No client results yet.

Every dollar figure on this site is one of three things, and says which: our price, arithmetic from a public schedule (Amazon's own fee cards, dated), or an estimate modeled from public data. None is a client's result. When one is, it will be on the wall above and nowhere else first.

An estimate is a range, never a point. We simulate the uncertain inputs thousands of times and report where the outcomes fall: P10 is the line one simulated year in ten comes in below, P90 the line one in ten comes in above. The band between them holds eight years in ten, the bad end included. Ranges are rounded down.

The largest uncertainty is volume. A listing's public sales rank is turned into units a month by a curve fitted to published estimators, and that curve can be wrong by half or by double. Every estimate is simulated that wide.

How your Profit Record counts a dollar

These are the rules your Record is kept by. They are agreed before any number is on it, so neither of us can bend them once one is.

Before. Every move is written down with the dollars we expect it to earn over its first thirty days, before it goes live.

After. It is measured on your own later exports, and every dollar is graded by how we know it:

• Direct. The platform's own record shows the money moved, as when Amazon pays a reimbursement on a claim we filed. Counted in full.
• Isolated. Measured on the exact line the move named, against that line's own level before the change: the change, times the volume you actually sold.
• Attributable. Where demand moved too, measured against a stated counterfactual, from the same model that made the promise, at the least favourable end of its own range.
• Unmeasurable. Where we cannot separate our move from everything else, nothing is counted, and the Record says so.

Four rules apply to every dollar.

• Never above the promise. An isolated or attributable result counts at most what we said it would earn. Anything more is shown and named, never added. A direct dollar counts what the platform actually paid.
• Nothing under $25. A smaller result is noise: the move closes and counts nothing.
• Once. The same movement on the same SKU in the same period is credited to one move only, the oldest.
• It has to last. Every month, each move is measured again on that month's exports alone, against the level before it went live. A leak that stopped holding earns nothing that month.

Misses stay. A move that earned less than we expected is recorded at what it earned.

Every month stands on its own. About a week after a month ends, once its exports have landed, everything every move earned in it is added up, once. That one number is the month's: nothing found but not yet measured counts, and nothing carries from one month to the next.

Disputes are settled by inspection. Every graded dollar carries its working: the grade, the window of your exports it was measured on, the evidence, and a note in plain words. If you question one, we open that record together. If it does not settle the question, the dollar comes off your Record.

The same rules are in the terms, word for word.

A real store's orders. A UK online retailer published every order it took from December 2009 to December 2011 (UCI Machine Learning Repository, CC BY 4.0). We picked it because the data is complete and open to anyone, not because of what it showed. The engine made each call from the orders before a cut-off, the cut-off for repeat customers set by the engine's own rule, and every call is printed beside what happened, misses included. It cannot show profit: the data has no costs, advertising or stock. The study.

One real Amazon listing. The Proof page also models one real Amazon listing from its public page, captured on September 3, 2026: a car-care brand's best-selling paint scratch remover, sitting just past one of Amazon's weight steps. The estimate is $6,400 to $19,200 a year paid on that one step. It is labelled on every screen as modeled from public data, not a client and not a result. The brand is not named.

We picked that listing because it had a finding. Most do not.

When the numbers say there is nothing to fix, we say so on the call.

• No client results. Zero paying clients, zero published results, no testimonial and no logo. None will appear until it is real.
• The Seal has no outside witness yet. Every move is fingerprinted onto your Record before its email goes out, and the chain can be checked by anyone with the export. It has no external timestamp yet, so whoever owns the database could rewrite it consistently; the Seal would make that disagree with what your inbox already holds. That is evidence, not prevention.
• Public data cannot see your inside. Ads, returns, your real landed cost, the packed weight and how long each unit has sat are invisible from a listing page. That is what the call is for.

A client's scoreboard reads, in this order: what the Record has proven since day one, what it measured this month, the bill, and the difference. Under that, every move: what it was, the dollars expected before it went live, the dollars measured after, and how we know. The first results published here will be those same numbers, with the client's consent, rounded down, without their name.