The number moves because your speed does

Your effective hourly rate keeps changing because, at a fixed price per job, it is the fee divided by however long the job took, and in the first months your time per job falls fast. Later, unbilled admin and scope drift are what push it back down.

Early on, the dominant variable in that division is not the price. It is you getting faster.

The arithmetic nobody runs

Take a job priced at a fixed fee. In month one it takes you seventy minutes end to end, including reading the brief, doing the work, delivering it and answering the follow-up message. By month four the same job takes thirty-five.

Nothing about the price changed. Your effective hourly rate doubled.

The table is an illustration rather than a measurement, and the shape of it is what matters: most of the improvement arrives in the first two months, and it arrives without you doing anything deliberate about it.

This is why a first month tells you almost nothing about what the work pays. What a realistic first month looks like covers that from the volume side; this is the same problem from the clock side.

What actually gets faster

Three things, in roughly this order.

The first is decision time. A new earner spends a surprising share of a job deciding what to say rather than saying it. That share collapses once you have a repeatable structure to hang observations on, which is most of the argument for a reusable rubric.

The second is setup and teardown. Opening the files, finding the template, formatting the delivery, writing the handover message. This is pure overhead, it is identical on every job, and it is the part that responds best to being made boring.

The third is the writing or recording itself, and it improves least. Producing a genuinely specific assessment takes about as long in year two as in month two, because the thinking is the product.

That ordering has a consequence worth noticing. Almost all of your speed gain comes from the parts of the job that are not the work, which means chasing it costs you nothing in quality. Once those parts are optimised, further speed does start costing quality, and that is the point to stop.

Measure by the month, not by the job

Per-job timing is useful exactly once, when you are setting a first price and need a starting number. After that it misleads, because the variance between two jobs is larger than the trend you are trying to see.

A thin brief takes twice as long as a detailed one. A chatty buyer adds fifteen minutes of messages that were never in your estimate. One job in ten goes sideways for reasons that have nothing to do with your speed.

The monthly figure absorbs all of that. Total money received in the month, divided by total hours spent on anything related to the work, including admin and messages and the enquiries that never converted. That single number is the honest one, and it is usually a fair bit lower than the per-job calculation suggested. The gap can be large: in Hara et al. (2018), a study of 2,676 Mechanical Turk workers, the median came out at about $2 an hour, and the authors note that how unpaid time - searching for tasks, rejected and unsubmitted work - is counted drives the result.

Run it monthly, at the same point in the month, whether or not the month felt good. Reviewing your numbers only when you are anxious guarantees you review them at the least informative moment, which is why the habit belongs on a calendar rather than in a mood.

The plateau, and what it means

Somewhere between month four and month eight, speed stops improving. The overhead is gone, the structure is automatic, and what remains is the irreducible time the work takes.

This is not a problem, but it is a hinge. Up to this point your income has been rising on its own. After it, every further gain has to be bought with a decision.

There are three decisions available, and only three.

Raise the price, which works only if you have a record and a backlog behind it - the preconditions for a first rise are both required, not either. Change the format mix towards work that pays better per hour, which requires knowing which format that is rather than assuming. Add hours, which is not leverage and runs into a ceiling faster than most people expect.

The reason the second one is usually the answer is that earners are consistently wrong about which of their formats pays best. The one that feels quick is often the one with the most unbilled messaging attached, and the one that feels laborious is often the one buyers accept a premium for. You cannot resolve that by intuition; you resolve it by tracking take-home per format for thirty jobs and reading the answer.

The thing that quietly reverses it

One force pushes your effective rate back down, and it is scope drift.

A job you have done two hundred times accumulates small courtesies: an extra paragraph, a quick follow-up answer, a second look because they seemed unsure. None of them is refusable in the moment and all of them are unpriced. An unnamed revision policy means the answer is always yes, which means the answer is always free.

If your monthly hourly rate falls while your speed has not, this is nearly always the cause, and the fix is a scope line rather than a price change.

Time-per-job is one of the few things in this trade you can measure precisely, which makes it unusually trustworthy compared to most of what an earner has to reason with. Anyone in the habit of quantifying something will recognise the discipline: an honest number depends on a stated method, and the conventions for making a measurement repeatable apply just as well to a stopwatch on your own work. It is also worth understanding why speed alone cannot save the cheapest end of the market: automated assessment already produces a generic result in seconds, and what an automated score actually does sets the floor no amount of personal efficiency will get under. Buyers, for their part, read scores rather than clocks - what a rating out of ten conveys to the person receiving it is the output they are paying for, and it costs the same to produce whether you took thirty minutes or seventy.

If you want the platform's own view of what a judge's throughput and earnings look like in practice, Rate Cock's judges page is where that sits.