Skills coverage
A certificate records that someone passed. A job title records where they sit. Neither records what they have actually done, who taught them, or whether the one person who has done it twice is on the evening shift. When the work needs finding, the list you have is not the list you need — so who can actually do this?
What a capability is made of
recorded
Qualification
partly recorded
Experience
rarely instantiated
Tacit knowledge
Managers do write the second layer down, once or twice a year. Performance reviews are full of it: what this person handled, what they recovered, where they were the one who knew. But the review is written as prose for a conversation, filed as a document, and read once — so the most informed description of what someone can do ends up as disposable text instead of an asset the organisation can use. Nothing about the exercise is wasted except its output.
An ability is all three, and an organisation that can only see the first mistakes a record for a readiness.
Today
01 · what the records miss
A skill with exactly one holder on the evening shift is not a gap in any system — every field is filled in. The certificate is valid, the role is assigned, the roster is complete. What nobody can see is that the coverage is one deep, because coverage is not a field. It is a count across a relationship.
02 · both sides lose
This is not only a staffing problem, and saying so is what separates this page from a workforce-planning pitch:
the organisation
the person
03 · the way out
The answer is not a better survey. Asking people to self-report tacit knowledge produces a list that is out of date the week it is collected, and it asks them to describe the thing they are least able to articulate. What is needed is not a new record. It is the relationships you already have, typed.
One model, one query
as a field
Someone types a skill name onto a profile. It is a claim, maintained by hand, and it decays the moment the person moves on.
as a relationship
The person is connected to the processes they ran, the assets they operated, the incidents they resolved. Nobody asserts the skill — it is the shape of the edges around them.
Which is why the question stops being who is certified and becomes a traversal: start from the work, follow what it needs, and count who reaches it.
Process
Specialized Paint
Technical Skill
Metallic Basecoat Application
People · and nobody else
Ren J Ishikawa-Reed
Three hops, and the finding is not in any of the three nodes. It is in how few of them there are. A count across a typed relationship is something a model can be checked against; it is not something a profile page can hold.
Unprompted
01
Detected
02
Routed
03
Decided by a person
04
Recorded
A skill with one holder is not a gap in any system. Every field is filled in.
Autonomy of attention, not of authority. The agent identifies, reports, monitors and escalates. It does not change your organisation.
What it takes
You do not begin by naming every ability in the organisation — that project is the reason most skills initiatives are abandoned. You begin with the processes and assets you already model, and the relationships between them and the people who ran them. The ability emerges from the edges.
And the honest limit, which is about filling and not about structure: the model has a type for tacit knowledge, and almost nobody has ever written one down. A model can show that experience is concentrated, who holds it, and where the coverage is one deep. What it cannot do is populate itself — nothing infers what someone knows and never had to use. What it can do is tell you exactly where to go and ask, and hold the answer once you have it.
Ask the model
Adventure Works is modelled in the demo tenant, down to the shift. Ask which capabilities have one holder, and read the path it answers with.