LOTGEN

Academy · Money

Reading Usage

Spent, kept and discarded are three different numbers, and the difference between them is the only honest measure of what a finished cut cost.

Six figures in a row: $66.05 spent at the labs, $41.87 kept across 204 winners, $24.18 discarded across 98 takes, 1.48 to 1 takes per winner, $145.32 Higgsfield-shaped and marked illustrative, and 87 per cent of spend confirmed by the provider. Under them, two lines: $51.99 is on providers that report no per-request cost, and 4 charged takes are on a model with no price.
Spent, kept and discarded are three different numbers, and the ratio between them is what a finished cut actually cost. The two lines underneath are the page refusing to round off what it cannot measure.Usage · 2026-09-08

Every take this workspace renders lands on Usage with what it actually cost. The page answers four questions, how much, on what, for whom, and how efficiently, and the four are easier to read once you know which number belongs to which.

Spent, kept and discarded are three different numbers

FigureWhat it is
SpentWhat the providers charged for every take, kept or not. This is the money
KeptThe part of it that is in a mounted take, the winners
DiscardedThe rest. Takes you paid for and did not use
Takes per winnerHow many renders it took to get each one you kept

The last one is the only honest measure of what a finished cut cost. A shot that took four attempts cost four renders, and a per-clip price that counted only the winner would be describing a project nobody ran.

It is also the number that improves fastest. Rendering one shot before queueing five, and fixing a take rather than rolling it again, both move it directly.

Part measured, part estimated, and the page says which

Some providers report what a job actually cost. One large gateway reports nothing at all.

Where the provider tells us, the figure is what they charged. Where it does not, the figure is our own estimate, and every such row is marked. So a month’s total can be part measured and part estimated, and you can see which part.

That is not a failure to reconcile. It is a property of the provider, and it is worth knowing before you build a client invoice on a total: rows that carry an estimate are the ones to check against the provider’s own dashboard.

The four breakdowns

CutAnswers
By dayWhen the month happened. Quiet days are drawn, not omitted
By providerWhich shelf the money went to
By lotWhat it was for
By clientWho it was for, the billing question

A long tail is summed and counted on a line of its own rather than dropped, so eleven providers where four are under a dollar still add up to the total on the face of the panel.

Getting an invoice out

Filter to a client, then export. The CSV is built from the rows on screen rather than from a second request, so the filters you set are the filters you get, a CSV that quietly ignored them would be a per-client invoice containing another client’s work.

The client and the month are in the filename, so a folder of these is self-describing a year later.

By person, and the two rows that confuse people

Usage also splits spend by member. Two things about that split are deliberate and read wrongly if you do not know them.

An agent is not a member

A workspace key’s spend belongs to whoever minted the key. An agent render is that person’s money wearing a tag, and the tag is the door it came through, so each person’s row carries a separate figure for the part of their spend that arrived from somewhere other than the web app.

Giving the agent a row of its own would take spend off the person who is accountable for it, which is precisely backwards.

The unattributed row is a row, not a remainder

Takes made before authorship was recorded, and anything a bootstrap token enqueued, sit in their own entry with no person on it. A member whose account was removed folds into it too, their takes arrive with an id and no name, and printing a bare identifier on a board of names would be worse than saying the take has no author, which is the honest description.

Because it is a row, the people sum to the total. Do not compute it by subtraction and do not describe it as missing money.

The comparison figure, and what it is not

Beside your spend, the page shows what the same work would have shaped up to on credit-based pricing. It is labelled illustrative every time it appears, and that label is not the caller’s to remove.

It is a multiple applied to your own spend, not a quote, not a measurement, and not something a competitor told us. Presenting it as a price we checked would be exactly the dishonesty the rest of this machinery exists to refuse, on the one figure where we have an interest in the answer being large.

It is never shown to a client. Not on a review link, not in an export pack, cost is the studio’s business, and a comparison to somebody else’s pricing is a message to ourselves.

The month, again

Usage’s month is the same UTC calendar month the cap enforces, from the same definition. One boundary, one answer, on both sides of the wire.

Want to try this on your own keys? Request an invitation, or get a workspace today.

Something wrong here, or a lesson you want that is not on the list? Write to support@lotgen.ai.