What goes wrong with levels
Most level systems are designed backwards. Somebody decides there should be five levels — five is a comfortable number, it fits a brochure, it matches the neighbouring programme — and then content is distributed across them until each looks full.
The result is levels that describe an administrative division rather than a capability threshold. Learners cross them without anything changing, or hit a wall in the middle of one because the real threshold sits somewhere the grid did not put a line.
There is a worse version, which is where this system started. If a level is allowed to encode several things at once — age, cognitive depth, course length, certificate, cohort — it becomes impossible to move any one of them. That was the design we replaced, and the reason the current system separates stage from level so insistently.
Three parts, and all three are required
A descriptor states what the learner knows, what they can do with it, and how independently. That last dimension is the one that most distinguishes a genuine threshold from a topic boundary — skill-acquisition accounts describe progression precisely as a shift from following rules, through situational judgement, to fluent practice[18], and structural accounts of learning outcomes describe the same movement from handling one element to relating several and then generalising beyond the task[20].
A volume states how much work this actually represents, in notional learning hours. Without it, a level is a claim with no scale attached, and two programmes calling something “Level 2” may differ by a factor of five.
An artifact is the assessable piece of work that evidences the level. This is the part that goes missing most often, and its absence is the clearest signal that a level was designed as a container rather than as a threshold. If nobody can say what a learner makes to demonstrate a level, the level is not finished.
Derived, not chosen
The method is deliberately mechanical, because that is what stops the number drifting.
- Anchor the entry. No prior making in this subject.
- Anchor the exit. The skilled maker — someone who can independently take a self-chosen project in this subject all the way to the completion standard. That is an observable event, not a duration.
- Count the genuine capability thresholds between them, dividing by the volume of one deliverable block.
The number that falls out is the number. Sometimes it is three. Sometimes five. Once it was two, because the thing in question turned out not to be an independent progression at all but the upper half of one subject joined to a module of another — which is a useful finding in itself, and one a pre-chosen grid would have hidden.
The grades describe the delivery context — who the material was written for — rather than the levels themselves. Used directly as level labels they would reintroduce age banding, so the progression is restated in capability terms first.
Why there is no tidy grid
A table in which every subject has the same number of levels looks authoritative. It is also an assertion — that depth is a single universal ladder and that all subjects are the same shape — and it is an assertion this system explicitly denies.
So the subjects do not line up into equal columns, and we do not make them. Above the Maker stage there are no levels at all: the Innovator and Changemaker stages carry named milestones instead, because at that point the useful question stops being “how deep” and becomes “did it work, and for whom”.
The same discipline applies to the empty cells. Where a subject has real, continuous delivery but no published progression, the entry says there is no level count yet. A number in that cell would imply a placement check, an exit condition and an assessable artifact at each threshold, none of which have been written.
Levels belong to programmes, not to people
This is the consequence that matters most in practice. “Robotics Systems — Level 3” is something you enrol in. “A Level 3 person” is not a thing.
Robotics 4 · AI 2 · Music 1.
Three facts about one learner, each independently useful, and no useful way to average them into one grade.
A learner’s depth reading is their maker profile. It is what lets an experienced software learner start electronics at the foundation without being treated as a beginner in every sense — and it is why the record is per subject rather than per person.