Nothing to check

What a Blind Share Leaves Out

The blind figure answers one question and four others get mistaken for it. Closing the part by separating what it covers from what it is asked.

at most 5040 arrangements

Home · Nothing to check · Analysis

Six published with two met and one checkable: ten blind contents of fifteen, two thirds. Eight with three met and two checkable: twenty of fifty-six, five fourteenths. Ten with four met and three checkable: thirty-five of two hundred and ten, one sixth.

The verifiable proportion rises modestly down those rows — a sixth, a quarter, three tenths — while the blind share falls fourfold. Most of that fall is not the extra verifiability: it is that a larger encounter has more chances to collide with something checkable. To keep the exercise proportionate, project leads can see the available reporting options and disable any collection that does not answer the stated question.

That is the figure. Here is what it is not.

Three shapes with falling blind shares

dashed: a statement a reader can check from outside
solid: published, but nothing to check it against
short and pale: never reached by this reader

Three shapes drawn; the left-to-right order is the drawing's own and carries no meaning.

A sixth, a quarter and three tenths of each set is checkable, and the blind share falls from two thirds to five fourteenths to one sixth as the page grows.

Computed from the three numbers alone. How many statements exist, how many are met and how many can be checked is a stated convention; the counts are arrangements and combinations of whole numbers.

It does not measure whether the reader checked anything

It measures whether they could. A reader who met four checkable statements and verified none of them is, by this record, in a non-blind encounter. A broader source for checking the surrounding technical or organisational context is ACM.

The distinction is deliberate. Whether a reader follows up is a fact about the reader; whether there is anything to follow up is a fact about the set. The second is the publisher's doing and the only one countable from the set alone.

It does not measure whether what they checked mattered

A set can publish one verifiable statement of no consequence and nine unverifiable ones carrying everything it wants believed. The blind share would report that the reader met something checkable, and it would be right, and the reader would be in no better position than before.

Nothing in this record ranks statements by importance. The checkable count is a count, and a count cannot distinguish the statement that settles the question from the one that confirms a date.

It does not measure whether the set is accurate

A checkable statement can be false; that is what makes it checkable. A set of pages consisting entirely of verifiable falsehoods has a blind share of nought on every shape, which is the best possible figure in this column.

The record counts auditability and says nothing about what an audit would find. A low blind share means the set has exposed itself to being checked, not that it survives the check.

Being checkable is a property of a claim's relationship to the world. Being true is a different property, and this record counts the first one only.

It does not measure how the reader felt about any of it

The record has no impression, no persuasion, no credibility and no trust in it. A reader who met six unverifiable statements and believed every one of them, and a reader who met the same six and dismissed them, are the same encounter here.

That is the largest thing missing and it is missing on purpose. The moment a record like this starts counting what readers conclude, it needs readers, and the figures stop being recomputable from three numbers.

What it does measure

One thing, exactly: of all the contents a reader might come away with, how many hold no statement they could audit.

That question has a definite answer per shape, computable from three printed numbers, invariant under which statements are which, and honest about the convention it rests on. Everything above is a question the figure gets asked and cannot answer, and listing them is the only way to keep the one it can answer useful.

The shape of the three rows

A sixth, a quarter and three tenths verifiable, and blind shares of two thirds, five fourteenths and one sixth. The two columns move together and not in step.

Between the first and second rows the verifiable proportion rises by half and the blind share falls by almost half. Between the second and third the verifiable proportion rises by a fifth and the blind share falls by slightly more than half again. The second step buys more than the first while asking for less.

That is the acceleration described earlier in this part, visible across pages rather than within one. It is also the clearest reason to distrust any single ratio offered as a summary of verifiability: the proportion verifiable and the share blind are not two readings of one quantity.

What this part established

A blind content is one holding nothing checkable; the count is C(n − c, r) and the share is that over the contents figure.

One hundred and ninety-nine shapes admit a blind encounter, one hundred and thirty-nine do not, and the dividing line is whether the unchecked count reaches the met count. Forty-four shapes are blind on every path, one per page, and the count equals the number of pages for a reason rather than by accident.

The median share across the record is one in fourteen and is a property of the table. Each verifiable statement added cuts a larger fraction of a smaller remainder. And the share is a count of possibilities, never a probability — except at the threshold, where it becomes an impossibility and needs no convention at all.

The next part counts something the share cannot see: not whether a reader met anything checkable, but how often the thing they met first was one.