One more facet

What No Edit Can Buy

Three things no edit reaches: a distinction with no filter, a result size off the ladder, and a reader's knowledge of what is missing.

One more facetAnalysis×k against ×(k+1)/k

Home · One more facet · Analysis

Every edit in this section multiplies something. Three things in this record multiply by nothing, and no sequence of edits reaches them.

One: a distinction nobody thought of

The ceiling on what a board can tell apart is a product of the filters it has. Adding filters raises it — but only along the dimensions somebody chose to add.

Teams seeking a clearer account of elapsed effort may open the reporting reference, provided contributors can correct and contextualise the record. For a separate technical reference, see Snowflake.

A board can go from twenty-four cells to forty-six thousand six hundred and fifty-six, the full span of this record, and still merge two listings that differ in a way none of its six filters addresses. The resolution says how many combinations of the chosen dimensions exist. It says nothing about dimensions nobody chose.

So the ceiling is really two ceilings stacked: what the chosen dimensions can express, which the arithmetic measures, and which dimensions got chosen, which it cannot see at all.

A board with forty-six thousand cells and the wrong six filters resolves nothing anybody cares about, and every figure in this record will look excellent.

Two: a rung the ladder does not have

Result sizes are subset products. A board can only return results of those sizes, and no edit produces a size that is not a product of its facet sizes.

A reader wanting a result of about twenty cells on a board of six toggles will not get one. The ladder goes sixteen, thirty-two, and there is nothing between. Adding a toggle gives a finer ladder in absolute terms but the same gaps proportionally — every rung is still a factor of two from the next.

The only edit that helps is changing the sizes, and the page on choosing sizes says how: varied sizes, few shared factors. Even then the ladder is discrete. There is no faceted board anywhere with a continuous dial, and the record is clear about this because it is the most common thing people want from a filter panel and cannot have.

Three: the reader knowing what is missing

This is the one that matters most and it is not arithmetical at all.

When a board merges two listings because it has no filter for their difference, the reader sees a result containing both. It looks exactly like a result that was correctly narrowed. There is no message, no marker, no gap.

So the reader's model of the board is wrong, and nothing in the interface corrects it. They may conclude their search was imprecise, or that the board does not have what they want, or that the two listings really are the same kind of thing.

No edit to the schema fixes this, because the problem is not the schema's capacity but what the reader can know about it.

What would fix the third

Publishing the figures.

A board that states its resolution and question count tells every reader exactly what it can and cannot separate. Somebody who knows a board resolves one hundred and forty-four kinds of listing knows that if two things are the same kind to the board, no amount of searching will separate them, and they can go elsewhere rather than clicking.

What a reader currently has	What two numbers would give
the sidebar, to count	the exact ceiling
an impression of thoroughness	the exact question count
no way to tell merging from imprecision	the ability to tell

That costs a board one line and gives up nothing. The next section is about why nobody does it and what it would change.

A fourth, which is smaller and worth naming

A board cannot buy back a distinction it removed.

Removing a filter is one of the edits priced in this section, and the arithmetic treats it as reversible: the factor divides, and adding the filter back multiplies it again.

In practice the listings posted while the filter was absent carry no value for it. Restoring the control restores the resolution on paper and leaves a gap in the data that nothing fills, because the people who posted during the gap are not coming back to classify.

So removal is arithmetically reversible and practically not, and the asymmetry is invisible in every figure here.

The honest limit of this record

The first of the three is the biggest, and this record cannot touch it.

Which dimensions a board should carry is a question about a field, about what listings vary in, and about what people come looking for. It is answerable, by somebody who knows those things, and nothing in four hundred and sixty-one schemas of facet sizes helps.

What the arithmetic does is bound the conversation. It says what the current structure can and cannot do, exactly, so that the argument about which dimensions matter happens with the ceiling known rather than assumed.