Method
Where every number comes from
A calculator is only worth the weakest number inside it. This page sets out how each figure got there, how confident we are in it, and the cases where the honest answer was to refuse to produce a number at all.
Four grades of evidence
Every fee on the site carries one of four badges. They are not decoration — they are the whole basis on which you should decide how much weight to give a result.
| Badge | What it means | How it was obtained |
|---|---|---|
| Settlement | The figure was read from a real payout statement in a live seller account. | Downloaded transaction reports, reconciled line by line against the order they belong to. |
| Official | The platform's own published fee schedule. | Read from the platform's fee page, which is linked on the line so you can check it yourself. |
| Published | A public rate that has never been cross-checked against an actual payout. | Treat as indicative. Real settlements routinely differ from rate cards. |
| Estimate | A rule inferred from limited observation. | There is exactly one of these in the database and it names the single observation it rests on. |
Every line carries a date
Fee schedules move, sometimes with a fortnight's notice and sometimes with none. Every rate on the site shows the date it was last checked, and that date is the first thing you should look at. A number checked last week is worth acting on. A number checked eight months ago is a starting point for your own verification, not a conclusion.
This is also the reason there is a verify link on almost every line. It goes to the platform's own page. If our figure and theirs disagree, theirs is right and ours is stale — tell us and it gets fixed.
What we refuse to estimate
The most useful thing a tool like this can do is decline to answer when there is no answer.
Canva Creators is the clearest case. Canva pays contributors from a Royalty Pool, not a rate. It sets the pool's size each month from its own subscription revenue, divides it by every Pro asset export across the entire platform, and pays proportionally with content weightings it does not publish. Your earnings therefore depend on how many other creators were used that month — a number you cannot see, cannot predict and cannot influence. There is no percentage to put in a calculator. Every "Canva earnings per template" figure in circulation is one person's account in one month, presented as a rule. The tab exists, and it explains this instead of inventing a rate.
The same discipline applies inside the priced channels. Amazon's long-term storage fees are not modelled per unit, because they depend on cubic feet and time of year. Against the reference settlement this leaves the model about $1.67 optimistic across a two-week period — it errs in the platform's favour, never in yours, and it says so on the page.
The four lines nobody else includes
Marketplace calculators are not wrong about their own commission. They are wrong because of what sits outside it.
- Advertising. Spread across the units it actually sold. On most marketplace channels this is the largest single line, and it appears in no official calculator.
- Returns. Not just the lost sale — the fees that are not given back. Amazon keeps up to $5 of the referral fee and never returns the fulfilment fee. TikTok Shop returns the commission in full. Gumroad returns nothing at all. The difference between those three policies is worth more than the difference between their headline rates.
- Fixed monthly costs. The subscription is owed whether you sell forty units or none, so it costs far more per unit in month one than in month twelve.
- Surcharges that appear only on settlements. The digital services fee on Amazon is the example that started this project: it is on the statement, it is in the Revenue Calculator, and it is in no public fee schedule.
How the idea score is built
The score is arithmetic on numbers already visible on the page, not a judgement. Six components, one hundred points:
- Margin after every cost — 30 points. Zero margin scores nothing; forty percent or better scores full marks.
- Platform's grip — 15 points. The share of the sale price the platform holds. Full marks at zero, nothing at thirty-five percent or more.
- Fixed-cost drag — 15 points. What proportion of your monthly volume goes purely to covering subscriptions and advertising before you earn anything.
- Payback time — 20 points. The month your start-up money comes back. Month one scores full marks, month nineteen or later scores nothing.
- Discount fragility — 10 points. How much of the profit a ten percent price cut destroys.
- Evidence quality — 10 points. The average confidence weight of the fee lines used, so a channel priced from a real settlement scores above one priced from a rate card.
Every component displays its own points, its own value and the reason it scored that way. If you disagree with one, you can check its arithmetic against the ledger above it.
What the projection is, and is not
The twenty-four month curve is the one place on the site where the inputs are not sourced, and it is labelled as such wherever it appears. The fees in it are sourced. The ramp is your assumption: it climbs in a straight line to your stated volume over the number of months you choose, charges fixed costs in full from month one, and scales advertising with the ramp. It does not know your market, your competition or your luck. Its purpose is to make the shape of the cash requirement visible — how deep the hole gets before it fills — not to predict your business.
What is outside the model entirely
- Tax. Everything is pre-tax. Income tax, self-employment tax, corporation tax, VAT and sales tax are all outside the model. The name describes what platforms take, not what governments take.
- Your hours. A channel paying $2 a unit for four hours a day is not a good channel, and no number here will tell you that.
- The cost of building an audience, which is the entire real cost of the direct-sales channel.
- Restocked returns. Returned stock is assumed unsellable. If you resell returns, your real figure beats the one shown.
- Chargebacks, which matter most on exactly the channel where they are hardest to predict.
Corrections
Errors get published, not quietly patched.
- One figure is knowingly unexplained: a refund credited $1.74 where the published rule predicts $1.68. Six cents are unaccounted for. It is left visible rather than smoothed over, because a model that always reconciles perfectly is usually a model that has been fitted to its own assumptions.
- Amazon monthly storage is not modelled per unit. Checked against the reference settlement this leaves the model roughly $1.67 optimistic over a two-week period.
- The Amazon digital services fee is the only estimate-grade line in the database. It rests on a single observation of $0.17 against a $2.10 referral fee.
Found something wrong?
If a rate here disagrees with the platform's own page, the platform is right. Send the line and the link to hello@thetaxcut.com and it will be corrected with a new checked date. Corrections from readers are the cheapest quality control this project has.