The problem appears to stem, not from training of engineers but from four problems ... almost orthogonal problems.
The first is that Openreach want faults to be reported, and handled, as either voice or data, not both, where demarcation is decided prior to any technical investigation, let alone impact assessment or resolution.
The second is an insistence that a DLM reset can only happen within a fault labelled as broadband, even when an engineer can see that a voice fix has impacted broadband performance.
These two combine to ensure that broadband connection can be left suboptimal, needlessly, because of simplistic administrative labeling.
The way around this problem is to order a separate broadband job. The third problem then turns up - Openreach are unwilling to take this on when the suboptimal operation still meets minimum standards.
No problem if DLM was seen to recover at some reasonable point in time. Leaving it to automation is a good answer. The fourth problem is that automation isn't doing the job here. The premise behind problems 2 and 3 - that DLM automation will recover - is turning out to be a false one. IMO banding is just too sticky right now.