how are the names in an answer counted?
Every other count in the package is about names somebody declared first. This one is the opposite direction: the questions carry no company in them, and what comes back is whatever the model already believed. Reading that list is how a customer finds out who it is actually competing with inside an answer, which is not the list it would have written down.
is a name a rival?
Nothing here is a rival. This counts names and judges none of them. A rival is a name a person recognises as one, and a table that decided for them would be the library asserting a market it cannot see.
how is a heading told from a company?
A model writing a structured answer capitalises its section titles exactly the way it capitalises a company, and the first pass at this reported Reliability and Key Considerations as names. A stop word list would have removed those two and waited for the next three, so the rule is positional instead: a candidate counts only if it appears at least once inside a sentence, somewhere that is not the first word. A heading never does. A company mentioned in prose always will.
what is the unit?
The question, not the answer. A name in all six draws of one question and a name in one draw of six questions are not the same finding, and counting either as "six" makes them look identical. The first is that question having a stable answer, and the second is the model reaching for that name whatever it is asked. So the count is kept per question and the number of questions is reported beside it.
is anything inferred?
Nothing is inferred, extracted or corrected. Every name printed appears character for character in a recorded answer, which is the same rule the evidence gate holds a quote to. There is no model call and no network anywhere near it: the same records produce the same table forever.
what does it export?
Every public name in lulumelon.mirror.names that its source says something about, with the first paragraph of what it says.
which commands reach it?
lulu rivals ± lulu draft ± lulu screened ± lulu publish
who does the model name when nobody suggests anybody?
every part of the method: intervals, variance, stability, compare, sources, ablation, lift, screen, names. back to the method