A status is not a score

Every checking point CONNECT Mastery runs ends in one of six words:

• Passed — the field matches the standard.
• Passed with Warning — it matches, but with a caveat worth a look.
• Failed — it does not match the standard.
• Cannot Check — there was nothing to compare against.
• Extracted — the value was read, but not yet compared.
• Not Found — the corresponding field was not there.

That is the entire vocabulary. There is no seventh status, and there is no number beside any of them.

What a result actually looks like

{
  "field": "Active date",
  "status": "Failed",
  "explanation": "The document's active period ends before the article production date."
}

A field, a status, and the reason. Nothing to interpret and nothing to threshold.

Why not a percentage

A percentage feels more informative than a word, which is exactly the problem. It reads as a measurement, so the reviewer treats it as one — and then has to invent a threshold. Is a mid-range value a pass? Does it depend on the document type? On the field? Nobody knows, so in practice everybody picks their own line and the review stops being repeatable between two people.

Worse, the number would not be measuring what it appears to measure. A confidence figure describes how sure the extractor was that it read the value correctly. It says nothing about whether that value conforms to the Connect TSS. Those are two different questions, and stapling the first one to the answer to the second is how a document ships with a comfortable-looking figure beside a value that does not meet the standard.

Cannot Check is the load-bearing one

The status people miss on first read is Cannot Check, and it is the one that keeps the rest honest.

The naive design has four outcomes: pass, warn, fail, and — for everything else — fail. That is tempting, because it makes the summary look decisive. It is also wrong: "this value does not meet the standard" and "there was no standard to compare it against" are different findings with different fixes. The first goes back to the material producer. The second means the Connect TSS row is missing, or the document type was not what the pipeline thought it was, and no amount of chasing the supplier will resolve it.

Collapsing the two produces a review queue full of failures that are not failures, which is how a checking tool trains its users to skim it.

What a reviewer does with each one

The test of a status vocabulary is whether each word maps to a different next action:

Status	What you do
Passed	Nothing. That is the point.
Passed with Warning	Read the line. Decide once.
Failed	Go back to the document's source.
Cannot Check	Fix the inputs — usually the Connect TSS or the document type.
Extracted	Nothing yet; the comparison has not run.
Not Found	Find out whether the field is genuinely absent or was missed.

Six words, six distinct actions. A percentage collapses all of them into one action — "use your judgement" — which is the action a checking tool exists to remove.

The one we would add

None, currently. If a seventh appears it will be because a real review action exists that none of the six describes, and the entry that justifies it will say what that action is. That is a higher bar than "the model was unsure", and it is deliberately higher.