Every number in this product carries its proof
These are the rules our code enforces, and where to check each one on a public page of this site.
The distinction everything else follows from
A reading can say three things, not two. A value exists. A fact exists without a value: "outside the ranking", "not cited". Or nothing was collected at all, because collection is still running, has failed, or the subject is too young. Collapsing the last two makes a metric crater during a network outage and paints a collapse every morning while collection runs.
Every number travels with its provenance
A figure shown without its origin is an opinion. We label ours: *measured* when the value was observed, *ceiling* when it is a limit of the product, *computed* when it derives from other published values.
The rule covers our screens and this site alike: a number with no entry in the truth registry cannot be displayed, and the page refuses to render rather than quietly drop it. A value also travels with its UNIT: "nineteen measured" does not say nineteen of what, and that is a defect we shipped once.
For example
Our log parser throughput is published in lines per second, with the *measured* label and the bench that produced it. The ranking capture depth carries the *ceiling* label instead: it is not a performance, it is a limit, and showing it without saying so would amount to boasting.
Three states, never two
A metric can be *measured*, *measured-negative* (both facts), or *not measured*. We show a dash, not a zero: a zero asserts that we counted and found none, which is a different sentence.
This is not an aesthetic preference. A dashboard that renders "0" during a collection outage triggers meetings, fixes, and sometimes decisions; all of them resting on a value nobody observed.
For example
A keyword found nowhere in the ranking is a FACT: it shows as one, and it counts towards the averages. A keyword whose reading failed today is nothing: it shows as a dash, and it leaves the denominator.
Every share travels with its denominator
A thirty per cent citation share across ten answers and across a hundred answers are not the same statement. We print the count next to the proportion, every time.
And an average over zero items is not zero: it is nothing. Dividing by an empty denominator renders a dash on screen and "not measured" in the export, never a floor percentage that would look like a measurement.
For example
A renormalised score drops the axes we could not measure, never the ones we measured and judged poor. Otherwise removing a check would RAISE the score of a site whose defect is still open.
Data residency is scoped, so it can be checked
We do not claim the whole platform is sovereign, that would be false. Third-party providers are involved in rankings, backlinks and model-generated interpretation, and we publish the full list along with the triggering condition of every transfer.
What we guarantee is narrow and held: **no IP address leaves your infrastructure**, because your logs are parsed on our side and none is ever written down. A narrow true promise beats a broad false one, and the second collapses at the first audit.
For example
The claim "one hundred per cent sovereign" was WITHDRAWN from our truth registry, and a test forbids its return in either language. It sells better than the truth; that is exactly why a machine refuses it.
We would rather say nothing
When a value falls outside plausible bounds we do not display a cap: an invented number is indistinguishable from a measurement, and nobody can tell them apart once it is on screen.
When a verdict is inconclusive we hide the figure it refuses to trust. Announcing "too early to tell" and printing a red delta in the same breath leaves the reader with the delta.
For example
A regression detector that could not re-examine the subjects already flagged does not conclude. It does not say "no anomaly": that sentence would resolve open signals during a collection outage.
No competitor can copy this discipline without exposing themselves. Which is exactly why we use it as an argument.
Check it yourself
Each of these rules can be observed on a public page of this site. None of them asks you to take our word for it.
-
Pricing
The ceilings on the pricing page are read from the registry that enforces them, not transcribed next to it.
-
Privacy
Retention periods are read from the purge commands the scheduler runs.
-
AI crawler directory
The answer-engine crawler directory is derived from the list the engine audits, and it says "partially" where that is the case.
-
Glossary
The glossary defines every term used here, including the ones two trades use differently.
-
NessFlow vs Semrush
Our comparison pages carry the date the named offering was reviewed, and separate a modelled estimate from an observed measurement.
Access opens in waves: we email you when yours is ready.