The system you will actually use

The best record-keeping system is the one still being used in month four. By that measure, a spreadsheet with six columns beats accounting software almost every time for someone at this scale.

This is not an argument for doing less. It is an argument that captured data beats well-designed data, because a well-designed system nobody fills in contains nothing at all.

The failure mode is abandonment, not inaccuracy

New earners rarely abandon records because they are hard. They abandon them because the record takes longer than the thing it records.

A rating job is often twenty to forty minutes of work. If logging it takes four minutes of category selection, client creation and invoice matching, the admin is a meaningful fraction of the job, and on the first busy evening it gets skipped.

The design constraint that follows is strict: the record must take less than a minute or it does not survive.

The minute

One row per job, entered at delivery, in the same sitting as the job.

Date. Gross amount, before the platform's cut. Fees deducted. Net received. Format. Buyer reference.

Gross and fees are separate columns rather than one net figure because a return generally asks about income and costs separately, and because the fee is usually a deductible cost rather than money that was never yours. That distinction is worked through in how fees change your effective rate, and getting it wrong is the most common structural error in a first year of records.

Everything else is derivable. Effective hourly rate, take-home by format, monthly totals, the set-aside figure: all of it comes out of those six columns, which is why a seventh column is almost always someone solving a problem they do not have.

Where it lives

Anywhere you already have open at the moment a job ends.

A spreadsheet on your phone works. The rule is that adding a row must not require opening something you were not already going to open. UK readers should note that Making Tax Digital for Income Tax has applied since 6 April 2026 above £50,000 of income, dropping to £30,000 in 2027 and £20,000 in 2028, and it means digital records and quarterly updates through compatible software.

Costs go in a second sheet with the same discipline, and their documentation gets captured at the moment of spending rather than sorted at the moment of filing, which is the receipts argument.

What this deliberately leaves out

It leaves out anything you can reconstruct.

Invoice numbers, unless a buyer asks for one. Categorised expense codes, which are a year-end job on a list, not a same-day job on an item. Time tracking per job, which is worth doing for a fortnight to calibrate your pricing and is not worth doing forever.

It also leaves out the delivered work itself, which is a different set with a different retention pressure, and the records you will wish you kept covers what belongs in that pile.

The payoff is not the deadline

Filing is the obvious reason and it is the smaller one.

Six columns across thirty jobs tell you which format actually pays, which is a pricing decision rather than an accounting one. They also make a platform's own report checkable rather than authoritative, and the two will disagree for reasons worth understanding, as the reconciliation piece explains.

Nothing in this habit is about the craft, which lives elsewhere in the network. What an assessment must contain to be defensible is documented as method; what the free automated version of it already does is described by AI Penis; and the tools most earners pay a subscription for are catalogued at Penis Rater, which is worth checking before adding another line to your cost sheet.

Your own six columns and a platform's export are two independent records of the same year, and the reason to keep both is that on Rate Cock, as anywhere, the exports attached to a judge account are only available while the account is.