How a number becomes defensible.
A figure that cannot be reproduced is not a finding. It is an assertion wearing a finding's clothes.
Methods have to survive being checked
Most of the analytical work worth doing is less about producing a number than about making sure the buffer analysis and spatial join behind it survive being checked by anyone entitled to ask how it was derived.
That standard has a practical consequence: given a choice between a clever model only I can run and a plain buffer analysis anyone can re-run, the plain one wins. Not out of modesty: a number anyone can re-run is a number that can be acted on rather than merely believed.
Definitions come before measurement
A surprising share of disagreement about numbers turns out to be disagreement about words. Before anything can be counted, the boundary has to be constructed: what counts as reachable, what separates one category from the next, which of two conflicting records wins when unique-key matching fails and the match rests on confidence scoring.
Where those definitions did not exist, I have written them, worked them through with the teams they bind, and then held them steady, because a definition that moves between reports is worse than no definition at all.
Sources disagree; say so
Reconciling systems that were never designed to agree is most of the job. Two platforms describing the same thing will differ; record linkage and duplicate detection narrow that gap, never close it. The honest output is not a single confident figure but a figure plus the residual: what still does not reconcile, and how much it could move the answer. Carrying that uncertainty forward is what makes the next decision safe.
The corollary is worth stating plainly: sometimes the defensible answer is that the plan needs a different shape. Carrying that upward is an obligation to the people who will act on the figure, not a failure of the analysis.
A method only its author can run is a dependency, not an asset.
It also cannot be checked, which is the more serious problem.
There are no maps on this page because the work it describes is commercially confidential. The reasoning is not, and the reasoning is the part that transfers.
One model, five altitudes. Every level stays traceable to the one below.
Real project maps are commercially confidential. Rather than redact one into meaninglessness or invent a fake, this shows the method at a level of abstraction that discloses nothing. With reduced motion enabled, and on any narrow screen, it becomes a small-multiple sequence: five frames read top to bottom, each labelled.