Receipts
Pick any number we show you and ask what stands behind it.
Most reporting tools answer that question with a support ticket. Signet answers it on the screen the number is on, because a figure whose provenance takes a week to retrieve is a figure nobody can act on in a meeting. This page is the anatomy of that answer, written for the person whose job is to interrogate it rather than to be reassured by it.
A receipt has five parts, and a number without all five is not published.
These are not five things we can produce on request. They are the shape of the record itself: the type that carries a measured value has a required evidence field, so a value with nothing behind it is not a weaker number in our system. It is not a value our system can express.
That distinction is the one worth checking with any vendor. Ask whether provenance is attached to the number or retrievable about it. The second answer means somebody has to go and reconstruct it, and reconstruction is where the disagreements start.
- BasisWhich kind of evidence produced the number, which attribution model it was read under, what time basis it counted on, and which currency it is denominated in. Two numbers computed on different bases are two different claims, so the basis travels with the number rather than living in a settings page.
- WindowThe exact period counted, resolved in your reporting time zone and your reporting calendar. A window opened at an instant has a partial boundary day, so a number is scoped to the days its window covers completely rather than to the days it touches.
- CoverageHow much of the thing being measured the number actually reaches: complete, partial, or not applicable, with the reason named when it is partial. A partial answer that does not say it is partial is the most expensive kind of wrong.
- MaturityWhether enough has resolved for the number to mean anything yet. A cohort three weeks old and a cohort two years old produce the same shaped figure and deserve very different confidence, so the count of resolved and pending observations travels with it.
- ProvenanceWhich version of the definition computed it, which read path produced it, which fields it drew on, and which subjects were excluded and why. This is the part that lets you reproduce a number you were shown last quarter.
Named from the fields the product's own metric contract requires. A field renamed there fails this page's typecheck rather than quietly leaving this list out of date.
A number and the thing wrong with it arrive together, or the number waits.
Every surface that carries a figure can carry three classes of fact beside it. One of the three can never be tucked behind a click, and which one it is decides whether a product is honest or merely honest-looking.
A qualification you have to go and find is a qualification most readers will not find. Putting the refusal, the disagreement or the too-young window behind a hover would make honesty opt-in, and almost nobody opts in on a Monday morning.
- DescriptiveMay be collapsedWhat the answer is scoped to: which lens, which filters, what it is being compared against, which benchmark it is measured beside. A description of an answer is not a qualification of it, and a stack of them beside every figure is noise.
- CaveatAlways visibleSomething wrong with the number: a refusal, a grade that says do not act on this, a disagreement between two readings, a window too young to judge. This is the class that can never be collapsed.
- SubjectAlways visibleWhat the numbers actually are: the window they were counted over, the origin a rate is measured from. Hiding this would not show you less detail, it would change the claim you are reading.
The badge on each class is read from the chrome's own vocabulary, so the rule about which may hide is the product's rather than this page's. The descriptions are written for this page, because the versions in the code were written for the people maintaining it.
The chain has five links, and a pound crosses all five or it counts as untraced.
Chain coverage is the share of the money credited to an ad whose whole path is evidenced rather than inferred. It is a ratio of two sums of money, never an average of per deal percentages, and there is no partial credit: a sixty percent traced chain is not something the evidence supports, so it is not something we report.
- The click was seenThe touch that earned the credit was observed by the pixel on your own site, not rebuilt afterwards from historical records. A reconstruction is a reasoned inference, and counting inferences as evidence would make a coverage figure a measure of our own confidence.
- The ad is knownThat touch names an ad we have actually synced, so the money can be attached to a real creative rather than to an identifier we have never seen.
- A person was reachedThe conversion's identity bridges to a contact in your CRM. Without this link there is a visit and a payment and nothing joining them.
- A call is attachedA call is reachable from the deal or from the bridged contact, which is what makes the conversation part of the record rather than a separate system.
- The cash is boundThe payment ledger has money bound to this conversion. Until this link closes there is a won deal, which is a promise, rather than collected revenue, which is a fact.
Carrying five separate verdicts rather than one traced flag is what makes the number diagnostic. A funnel that narrowed because few people bought and a funnel that narrowed because nothing was recorded produce the same shape in the totals, and only the link that failed tells them apart.
When a percentage is measured against a different total, the label says so.
Chain coverage is deliberately measured over every pound credited to an ad, which is a wider set than the attributed revenue figure printed above it on the same card. That is the honest choice and it is also the inconvenient one, because a reader who is not told will reasonably read the percentage against the number beside it.
So the label carries the words of all credited revenue every time it renders. Narrowing the denominator to the figure alongside would have produced a friendlier statistic that was pinned near one hundred percent by construction, which is a measurement that cannot fail and therefore cannot inform.
31.7% chain coverage of all credited revenue
The share of every penny credited to this ad whose chain is traced end to end: an observed click on a synced ad, reaching a contact, a linked call and bound cash. Its denominator is all credited revenue, which is wider than the attributed figure shown above.
Both strings are rendered from the module that ships them, so the sentence here is the sentence in the product.
A number we cannot produce is a sentence, never a zero and never a blank.
An unexplained gap where a percentage belongs reads as zero to everyone who sees it, and zero is a claim. There are two reasons a coverage figure can be absent, each one a different fact with a different fix, and the product says which rather than rendering the same shrug for both.
An ad nobody credited has no chain to measure, which is not a tracking failure and must not be drawn as one. A ledger we have not recorded yet is a real gap on our side, and saying so is how you know to ask us about it.
- Nothing creditedNo revenue is credited to this ad in this window, so there is no chain to measure. This is not a tracking failure.
- Binding not recordedThe payment-binding ledger has not been recorded for this organization yet, so whether revenue reached cash cannot be shown.
A closed list. A new way for this number to be missing cannot be added without writing the sentence a reader will see.
Below 30 observations, the receipt says so before the number does.
Maturity is a part of the receipt rather than a footnote on it. A verdict reading Needs more data is not a softer version of a real verdict, it is a different statement: we have not seen enough to have an opinion, and here is how much more we would need.
One floor applies on every surface. A product where confidence depends on which page you opened has no floor at all, it has a preference.
- Needs more data
- Maturing
- Solid
CAC $212 · 48 closed · 90d
CAC · 3 closed · 90d
Rendered by the product's own components. The evidence line prints the sample beside the rate at the same size, which is the fastest way to audit any figure on any screen.
Anything the model wrote carries where it came from, recorded when it was made.
Creative that Signet generates is registered at the moment of creation rather than labelled afterwards. The record holds which model produced it, the prompt that was submitted, the brief and brief version it was drawn from, the rule version in force, and the person whose account asked for it. That record is append only, so it cannot be quietly tidied later.
Born tracked rather than tracked on adoption is the load bearing part. A provenance record written at publish time can only describe what was published, so the variation somebody copied out and shipped by hand would have no history at all. Registering at creation and matching on content means the copied one is recognised when it comes back.
Audit us
Three questions that separate a real receipt from a confident font.
- Show me a rate and its denominator, at the same size, on the same screen.A percentage with its sample hidden is a rhetorical device rather than a measurement. If the evidence line looks too thin to support the claim beside it, believe the evidence line.
- Show me something the tool refuses to answer.Every measurement system has a boundary. A product that cannot show you one is either hiding the boundary or has not found it yet, and both are worse than being told.
- Ask what changed when the definition changed.Numbers move because the world moved or because the definition moved, and telling those apart after the fact is impossible unless the version was recorded at the time. Ours is part of the receipt.
The method behind the numbers, including the parts that make us look worse, is on how we count. How your data is kept apart from everyone else's is on security.
Ask for the receipt before you trust the number.
It is one chain rather than a set of tools you wire together yourself. You get all of it, and the tiers differ by scale rather than by which capabilities are switched on.