How to count entries without misleading anybody
A number is the most quotable thing a register produces and the easiest to get wrong. What it counts depends entirely on an inclusion rule, and a rule nobody published is not a rule. As of 2026-09-12.
| Failure | What prevents it |
|---|---|
| Counting inferences as statements | Separate the two in the rule |
| A total that no longer matches the table | Derive the number from the rows |
| Two numbers for one thing on one page | One noun, one value, checked |
| A rule invented after the count | Fix the rule before reading the pages |
Inclusion rule. Ways a published count in a register goes wrong. Order. From the most consequential failure to the least.
1A count is a claim about a rule, not about the world
Saying that several entries describe a mechanism is saying that several entries satisfy a definition somebody wrote. Change the definition and the number changes without anything in the world moving.
Which is why the rule belongs beside the table. A reader who disagrees with the rule can discount the number, and a reader who cannot see the rule has to take it on trust.
2Totals drift when tables grow
An entry added to a table does not update the sentences around it. A description saying six, an introduction saying five and a table with seven rows is a normal outcome of ordinary editing.
The fix is mechanical: derive counts from the rows where possible, and check the remaining prose against the table whenever the table changes. Nothing about this is interesting, and it is where most errors live.
3Numbers in a summary are the ones that travel
A title, a description and an opening answer are what other people quote. An error there propagates; an error deep in a page mostly does not.
So those three deserve more checking than the body, which is the opposite of how attention usually gets allocated. The shortest text carries the most weight.
4Subset claims are safer than totals
Saying that two entries do something is checkable against the table and does not depend on how many entries exist. Saying two of the sixteen also asserts a total that has to be maintained.
Where the total genuinely matters, it should come from one place and be repeated nowhere. Where it does not, the subset form is both shorter and less likely to become wrong later.
Background on mechanism and practice. Nothing here is attributed to a product, and nothing here is a reading of any instrument. The sourced material is on the generator table. Related: Asking the right party, Three kinds of marking.