Rehearse on something nobody owns

Rehearse on synthetic material you constructed yourself, or on a written brief with no material at all - never on a real buyer's submission, not once and not anonymised. A submission arrives under a specific permission that does not extend to your rehearsals or your public examples, and you still need something to time yourself on before you can price.

That leaves four sources, of which two are worth the effort.

What is off the table

Anything a person sent to you or to anyone else for assessment. The permission attached to a submission covers the job it was sent for, and nothing beyond it - not practice, not a template, not a portfolio piece with the identifying parts removed.

Anything scraped, reposted, or lifted from a forum. The absence of a name on something is not consent, and most platforms treat retention or reuse of submitted material as a suspension-grade breach rather than a warning-grade one.

A colleague's delivered work, used as a template. This one feels harmless and is the most common quiet breach in the trade, because delivered assessments are usually the buyer's property under the terms you both signed.

The two sources that work

Synthetic material. Generated or constructed material owned by nobody is the cleanest option available, and it has a property the alternatives lack: you can produce as many variants as you need, with known characteristics, which makes it far better for calibrating a rubric than a single real example would be.

You can also deliberately construct hard cases - the ambiguous one, the one at the edge of your stated scope, the one where two of your axes disagree. Those are exactly the cases that eat forty minutes on a live job, and rehearsing them is worth more than rehearsing an easy one.

A written brief with no material at all. For timing purposes most of what you are measuring is not looking, it is reading the brief, structuring the response, writing, and delivering. A fabricated brief with a description in place of an artefact gives you a usable number for the majority of the work at zero sourcing cost.

This is the fastest route to the number the timing run exists to produce, and it is usually accurate to within a few minutes of the real thing.

The two that mostly are not

Public domain or licensed stock. Legally clean, and for this category of work the available material is thin and unrepresentative, which limits what a timing run against it tells you. Check the licence covers commercial use if you intend the output to be public, because most free licences do not.

A consenting volunteer. Workable with explicit written permission covering practice and, separately, publication if you plan to publish. The friction is real and the permission has to be specific: "yes fine" in a message is not a licence to put something on a public page indefinitely.

Worth it only if you intend the output to be your one public example, in which case the extra rigour is proportionate.

The material determines what you can publish

Practice material and portfolio material have different requirements, and conflating them is how people end up with an example they have to take down.

Practice is private, so the bar is that you had a right to use it. A public worked example is a permanent publication, so the bar is that you have a documented right to publish it and that the right does not expire. Synthetic material clears both bars without a conversation, which is most of why it is the default - building the public worked example makes the case for having one at all.

Keep a note of where each piece came from and under what permission. It takes one line per item and it is the only thing that answers a challenge two years later.

Calibrating on it

Practice material is also how a rubric gets tested rather than merely written. Run the same piece through your structure twice, a week apart, and see whether you land in the same place; a rubric that produces different answers from the same input is a rubric with a vague axis in it, and finding that out privately is considerably cheaper than finding it out in a review. A rubric you can reuse is the artefact this exercise is really building.

The consistency question has a literature behind it. In one clinical-assessment study, two raters scoring 40 recorded exams against a shared rubric, with repeated discussion of its criteria, saw their agreement (weighted kappa) rise from 0.49 on the first ten to 0.82 on the last ten (Watari and colleagues, 2022). That is the entire argument for rehearsing against a fixed structure rather than practising judgement in the abstract.

The measurement conventions themselves are documented elsewhere - the method side of this is the reference if numerical precision is part of what you are selling, and how the data behaves across a population is why a single practice piece tells you less than you think.

Two other things worth knowing while you are rehearsing. Buyers arrive having already used the free automated options, and the tools that provide them set the baseline your practice output has to beat. And what a buyer actually values in a delivered review - described from their side - is specificity rather than length, which is a thing to practise deliberately rather than discover.

One last check before you rehearse: whether the platform you are joining permits you to retain or republish anything at all, since some forbid it outright and some are silent. Rate Cock's judges page states its position, and reading it first is cheaper than building a portfolio piece you are not allowed to keep.