what does lulu report do?
Prints what one recorded round measured about one brand: the design it was collected under, the rate with its interval, where the width of that interval came from, and every place the round refuses to give a number.
how do you run it?
In the parser’s own words, what one recorded round measured about one brand. Every flag this build accepts, on one line:
lulu report --snapshot SNAPSHOT --subject PATH --brand BRAND [--ledger LEDGER] [--tex PATH] [--pdf PATH]
lulu report --snapshot ROUND --subject PATH --brand BRAND ± on the terminallulu report --snapshot ROUND --subject PATH --brand BRAND --pdf report.pdf ± and as a document
what is checked before anything is computed?
The chain is re-derived first, through the same check lulu verify runs rather than a second copy of it, because a number computed from a round that does not re-derive is not a number.
what does it say about how the round was asked?
What it asked and what it asked it under, which is the half of this category a buyer is normally not shown. The report ends with every question, in the order the subject file states them, each with how many of its asks came back usable and a mark where it is left out of the rate.
Above the questions it says where the round asked from, which is nowhere: no location is sent on any call, and a report silent about that reads as a report that chose somewhere. Under the rate are the two lines that produce it, the mean of the per-prompt means and the clustered percentile bootstrap named with the draws and the seed it was run at, so the arithmetic can be repeated rather than believed.
why is a question that names the brand left out?
A question that names the brand cannot measure that brand. Detection matches declared literals, so a question carrying one of the brand's own forms is answered with that form whatever the model knows, and the answer scores as a mention. Which questions those are is computable from the subject file, so it is derived rather than labelled by hand. Such a prompt is left out of the rate and its interval and named on its own line: a number that quietly got smaller is its own defect.
That rule was written against this repository's first real round. One question in ten asked what the brand's own domain was, every answer to it named the brand, and every one of those answers was the model saying it had never heard of it. It was worth ten points of headline visibility on the arm collected with no search tool, which reads 0.0% over the nine questions that remain.
what does the document add?
The file the numbers came from: the snapshot id, how many records are on it, what the round's own seal says it did, the hash the chain ends on, the model as the response reported it, the surface, the dates and what the round cost. A number in a PDF with no way back to its evidence is the thing this repository exists against.
--tex PATH stops after the LaTeX and needs no TeX installed to get there. --pdf PATH writes that document beside the PDF and runs tectonic over it. Where there is no engine on the PATH the document is still written, the binary that was looked for is named, and the command exits 7, which is outside the range the refusals use.