less than the supervisor’s review time. If review time halves and accepted output doubles over the following months, the same arrangement turns positive. For any task family, the question is how quickly that crossover arrives and who carries the cost until it does.

Small firms often have real work and little else: no HR function, no training system and no spare supervisors. In focus groups, small Ontario contractors named training costs, administrative burden, supervision ratios, poaching and apprentices’ limited readiness as barriers. They valued an industry association’s mentoring, hiring resources and onboarding tools ^[3]. The review found the same pattern more broadly. Intermediaries reduce employers’ setup and administration costs, and the need for an intermediary shows that an open network has to pay for coordination rather than eliminate it.

So the network’s role isn’t to remove coordination. It is to make coordination cheaper and more reliable than each small firm doing it alone.

Registered apprenticeship handles supervision, liability, intellectual property, wages and standards through a sponsor or employer that answers for the apprentice. Suppose an apprentice learns one micro-skill with Trainer A and another with Trainer B, then works on Employer C’s project. That arrangement removes the boundary, so the governance has to be rebuilt explicitly. Someone has to be the employer, supervise, own the deliverables, carry liability, protect confidential data, pay the wage and receive credit for the training. Without an accountable sponsor, the review warns, an open network can drift into a marketplace for unpaid or underpaid trial work.

Four rules follow. Every apprenticeship in the network is paid. Every apprenticeship has one accountable sponsor, usually the capability family’s intermediary. The worker’s employment status is settled for each jurisdiction before launch. And intellectual property and confidentiality terms are standard across the capability family rather than negotiated case by case.

Every apprenticeship in the network is paid, and every one has someone who answers for it.

Training can also become measurable. “Trained 500 people” says little. A stronger record might read, illustratively, 500 apprentices trained, 467 independently passing assessment and 421 later performing well in paid work ^[CN]. That would reward trainers for producing people who can perform rather than for volume. It remains a direction rather than a market, because attributing later performance to one trainer is hard and small cohorts give noisy results.

Registered apprenticeship, apprenticeship intermediaries, community-college partnerships and group programs shared across employers all exist. So do rigorously evaluated technology-sector programs such as Year Up and Per Scholas, whose earnings gains have lasted for years ^[2]. The evidence used here exists because these programs invested in measuring themselves. What this architecture adds is apprenticeship spread across trainers and employers that produces evidence the worker can carry, with the governance rebuilt around it.