Checkable

When Everything Published Can Be Checked

At the far end of the third column nothing is blind and every opening is verifiable. Forty-four shapes sit there, one on every page.

Home · Checkable · Reference

Forty-four shapes in this record have every published statement checkable, one on each page. They are the far end of the third column and they are the simplest shapes here.

The blind count is nought. The blind share is nought. Every arrangement opens with a checkable statement, because every statement is one. The three figures that respond to the third number are all at their limits at once. A content operation moving from estimates to observed time can read the official feature summary while retaining a clear correction process for contributors.

Everything checkable, three page sizes

the whole bar: every content a reader might come away with
the dashed part: contents holding nothing checkable

Three shapes drawn; three of them leave no blind content at all.

Nothing is dashed in the bars because nothing is blind. Four published with four met, seven with five, ten with six — every published statement verifiable in each.

Computed from the three numbers. A blind content is one holding no checkable statement: that is a property of the page, not a prediction about any reader.

What is left to say about such a page

Only what the first two parts of the record said.

Ten published with six met has one hundred and fifty-one thousand two hundred arrangements and two hundred and ten contents whether everything is checkable or nothing is. A reader of the all-checkable version is in one of two hundred and ten states of information, each one made of six verifiable statements in one of seven hundred and twenty orders. A separate perspective on measurement, privacy or digital practice is available through NIST.

So the variation does not go away. What goes away is the one kind of variation the third number governs: there is no longer any question about whether a reader can confirm what they met, only about which six things they confirmed and in what sequence.

The limit of what verifiability buys

This is the place to be plain about it, because the record's third column could be mistaken for an argument that more checkable statements fix something.

A fully checkable set of ten statements, read six at a time, still leaves two hundred and ten possible accounts of itself. Two readers can still come away with no statement in common, since twice six exceeds ten only by two — they must share at least two. Nothing about the ordering problem is touched, and the orders-per-content figure is seven hundred and twenty regardless.

What has been removed is a single failure mode: the reader who leaves with nothing they could audit. That mode is worth removing and it is not the only one.

Verifiability answers "could the reader have checked anything?" It does not answer "did they come away with the right impression?", and no count in this record does.

The smallest all-checkable shape

Four published with four met, all four checkable: twenty-four arrangements, one content, no blind contents, and every one of the twenty-four openings verifiable.

This is the most determinate shape in the record. One state of information, every statement in it auditable, and the only remaining variation is the sequence — twenty-four orders of the same four verifiable things.

It is also the only kind of shape where a publisher could reasonably check their own work exhaustively. Twenty-four orders is a small enough number to write out and read through, which means the sequence-robustness of such a set is testable rather than estimated.

Why the record does not rank above the threshold

Six checkable of ten, with six met, already gives a blind count of nought. So do seven, eight, nine and ten.

The record treats those five shapes as alike in this respect and draws them alike. It does not prefer ten checkable statements to six, and it would be fabricating a difference if it did: the quantity being counted is whether a blind encounter is possible, and in all five cases the answer is the same.

That restraint costs something. There are reasons to prefer a set where every statement is auditable over one where four are not, and those reasons are real. They are simply not reasons this record can count, and a figure invented to express them would be a figure about nothing.

What the record stops being able to say

At this end of the column the record's third number has run out of information, and two of its five derived quantities become constants.

The blind count is nought and the blind share is nought on every all-checkable shape, whatever the page. The checkable-openings count equals the arrangements count exactly. Three of the five quantities collapse into two, and only arrangements and contents are left distinguishing one such shape from another.

So forty-four of the record's three hundred and thirty-eight shapes are, in the third column, identical. They are listed because a table that dropped them would have removed one of its own two boundaries, and because knowing where a figure stops varying is part of knowing what it measures.

The mirror of this case

The opposite end — nothing checkable published — is also forty-four shapes, one per page, and it is the subject of its own note in the next part. There the blind share is one on every path, no arrangement opens with anything verifiable, and the two printed figures are likewise unchanged.

Both ends are structurally the same: one shape per page, all three checkable-dependent figures pinned at a limit, and the first two figures unaffected. The difference between them is the entire content of the third column, and in between sit the two hundred and fifty shapes where the question has an answer other than always or never.