Assess across a client boundary¶
Consultancy work, cross-business-unit reviews and partner estates all share a condition: you are asked about repositories you cannot read.
The instrument handles this by naming what it could not see, rather than quietly assessing whatever happened to be reachable.
Name every subject, including the unreadable ones¶
Put all fourteen in the manifest even if six are readable:
version: 1
team: client-payments-tribe
subjects:
- id: orders-api
path: ./orders-api
- id: billing
path: ./billing
# ... four more you can read ...
- id: settlement
path: ./settlement # no access
- id: partner-gateway
path: ./partner-gateway # different business unit
# ... six more you cannot ...
The manifest is the claim about scope. Leaving the unreadable ones out would make the report say "6 of 6" — which is true about the manifest and false about the estate.
What the report does¶
The coverage ledger reports against the named subjects:
| Subject | Status | Report date | Note |
| orders-api | assessed | 2026-08-02 | — |
| settlement | unreachable | — | no access |
| partner-gateway | unreachable | — | different business unit |
Assessed 6 of 14 named subjects; 8 unreachable.
The denominator is the claim. "Assessed 6 subjects" and "assessed 6 of 14 named subjects" are different statements, and only the second one is honest about what was left out.
Under partial coverage the report also never describes the ceiling as estate-wide. It is the ceiling of the six that were read, and the report says so — because a common-weak dimension across six repositories is much weaker evidence about fourteen than about six.
Why not just assess what you can see¶
Because the result gets quoted without its caveats.
A portfolio report is a persuasive artefact — the matrix in particular travels well and travels alone. If it does not carry its own coverage on its face, "our estate is Distributed with an observability ceiling" becomes a sentence about fourteen repositories on the strength of six. The coverage ledger sits first in the report, above the matrix, for exactly this reason.
The engagement pattern¶
Where teams can run the assessment but you cannot read their code, the roll-up is designed to work from artefacts:
- Each team runs
/ai-readiness-assessin their own repository. - Each team keeps their own report — it stays in their
assessments/directory, and you never need access to their code. - They share only the report, or only its summary block.
- You roll up what you have received.
This separates who runs the assessment from who reads the summary, which is usually what makes a cross-boundary engagement possible at all. Teams that will not grant repository access will often happily share a report they generated and can read themselves first.
Reassessing later¶
Subjects that were unreachable stay in the manifest. When access
arrives, or a team sends their report, they move from unreachable to
assessed and the coverage sentence improves on its own.
An estate that goes from "6 of 14" to "11 of 14" between reviews has made real progress in visibility, and that is worth reporting alongside any change in the levels themselves.