Back to the observability problem
Definition
Warrant: Foundation
Industrial capability cannot be described in fewer than eight dimensions.
This is a property of the problem rather than a choice. Eight is not a taxonomy anyone found convenient. It is the count of independent things that have to be known before a claim about what a person can do is decidable at all. Every record industry currently keeps collapses it to one or two, and then asks a decision worth millions to rest on the projection.
01 · The eight, in full
Any description of what a person can do resolves onto eight dimensions.
They are independent in the only sense that matters here: knowing all of the others does not tell you this one. A worker's competency does not tell you the standard the work was accepted against. The standard does not tell you the regime that governs it. The regime does not tell you what happened on the plant six months later.
Eight dimensions, of which most industrial records carry two. That is the measurement problem, stated exactly.
Definition
Three groups, one decidable description.
What is being described
Capability is exercised by a person, on a specific class of work, against a specific standard. Acceptance is standard specific, so the standard travels with the claim or the claim carries nothing.
What conditions the reading
It is exercised inside a jurisdiction and its regime, for an employer whose tolerances decide how much of it is expressed, and observed through some instrument. The same skill is a different legal quantity across a border, and the instrument travels with the reading.
What settles it
It is deployed into some environment, under load, at whatever hour the work falls, and it is followed by an outcome that either confirms the judgement or refutes it. The outcome is the only thing that calibrates anything.
Remove any one and the description stops being decidable.
02 · Why eight, and not fewer
The test is removal, and every removal fails.
A description is decidable when two things can be done with it and nothing else is needed. First, someone can say from the description alone whether this person can be put on this work. Second, someone can say what would have to be observed later for that answer to be shown wrong. A description that supports the first but not the second is a preference. A description that supports neither is a document.
Take any one of the eight out and one of those two halves stops working. That is the whole argument for the number, and it is why the number is not negotiable downward for convenience.
- 01 Remove the work Capability with no task attached is a vocabulary exercise. There is no decision left to be decidable about, because nothing has been proposed.
- 02 Remove the standard Acceptance loses its criterion. The same piece of work passes and fails depending on a document that is no longer in the record, and both answers look equally defensible.
- 03 Remove how it was observed The reading arrives with no error bar. A result from a rehearsed demonstration and a result from a live shift are then stored as the same result, and no later reader can separate them.
- 04 Remove what happened afterwards Nothing calibrates. The description can no longer be wrong, which means it can no longer be right either, and the distinction it was built to support quietly stops existing.
The same removal runs on the remaining four and returns the same verdict. Eight is not a ceiling anyone reached for. It is a floor nothing can be taken from.
03 · What the record carries instead
A certificate describes a standard and a date.
A curriculum vitae describes an employer and a claim. Two dimensions of eight, at best, and the other six are not marked absent. They are simply not there.
Set the four records the industrial economy actually runs on against the eight, and the pattern repeats without exception. Each one is honest about what it holds. None of them is honest about what it dropped, because none of them has a place to say so.
The standard, and a date. Silent on the conditions the work was really done under, silent on who observed it and with what instrument, silent on what the plant found six months later.
The employing organisation, and a claim. Written by the person being described, in language chosen to succeed, about work that the reader has no route to check.
The employing organisation again, and its tolerance. It records that someone was retained, which is a statement about an employer at least as much as a statement about a worker.
The work and the standard, on one morning, in a rehearsed setting. The closest of the four to the quantity in question, and still silent on deployment and on outcome.
A projection loses what it loses silently. Nothing in any of these four announces the six missing coordinates, which is precisely why capability claims from different sources cannot be compared, weighted, or checked against what actually happened.
04 · Exposed, and protected
Warrant: Derivation
Not everything built to carry the eight is defensible.
A list of assets is a comfortable document. A sort is not, and the sort is the one worth publishing. Once the eight dimensions have to be carried by real machinery, that machinery divides into two kinds, and only one of them is worth treating as a position.
One kind is rule-following generation over licensed text: read a description of work, match it against a structured record of standards, emit the structured pieces. It is real work, it has to be right, and it sits on the same automation curve as every other symbolic task. Its cost falls every year and its scarcity falls with it. That layer is pushed toward full automation deliberately, because the efficiency is real and the strategic cost is nil.
The other kind cannot be manufactured afterwards at any price, because acquiring it required being present while something happened that will not happen again.
| Component | Exposed or protected | Why |
|---|---|---|
| Structured standards record | Exposed | A structured index built over text that is licensed rather than owned. A derivative index is defensible only as far as the licence permits, and that question is open and stated as open. |
| Matching pipeline | Exposed | A rule-following generation task over that record. It rides the same automation curve as any other symbolic work, so its cost and its scarcity fall together, year on year. |
| Recording of the event | Exposed | Cheap sensing industrialises acquisition. Capturing what happened becomes ordinary, in the way executing code became ordinary, and ordinary things do not defend anything. |
| Observation to later outcome | Protected | The causal link between a characteristic observed at one moment and a real, later, non-repeatable outcome. It requires having been there, and no subsequent reasoning reconstructs it. |
The licence question above is not settled here and is not described as settled anywhere else. Standards text is licensed by the bodies that publish it, a structured derivative sits inside those terms, and the legal answer is one that has to be obtained rather than assumed. It is carried as an open question, in public, for the same reason every other open question is.
What is settled is the sort. The plumbing is exposed. The calibration is not. No quantity of investment in the exposed half produces a defensible position, and a company that mistakes the two spends years building the part that gets cheaper.
05 · Coverage, stated per dimension
Definition
Warrant: Open question
Naming the description creates an obligation to report it.
Eight dimensions is a definition, not an achievement. Stating what a full description requires immediately produces a duty to say how much of it is actually held, and to say it one dimension at a time.
A single coverage number averaged across the eight hides exactly what a reader needs: which dimension is thin. A representation strong on work and standards and quiet on deployment and outcome is not most of a description. It is a description missing the half that calibrates the other half. Averaging turns that into a respectable-looking figure, and the figure conceals the specific gap that would change the decision.
- Coverage, per dimension
-
How much of each of the eight the representation actually holds, reported separately, dimension by dimension, never rolled into one number.
Measured against the description the problem requires, not against the description that is convenient to produce.
- A thin dimension
-
One where the representation holds little. It is named, and named as thin, in the same document that carries the finding it constrains.
Regions not covered are reported as not covered, rather than interpolated across.
A description that cannot be wrong cannot be right either.
The eight dimensions are not an ambition for a richer record. They are the condition under which a capability claim becomes the kind of statement that can be checked, and therefore the kind of statement worth putting money behind.
Every assessment that leaves here carries a dated, checkable claim about what happens next, and the claim is revisited against what actually happened. That is a condition of the work, not a feature of it, and it is the eighth dimension doing its job.
How it is constructed is not described here, and will not be.