The developer who needs a database with one setting the template does not expose ends up in the same place either way, writing a message to a person. Where there is no platform, that person is the one who knows how the database gets set up, which the CNCF post offers as the archetype of Level 1: knowledge passed hand to hand, provisioning by manual request [9]. Where there is a platform, that person sits on the platform team, because per the post everything outside the paved path still needs a human from that team to implement it [12].
That is the thing being done. The thing being pitched is the portal and its CLI, and in some cases an agent alongside them, in a room where adoption looks reasonable and the platform team is still fulfilling requests by hand [2]. The post's claim is that the two rooms are not opposite problems, and that the question worth asking is how developers interact with what has been built rather than whether a platform exists [3]. Read down the Interfaces levels and the shared plateau has a name: request fulfilment by a person, which Level 1 does openly and Level 2 does for everything off-template [16].
The model also refuses to give you a single grade. Five aspects are scored independently and each carries the same four levels, named Provisional, Operational, Scalable and Optimizing [5], which means an honest position is one of 4^5 = 1024 combinations rather than a headline number [15]. And it stops at locating you; by its own description it does not tell you how to reach the next stage [6].
The test the post offers for Level 3 is behavioural rather than metric: teams stop filing tickets for routine provisioning [14]. Sign-in counts and onboarding time cannot separate a paved road from a self-serviceable one, because the developer who logs into the portal and then files a ticket has done both.
That points to a count worth running instead: last month's requests, sorted on two axes, whether the request fell inside a documented template, and whether a platform engineer had to do anything. Inside with no engineer is the golden path earning its keep. Inside with an engineer is usually an approval or review step that can be deleted. Outside with an engineer is where the week goes. Outside with no engineer is what the model's Level 3 actually describes, one-click provisioning for most asks with maintainers out of the loop [13].
The recommendation is to clear the second cell and then the highest-frequency entries in the third, ahead of adding capabilities, and the tradeoff belongs in the same breath: every request you make self-serviceable is a guardrail you now own, and "little support from maintainers" [13] also means nobody reviewing the moment a developer misconfigures something. A month of counting that third cell yields both the Interfaces score the model asks for and an automation list in priority order.
