Attribution Is a Judgment Problem Before It Is a Tracking Problem.

Attribution is a three-layer stack. Tracking integrity is the first layer, counting conventions is the second, operator judgment is the third. Most disputes are framed as tracking failures and lived as judgment failures. The diagnosis names which layer is failing.

The claim has two parts. The first is structural: every operator-facing attribution problem can be sorted into one of the three layers, and the right intervention depends on which layer is named. A pixel that does not fire is a layer-one problem. A platform that counts view-through against another platform's click is a layer-two problem. An operator who has not declared which read is the source of truth is a layer-three problem. The interventions are different. The cost of treating a layer-three problem as a layer-one problem is six months of measurement-engineering work that does not close the disagreement.

The second part is operational: layer-three problems compound. The longer the operator goes without picking a source of truth, the more vendor relationships, agency contracts, and budget allocations get scoped against contradictory reads. The fix at month one is one document. The fix at year three is a renegotiation across every party that has been steering against a different read.

The position is not "tracking does not matter." Tracking matters. The position is tracking is necessary and not sufficient. Without the judgment layer, tracking improvements close one gap and surface a louder one.

The structural argument is that tracking improvements at layer one cannot resolve disputes at layer three. The two layers are not stacked in the sense of fixing the lower one fixes the upper one; they are stacked in the sense of each is necessary and the upper layers are entirely independent of the lower ones once the lower ones are roughly intact.

Five failure modes follow.

Failure mode one. Tracking is mostly already fine. On most operating Shopify-plus-Google-plus-Meta accounts, the technical layer is already roughly intact. Pixels fire. The Conversions API matches client-side at ninety-percent-plus. GA4 events fire on the right pages. The disagreement at the platform level is twenty to seventy-five percent. Two-percent technical defects do not produce that scale of disagreement. The team that audits tracking finds defects, fixes them, and is surprised the disagreement persists. The disagreement persists because it was never technical.

Failure mode two. Tracking improvements surface conventional disagreement. When server-side conversions are installed, the platforms now have cleaner data. The cleaner data still applies different windows, different definitions, different de-dupe rules. The reads are now cleaner and louder in their disagreement. The team that read the post-install report concluded the installation was wrong. The installation was right; the disagreement was conventional all along.

Failure mode three. Conventional fixes do not survive without the judgment layer. An operator can write down the canonical conventions (click window, view-through, de-dupe, recurring orders) and still leave the source-of-truth question unanswered. Each vendor will read the canonical conventions in the way that favors the vendor's work. The Google specialist will defend the Google convention; the Meta specialist will defend the Meta convention. The conventions document on its own does not pick. The judgment-layer pick is what closes the argument.

Failure mode four. The longer judgment is absent, the more expensive the fix becomes. Vendors get hired against contradictory reads. Agencies sign contracts with KPIs that reference the read favoring the agency's work. The board pack settles into a slide that consolidates four reads into one without naming any. Reversing all of that takes a renegotiation, not a document. The cost of the absent decision compounds at the rate of new contracts that calcify around the absence.

Failure mode five. The CFO's question stays open. The CFO asks: which channel produced which dollar. The marketing team answers with whichever number favors the team's preferred reading. The CFO loses confidence in the marketing team's read and starts triangulating against the bank statement. The relationship between marketing and finance becomes adversarial, not because the marketing team is dishonest but because the marketing team has no source-of-truth document and therefore no defensible answer. The fix is the document. The relationship cannot be repaired without it.

The working spec runs three numbered moves per layer. Validate, audit, install. The moves complete in writing and the operator signs off before moving up the stack. The whole diagnostic completes in roughly seventy-two hours of audit time on a typical operating account.

Every methodology has assumptions. Naming the assumptions is part of defending the position. The three-layer diagnostic assumes layer-one data hygiene is roughly intact or known-broken in named ways. The methodology does not handle every operator-side configuration.

The Reference pages in the Attribution cluster, the case files this position was written against, the companion position, and the hub. The graph below is the cluster map.