Methodology
How a job role is built, versioned and retired
A level is only as good as the model behind it. This page is the model of the models: written so an employer can audit us and a participant can argue with us.
1 · Where the skills come from
Live job descriptions and the work people actually did last quarter, not a taxonomy. Anything appearing in fewer than half the sampled roles is cut.
Each skill must be observable, or it goes back. If two verifiers cannot look at the same submission and reach the same level, the skill is written badly.
2 · How the target level is set
Shared with another function: L2 Caterpillar. Solely accountable: L2 Caterpillar or above.
The level at which the role can be trusted to own the outcome: set by three or more employers, reconciled in the open. Where they disagree, we publish the disagreement rather than average it away.
3 · Versioning and what happens to your level
Every model carries a version and a review date. Issued levels keep the version they were issued under. Your issued levels do not change retrospectively.
If a new version moves a target above your level, coverage drops and we name the version that did it.
4 · How to challenge a model
Anyone can submit a challenge against a skill, a descriptor or a target, with reasoning. Challenges and their outcomes are published against the model.
2 challenges upheld to date, each published below with its reasoning.
5 · Changelog: what challenges changed
Every challenge we resolve is published here with its outcome and the reasoning behind it, so a model's history is auditable in the open.
Declined. Intent classification is assessed separately from execution on the page, and the evidence submitted for each is different in kind. The skill stands.
Upheld. Accessibility requirements are now part of the L2 descriptor, effective from v1.1.
Partly upheld. The L3 descriptor now requires attribution modelling and a decision taken from the data. The target stays at L3 for marketing roles.
6 · Open a challenge
No account needed. Reasoning required; anonymity available. We publish the decision, and why.
On this page
Challenge a model
You do not need an account. Reasoning required, anonymity available.
Open a challenge