Methodology
Known, supported, unknown
Two figures can sit side by side in the same report and deserve completely different levels of trust. Autopsify marks which is which, on every number, and refuses to manufacture the ones the evidence cannot reach.
Operator-led · Evidence-led · Human-reviewed
The classifications
Three states, applied to every figure
Known
Directly supported by the available source data. The figure is present as a field in the merchant's own records — an order total, a recorded unit cost, a spend figure from a connected advertising account. Reading it requires no transformation and no assumption.
Supported
Derived, reconciled or inferred from documented source data by a stated method. The arithmetic is deterministic and the scope is declared, but the figure is not a source field: something was computed, allocated across a period, or reconciled between two records. Supported figures are trustworthy and auditable — they simply carry a step a reader is entitled to inspect.
Unknown
The evidence is insufficient to establish the figure reliably. Not “small”, not “approximately nothing”, not “we will assume a typical value” — unestablished. The Blueprint states the figure is unavailable, names the input that is missing, and withholds every figure that depended on it.
The rule
Unknown never equals zero
This is the single most consequential rule in the method, and it is the one most commonly broken elsewhere. An absent measurement and a measured zero look identical in a spreadsheet cell. They mean opposite things.
Consider a store with no advertising figures in the period. One possibility is that the business genuinely spent nothing — a real, measured zero, and a meaningful finding. Another is that the advertising account was never connected, or the connection lapsed, or the spend sits in a channel the data does not reach. In the second case the true figure might be substantial.
If both are recorded as zero, contribution after advertising is computed identically in both cases — and in one of them it is simply wrong, by the entire amount of the unmeasured spend. The merchant reads a healthy number and has no way to know which situation they are in. Autopsify distinguishes them: a verified zero is reported as zero, and an absent measurement is reported as unavailable, with the dependent figures withheld.
The same distinction applies throughout. Incomplete attribution is not proof that the unattributed revenue was organic; it is revenue whose source is unknown. Unavailable payment fees are not evidence that payment processing is free. In each case the honest statement is narrower than the convenient one.
The argument
Why false precision is dangerous
A missing figure is an obvious problem. A fabricated one is an invisible problem, and it is worse in three specific ways.
It cannot be audited
An assumed cost looks exactly like a measured one in the output. Nothing in the report tells the reader which figures to distrust, so the whole report has to be taken on faith or discarded entirely.
It propagates
Every figure below an assumption inherits it. One benchmarked cost quietly contaminates gross profit, contribution, margin percentages and any ranking built on them.
It changes decisions
A merchant acts on a profit figure. If that figure rests on an assumption that happens to be wrong for their business, the action is aimed at the wrong constraint — and the cost of the mistake is real even though the number looked precise.
An admitted gap does none of this. It tells the operator exactly what to go and find, and it leaves the figures that were established standing on their own.
In practice
What the classification is for
The classification is not a disclaimer. It is the thing that makes the rest of the method work: it defines where the reconstruction has to stop, it is what coverage is measured against, and it is what makes a later re-diagnosis comparable — two readings can only be compared if both were computed on the same evidence basis.
Every Blueprint is also read by a person before it is delivered. The classifications are produced deterministically by the engine; the review checks that what the report concludes is genuinely what those classifications will bear.
Methodology
The rest of the method
- The methodHow the whole diagnostic fits together.
- Economic reconstructionRebuilding the ladder from gross sales to whatever the evidence reaches.
- Measurement coverageHaving data is not the same as having enough of it.
- Verification & re-diagnosisWhat the reading about 30 days later can and cannot prove.
Find out which of your numbers are actually known
£497 for one diagnostic cycle. Every figure classified, every gap named, nothing assumed into existence.