Telling apart

The Ceiling Is Set Before Anything Arrives

The filters are chosen once; the listings arrive afterwards, for years. Everything countable about a board was fixed in that first decision.

Telling apartAnalysis2 to 46 656

Home · Telling apart · Analysis

There is an order to how a board comes into being, and the order matters more than anything that follows it.

Somebody decides what the filters will be. Then listings arrive — for months, for years, in any quantity, from anybody. The second part is the visible part, the part that changes daily and the part everyone talks about. The first part happened once, probably quickly, possibly without much deliberation, and it fixed both ceilings this record measures.

A data team can use transparent workforce analytics to connect project hours with deliverables, provided the report remains one source of context rather than a substitute for technical judgement. For a separate technical reference, see scikit-learn.

What was settled in that first decision

The count of listings the board can ever tell apart. The count of different questions anybody can ever put to it. How many different sizes of result are reachable. The share of available questions that narrow to a single answer.

All four are products of the facet sizes. None of them involves a single listing. A board with no listings at all has the same four figures as the same board three years later carrying two hundred thousand.

Content is the part that moves and the part nobody can summarise. Structure is the part that does not move and can be written down completely in two numbers.

Why the first decision gets made badly

Not through carelessness. Through sequence.

At the moment the filters are chosen, there is nothing to test them against. No listings exist yet, so nobody can see which distinctions will turn out to matter, which filters will sit unused, or which axis everybody will want and nobody thought of. The decision has to be made from a guess about a population that does not exist.

Afterwards the position reverses: the listings exist in quantity, the gaps are obvious, and changing a filter now means re-tagging everything already posted. The information arrives precisely when acting on it has become expensive.

So the common outcome is a schema chosen under maximum uncertainty and kept under maximum inertia, which is roughly the worst available combination and nobody's fault.

What can still be done afterwards

Two things, and the section on edits prices both.

A new facet multiplies the resolution by its size. Adding a filter of four choices to a board of one hundred and forty-four cells gives five hundred and seventy-six. It is the larger of the two edits by a wide margin and the more expensive to carry out, because every existing listing needs a value for the new filter.

A new value inside an existing facet multiplies it by a smaller amount. Taking a facet from four choices to five multiplies the cells by five fourths. Existing listings mostly keep their values, so the edit is cheap — and it buys correspondingly less.

Edit to a board with 144 cells	Cells after
add a facet of four choices	576
add a facet of two choices	288
take a facet from 4 values to 5	180
take a facet from 6 values to 7	past this record's bound

The last row is not a dodge. The bound this record sets is six, and figures past it are not in the table because the record does not hold them — though the arithmetic, as the page on going beyond the bounds says, carries on working.

The asymmetry worth noticing

The cheap edit is the one that preserves the existing arrangement, and the expensive edit is the one that changes what the board is about. That is not a coincidence; it is the same fact twice.

A new value inside an existing facet says: there is one more kind of the thing we were already sorting by. A new facet says: there is a dimension we were not sorting by at all. The second is a bigger claim, and it pays bigger.

The three-year version

A board opens with three filters. Two years later it carries five, because two were added when complaints made the gaps obvious.

The first three were chosen with no listings to test against. The two added later were chosen with plenty — and were chosen to address complaints, which means they address distinctions readers could already name.

Distinctions readers cannot name generate no complaints. So the filters added later are systematically the ones for gaps that were visible, and the gaps that were never visible stay open regardless of how long the board runs or how much feedback it receives.

That is not a failure of attention. It is what happens when the only signal is complaints and complaints require a vocabulary.

What this means for reading a board

It means the structure is the honest thing to look at.

Volume figures are a statement about now and go out of date in a week. Interface quality is a matter of opinion. The number of filters and the choices in each are visible on the page, do not change without somebody deciding to change them, and determine everything in this record.

A board that carries three filters of four, six and six choices resolves one hundred and forty-four kinds of listing. That sentence was true before the board had any listings, is true now, and will be true for as long as the filters stay as they are. Very little else anybody can say about a website has that property.