Most specifications are a list of things to build. This one has to survive being paid on.
Every engagement here begins with a single document, co-authored before anything exists. It says what will be built, what number counts as success, and what that number would have done without us. Nothing is built from it until it passes a gate.
What follows is that standard, shown field by field through an illustrative example — and annotated twice, because the same document is read by two people for two different reasons. A business owner reads it to check the deal is honest. A builder reads it to know what they are being held to.
Both are reading the same lines.
Before anyone writes code, something has to be true. The question is what.
In most arrangements the answer is: enough agreement to start. A scope, a timeline, a price. Whether the thing works is settled later, by argument, using evidence that didn't exist when the argument's terms were set.
Beyond a capped base that covers our cost, we're paid only out of the growth we create. That makes the question sharper, because a specification we cannot measure against is one we cannot be paid on beyond the base. Vagueness stops being a communication problem and becomes our problem.
So the standard below isn't a preference about documentation. It's the minimum set of things that must be settled in advance for the arrangement to be checkable by either side afterwards.
The document has a name in each of its two states, and the distinction does real work.
Filled in and co-signed, before it has been tested, it is a Golden Copy — complete, agreed, and not yet cleared. It carries everything a specification needs and has not yet been checked against the standard.
Once it clears the gate, the same document is a PRD — the thing that gets built from.
Clearing changes the status, not the text. Not one word is edited by passing. It is the difference between a bill and an act: the same sentences, and a different authority. Revision happens only on failure — a document that doesn't clear goes back to co-authoring and returns.
Nothing is built from a Golden Copy. Things are built from a PRD.
The first field is the narrowest one. Not a strategy, not a transformation — a discrete thing that will exist when it didn't before.
The exclusion line matters more than the inclusion line. A scope that says only what is in it will be read generously by whoever benefits from a generous reading.
One number. It has to be growth that did not exist before, in a form someone outside the room could verify.
The substitution clause is the whole test in miniature. Without it, every booking that merely changed channel counts as value created — a number that would rise on day one and mean nothing. Its presence is the difference between measuring growth and measuring adoption.
A metric nobody could fail is not a metric. If you cannot describe the reading that would count as disappointing, the field isn't finished.
This is the hard one, and the reason the document takes as long as it does.
The baseline is the counterfactual: what the metric would have done anyway, without the thing being built. Value created is the gap above it — never the whole line.
Two things make this survivable. It is fixed before the outcome is known, when neither side can tell which way the argument favours them. And it names its own dispute procedure — a source of record both parties can query, rather than a conversation.
A measurement that depends on someone remembering to measure is not a measurement.
The build carries its own scoreboard. This is what makes the invoice and the evidence the same object: there is no separate act of assembling a case, because the case is a by-product of the product running.
Two of the fields above exist for the case where the number does not move. The date, and the level of use that counts as adoption, are both fixed before anyone builds anything — which is the only point at which they can be set by agreement rather than by whoever the answer happens to favour.
Ownership is stated in the specification, not left to the contract to imply.
This one is deliberate rather than generous. An arrangement in which we could switch off what we built would corrupt every other clause above it, because it would give us a way to be paid, beyond the capped base, that has nothing to do with growth.
And here is the field that makes the document unusual. Read it against the success metric above.
The payment clause does not define what it is paid on. It references the section that already did. One definition, cited twice.
The one fixed part is the capped base. The engine is built before it can create anything, and the base pays for that work — the builders’ time and the tools. It is written into this field before anything is built, it never grows, and it covers cost, not profit. It is the only thing owed regardless of result, which is why it sits in plain sight in the terms rather than in a schedule at the back.
And the clock starts with the measurement rather than the signature. Time spent building is not time spent earning a share, so the term runs from the day the measure is verified — with an outer date past which it starts anyway, so that neither side gains from delay.
Now the inspection this page exists for, and it takes about thirty seconds.
Find the line we are paid on. Find the line success is measured by. They are the same line.
That is the whole of it. Not a stated intention to be fair — a document in which the sentence describing our share cannot be satisfied except by the sentence describing your growth, because the second is what the first points at.
It has a consequence worth stating plainly: we cannot specify work we cannot measure. Not as a policy we follow, but because the acceptance criterion and the payment metric are one field. A build with no honest measure attached is a build with no way to be paid for it beyond the base.
This is what makes the arrangement positive-sum by architecture rather than by intention. Intentions decay under pressure. A document that has to be satisfied in one place to be satisfied in the other does not.
A completed Golden Copy is checked against six conditions before anything is built — one for each field above. The build must be the leanest change that could move the metric. The metric must be well-formed. The baseline must be numerically bound to the agreement. The instrumentation must be specified. The asset must be owned by you. The terms must point at the metric rather than defining success separately. Each field is checked for precisely the thing it exists to carry.
What a machine checks is the form of those six conditions: that a metric resolves to a single computable figure, that a baseline is a fixed number bound to the agreement, that instrumentation names the events it will record, that the scope carries no item without a stated link to the metric, that ownership transfers, that the terms cite the metric field instead of restating it. Form is checkable, and all six have a form.
What a machine cannot check runs across all six rather than being three of them. Whether the baseline that is numerically bound is also honest. Whether the metric that is well-formed also measures real growth rather than flattering movement. Whether a scope that is lean is lean around the right thing. These are judgments about whether a number tells the truth rather than whether it is well-formed, and no machine settles them — which is why a person signs.
Failing is ordinary. A Golden Copy that doesn't clear returns to co-authoring and comes back — no work has been wasted, because no work had started. That is the entire purpose of putting the gate before the build rather than after it.
What the gate does not ask is whether the growth will appear. Nobody can settle that in advance, and a standard that pretended to would be the first dishonest thing in the document.
Not flawless. Finished in form.
A perfect specification carries what it must and stops. Six fields, each settled. It is not longer than that, not more elaborate, not more beautifully written. A document twice the length is not twice as good; it is usually a document whose author was uncertain and wrote more to cover it.
The standard is deliberately hostile to thoroughness for its own sake. One of the six conditions is that the build be the leanest change that could move the metric — so a specification padded with ambition fails the gate on its own terms.
The test is not "is this impressive." It is could someone who has never met either of us tell, from this document alone, whether the thing worked? When the answer is yes, the document is finished. Everything after that is decoration, and decoration in a specification is a liability.
A standard held privately is a claim. A standard published is a thing you can be held to.
We are not concerned about this being copied. Any firm can adopt the form — write the metric, fix the baseline, weld the payment clause to the measure. What almost none will do is the thing that makes the form worth having: agree that, beyond a capped base that covers their cost, they are paid only out of the growth above that baseline. On a fee, welding your invoice to a measured delta buys you nothing and costs you everything. The document only pays off for someone who has already given up being paid beyond their cost when the work doesn't.
So the barrier was never the document. It is what the document commits its author to — and that cannot be copied, only accepted.
Which is also why publishing it is useful rather than risky. If you are considering working with us on either side of this, the standard above is the thing you would be inside. Read it before deciding, not after.