what does lulu usage do?
Says what the rounds already on disk cost, from what the provider said about them, and checks the chain before a single figure is printed.
how do you run it?
In the parser’s own words, what the recorded rounds cost, from the provider's own figures. Every flag this build accepts, on one line:
lulu usage [--ledger LEDGER] [--snapshot SNAPSHOT]
lulu usage ± every round in ./ledgerlulu usage --snapshot ROUND ± one round
which three bases does it keep apart?
An amount the provider stated, a cost computed from tokens it reported, and a floor for calls it said nothing about. They are never merged. Failed calls are counted and priced at nothing, because no response says whether a rejected call is billed.
A count nobody reported is never printed as a zero. Records written before usage was recorded at all are kept apart from the calls a provider said nothing about, because "this build did not collect it" and "the provider did not report it" are different facts about different parties.
why does it take no model?
Each record names the model that answered it, and that is the only model that was billed, so the rate is read per record rather than chosen once by whoever runs the command. A flag here would let a round collected on one model be priced at another's rate without anything on screen disagreeing. A model with no published price is counted as unpriced, reported out loud, and kept out of every figure.
why is the chain checked first?
An invoice computed from a file that does not verify is not a cheaper invoice, it is an unknown one, and printing it under a heading that says COST would be the exact move this library exists to argue against.