Set the limit before you start, not when you notice

A fixed fee is a bet that the job takes about as long as you expected. Nothing in the arrangement tells you when you have lost that bet, which is why the losses accumulate quietly for months.

A time box is the missing signal. You decide, per tier, how many minutes the work gets, and you find out you are over while there is still something you can do about it.

Set the number from your own data

Time one complete job per tier, honestly, including the brief, the work, the delivery message and the admin afterwards. That measured number is your box, plus about a fifth for the jobs that arrive awkward. Do not try to replace the margin with a pessimistic guess: in Buehler, Griffin and Ross's 1994 study, fewer than half of students (48.7%) finished even by their worst-case estimate. The argument for measuring rather than guessing is in the piece on timing yourself.

An illustration of the shape, not a claim about any real rate card:

A fee with no box attached has no implied hourly, so it cannot be compared to anything, including your other tiers. That is the reasoning under the choice between hourly and per-job pricing: per-job is right for this trade, and it only works if you know the hours it hides.

Watch the box, not the clock

Start a timer, put it out of sight, and check it once at the point where you would normally be about two thirds done. That single mid-job check is where the whole technique lives. Over at two thirds means you are not going to make it, and you now have twenty minutes of options instead of a finished job that took double.

When it overruns

Three causes, three different answers, and telling them apart matters more than the fix does.

The brief was bigger than the tier. Deliver the tier you sold, to its stated scope, say so plainly in the handover, then reprice that shape of work because it will arrive again. This is not the buyer's fault, it is a gap in your scope box.

You are gold-plating. Very common, and it looks like conscientiousness. A third pass over the wording adds almost nothing a buyer notices and costs a third of the fee.

The format is wrong for you. If one tier overruns every time, the tier is mispriced or you should not be offering it. Thirty jobs of records will tell you which, and tracking take-home per format is how that becomes visible rather than felt.

Finish the job either way; abandoning work mid-box costs a review, and a review outlives every fee.

What the box protects

It protects the hourly figure, which is the only number in this trade that improves with experience, and it protects the batching that makes a working session efficient rather than an evening that ate itself. Five boxed jobs in one sitting behave like a session; five unboxed ones behave like five separate evenings with the setup paid for each time.

Automated scoring is instant and free, which is exactly why your box has a floor: anything you spend twenty minutes producing that a model produces in two seconds is time you cannot bill for at all. Where the minutes genuinely go, if precision is the thing you sell, is the setup and the measuring conventions rather than the judgement, and the method notes on that explain why careful work has an irreducible floor. The tooling around automated scores is what buyers have usually already run before they book a person, which shortens what they need from you.

Platform turnaround expectations set the outer edge of any box you choose, and Rate Cock publishes what its judges commit to.

Review the boxes twice a year. Your minutes per job fall steadily for the first few months, and a box that no longer binds has stopped being a control and become a decoration.