what does lulu publish do?
Writes a recorded round out as pages the web can serve, one per question, with every name it found and the interval the sample supports. A round whose questions name no company is a measurement of a market rather than of a brand, and that is worth reading by more people than the one who paid for it.
how do you run it?
In the parser’s own words, write a recorded round out as pages the web can serve. Every flag this build accepts, on one line:
lulu publish --snapshot SNAPSHOT --questions PATH [--out DIR] [--ledger LEDGER] [--drop NAMES] [--least N]
lulu publish --snapshot ROUND --questions PATH --out data/published --least 2 ± names seen in at least two draws
what does it refuse?
A round whose questions carry a tracked name, because that round is about a brand. Nothing it writes is served until somebody commits it: a ledger is the customer's own record and is not in this repository at all, so there is no path from a collected round to a public url that nobody chose.
The questions come from a file rather than from the ledger, because a record carries a prompt's id and not its sentence, and a page that invented the question it is reporting on would be inventing the whole page.
what does --drop leave out?
--drop is the one place a person's judgement enters the numbers. The name counter is positional and it cannot tell a company from a genre word a model writes in title case, so API, LLMs and AI Agents come through it. What is dropped is written into the file and printed on the page beside the table, because a table that quietly lost rows is a table nobody can check against the round it claims to come from.