Subject, habitat, team¶
The single-repo instrument assumes the repository it runs in is the habitat under examination. That assumption holds in exactly one case, and quietly conflates three different things everywhere else.
| Concept | What it is | Natural scope |
|---|---|---|
| Subject | The code artefacts the habitat governs | A repo, a submodule, a directory in a monorepo |
| Habitat | The harness and control surfaces governing it | May live in the subject, a sibling repo, a parent, a package, or an org |
| Team | Whose behaviour produces the cognitive read | People — never a repo |
Separating them gives one primitive that covers every shape:
Assessment unit = subject × governing habitat, with the cognitive read scoped to the team.
Why the team is not a repository¶
The cognitive read measures what people can think and do — how directive their relationship to the agent is, what they actually inspect before accepting work, how much of a unit of work they hand over. None of that is a property of a directory.
A team of six that owns four services does not become four teams. They carry the same habits into all four. What changes between the four is the environment those habits meet.
Which is why the scope run asks the behavioural questions once. Asking them per repository does not produce four readings; it produces four copies of one reading, at four times the cost, with four opportunities to answer slightly differently out of fatigue.
Why the spread is the finding¶
One cognitive read measured against four habitats produces four gaps. Their spread is a signal the single-repo instrument cannot produce, because a single assessment has only one reading to compare.
A team can be Coherent in the greenfield service and a full level and a half ahead of its environment in the legacy one. Both facts are true at once, and the work they imply is opposite: hold the first, invest in habitat for the second. Average them and you get a number describing neither — which is why there is no portfolio score.
When practice genuinely does differ¶
Sometimes it does. A team really can work one way in the service they deploy daily and another in the batch job they touch twice a year.
So each subject gets exactly one cheap question — is your way of
working here materially different? — and if the answer is yes, the six
behavioural dimensions are asked again for that subject. Its report
records cognitive_source: subject, and the portfolio shows its gap
separately from the team's general pattern.
One question is the right price. Asking nothing assumes uniformity that may not hold; asking everything again assumes difference that usually does not.
Reuse is always declared¶
Where the team read is applied to a subject, the report says so in its body — not in a footnote.
This matters more than it first appears. A reader picking up the report for one repository has no way to tell a placement gathered there from one carried in from a conversation about a different repository. Both render as "L3". Silent reuse is not a shortcut; it is a false claim about where the evidence came from.
What this does not change¶
The model is untouched. Fourteen dimensions, the same verbs, the same six-level ladder, the same three disciplines, the same gap formula and the same three regimes. This is a change of unit, not of measure — and with no scope manifest present, nothing about a single-repo assessment is different at all.