McNamara fallacy¶

The decision-making error in which the measurable progressively displaces the important through a four-step ratchet — measure the measurable, arbitrary-value the rest, presume it unimportant, then declare it nonexistent — until an institution optimizes a proxy while its true objective silently drifts away.

Core Idea¶

The McNamara fallacy is the decision-making error in which the measurable progressively displaces the important through a four-step ratchet: first, measure what can readily be measured (defensible); second, disregard or assign arbitrary values to what cannot easily be measured (artificial); third, presume that the unmeasured is unimportant (blind); fourth, declare that what cannot easily be measured does not really exist (suicidal). Named for US Defense Secretary Robert McNamara's management of the Vietnam War — where enemy body counts served as the dominant metric of progress despite weak correlation with the actual objective of defeating insurgent will and political legitimacy — and formalized as a fallacy by Daniel Yankelovich in 1972.

The structural commitment is that the inventory of available measurements is endogenously narrower than the inventory of decision-relevant facts, and that institutional pressure to be "data-driven" silently substitutes the former for the latter. What makes this a ratchet rather than a single mistake is escalation: each step makes the distortion harder to detect because the substituted metric becomes the institutional record, and what the record does not capture is gradually treated as not there. The four-step ladder is the fallacy's distinctive contribution — not merely the observation that metrics are imperfect proxies, but the staged progression by which proxy substitution hardens into ontological erasure of the residual.

Structural Signature¶

Sig role-phrases:

the decision-relevant objective — the thing actually cared about, with both measurable and hard-to-measure components
the available-measurement inventory — the set of facts the instruments reach, endogenously narrower than the objective
the recorded proxy — the measurable stand-in that becomes the institutional record
the unmeasured residual — the decision-relevant facts the proxy is silent on, the thing the ratchet progressively erases
the institutional reinforcer — the dashboard, KPI, or evaluation regime that locks the proxy substitution in
rung 1 — measure the measurable — track what can readily be measured (defensible)
rung 2 — arbitrary-value the rest — disregard or assign the residual arbitrary numbers or zero (artificial)
rung 3 — presume unimportant — treat the unmeasured as not mattering, diverting budget and attention (blind)
rung 4 — declare nonexistent — write the residual out of the strategic vocabulary, an erasure no one decides on purpose (suicidal)
the ratchet property — each rung makes the distortion harder to detect, the late rungs go silent rather than loud, and descending costs far more than the local step that climbed it
the drift failure mode — the system optimizes the proxy while the true objective silently drifts away

What It Is Not¶

Not merely "you can't measure everything that matters." That flat complaint is the raw material; the fallacy's distinctive contribution is the staged ratchet — measure the measurable, arbitrary-value the rest, presume it unimportant, declare it nonexistent. Without the four rungs the critique cannot tell that step one has quietly become step four, which is exactly the diagnostic the concept supplies.
Not the mere use of a dominant metric. Measuring the measurable is rung one and is defensible; a metric-heavy regime is not by itself the fallacy. The error begins only at rung three, the first treatment of the unmeasured as unimportant rather than merely unmeasured. Branding every dashboard pathological over-applies the diagnosis as badly as excusing erasure under-applies it.
Not Goodhart's law. Goodhart names the response — agents gaming a measure once it becomes a target, so the measure degrades. McNamara names the prior move: the institution substituting the proxy for the objective and then erasing the residual. They compound but are distinct steps; the unmeasured can be erased with no one gaming anything.
Not loud overclaiming. The dangerous late rungs are not argued for — a fact treated as nonexistent generates no debate, so the signature is silence in the official account, not conspicuous boasting about the metric. Looking for overclaiming misses the erasure, which advances precisely by going quiet.
Not a single misjudgment. It is an escalating sequence, not a one-off mistake: each rung makes the distortion harder to detect because the proxy has become the record and the record resists contradiction by what it omits. A regime at rung two tends to advance to three rather than retreat to one, and descending the ladder costs far more than the locally-reasonable step that climbed it.
Not an independent causal mechanism. Its deep structural content is already carried by goodharts_law, legibility, streetlight_effect, commensurability, and proxy. What McNamara adds, and the only thing that does not reduce to those, is the staged-escalation framing — a diagnostic instrument that locates how far a regime has slid, not a distinct force doing explanatory work those primes do not already do.

Scope of Application¶

The McNamara fallacy lives wherever decisions are made under measurement constraints — across the management, public-policy, and evaluation subfields where an institution runs a quantitative regime against an objective with an unmeasurable residual. Its home is that decision-making-under-metrics domain; the loose proverbial sense that "numbers can mislead" is carried by the parent primes (goodharts_law, legibility), so the map below is confined to the literal institutional settings where the four-step ladder genuinely re-runs.

Management and corporate strategy — OKRs, KPI cultures, and executive dashboards, where the standing risk is optimizing the dashboard rather than the business and letting un-dashboard value silently fall off the strategic agenda.
Public policy and government performance — GDP standing in for national welfare, performance-based budgeting, and CompStat-style metric regimes; the Stiglitz–Sen–Fitoussi critique of GDP-centered policy is this fallacy applied to a national accounting metric.
Education — standardized testing operating as the de facto curriculum, with civic, creative, and artistic dimensions that the test does not score progressively reclassified as "non-academic" and cut.
Healthcare — process measures (e.g., the fraction of diabetics tested) substituting for outcome measures (actual morbidity), and pay-for-performance schemes optimizing the recorded measure rather than the patient.
Military strategy — body counts, kinetic-strike counts, and drone-strike metrics as proxies for progress; McNamara's own Vietnam-War body-count regime is the historical anchor and direct lineage.
AI/ML evaluation — benchmark optimization and leaderboard-chasing, with the AI-safety subfield reviving the diagnosis around reward misspecification and benchmark-gaming that crowd out unbenchmarked dangerous behavior.
Software engineering — velocity, story points, lines of code, and test-coverage as the operative metrics of "metric-driven" engineering cultures.
Academic assessment — citation counts, the h-index, and impact factor as proxies for research quality; the San Francisco Declaration on Research Assessment (DORA) is the fallacy applied to research-evaluation metrics.

Clarity¶

Naming the McNamara fallacy makes a slow institutional drift legible while it is happening, when at each step the move looks locally reasonable. The fallacy's clarifying force is its four-step ladder: it converts the flat complaint "you can't measure everything that matters" into a staged diagnostic that forces the question which step are we on? — are we merely measuring the measurable (defensible), or have we already crossed into presuming the unmeasured unimportant (blind) or declaring it nonexistent (suicidal)? Without the ladder, an organization sees only its current state and cannot tell that step one has quietly become step four; with it, the dangerous late steps — the ones least likely to be admitted aloud, because no one decides to erase the residual on purpose — become nameable and therefore contestable.

It also sharpens a distinction that "data-driven" culture works to blur: the gap between the decision-relevant objective and the available measurement inventory. The fallacy's claim is that the second is endogenously narrower than the first, so treating reported metrics as the full account silently amputates whatever the instruments do not reach. The sharper question a practitioner can now ask is not "is this metric accurate?" but "what is this metric silent on, and is the institution treating that silence as absence?" — separating a genuine measurement (a good proxy, honestly bounded) from the institutional habit of letting the proxy become the record and the record become reality.

Manages Complexity¶

Measurement-driven institutions generate an unruly catalogue of failures — body counts standing in for military victory, GDP for national welfare, test scores for education, process measures for patient health, benchmark leaderboards for model capability, citation counts for research quality. Each sits in a different bureaucracy with its own metric, its own objective, its own residual of unmeasured facts, and absent a frame they read as unrelated scandals each requiring its own post-mortem. The McNamara fallacy compresses that sprawl to a single staged regularity: in every case the measurable inventory is endogenously narrower than the decision-relevant inventory, and institutional pressure to be data-driven walks the same four rungs — measure the measurable, assign the rest arbitrary values, presume the unmeasured unimportant, declare it nonexistent. The whole catalogue collapses to one ladder applied to different objects.

What an analyst then tracks shrinks to a handful of quantities readable off any metric regime without re-deriving the pathology. Name the objective, name the proxy actually recorded, locate the gap — the decision-relevant facts the instrument does not reach — and read which of the four rungs the institution currently occupies: is it merely measuring the measurable (defensible), has it begun assigning the residual arbitrary numbers or zero (artificial), is it treating the residual as unimportant (blind), or has it written the residual out of existence (suicidal)? The rung is the qualitative outcome, and it follows from one diagnostic — what is this metric silent on, and how is the institution treating that silence? — rather than from a full audit of the measurement's accuracy. The branch structure is the four-rung ladder itself: each rung carries a known signature (the early rungs locally reasonable and admitted, the late rungs unadmitted because no one decides on purpose to erase the residual), so placing a case on the ladder both diagnoses its severity and predicts which corrective counterweight — honest bounding, qualitative review, independent audit — the regime is missing. The move is from a high-dimensional "which of these many metric regimes has gone wrong, and how badly" problem to a low-dimensional one: a proxy, an objective, the gap between them, and a position on a four-step staircase.

Abstract Reasoning¶

The fallacy licenses a tight set of reasoning moves, all keyed to the gap between the recorded proxy and the decision-relevant objective, and all sharper than the bare warning that metrics are imperfect.

Diagnostic — read the rung from the institutional posture, not the metric's accuracy. The characteristic inference runs from a surface symptom to a position on the ladder. Observe how the organization talks about what its instruments miss: if the unmeasured is still named and held open ("we lack a number for morale, so we judge it qualitatively"), the regime sits at rung one or two; if budget and attention flow away from the unmeasured residual on the unstated assumption that it matters less, infer rung three (blind); if the residual has dropped out of the strategic vocabulary altogether — reclassified as "non-core," "non-academic," "soft" — infer rung four (suicidal), the erasure no one decided on purpose. The tell of the late rungs is precisely that they are not argued for: a fact treated as nonexistent generates no debate, so silence in the official account, not loud overclaiming, is the diagnostic signature. From a flat valuation of an unmeasured good, infer the co-occurring marks — the proxy has become the institutional record, dissent about the residual has gone quiet, and the post-mortems will arrive only after the objective has already drifted.

Interventionist — insert counterweights matched to the rung, and predict where each bites. Because each rung has a known signature, the corrective follows from placement rather than from a full audit. At rung one the move is honest bounding: keep the proxy but annotate what it is silent on, and predict this alone arrests the slide if the silence stays visible. At rung two (arbitrary values, residual zeroed) the move is to refuse the false zero and restore qualitative review, predicting that re-naming the residual reopens the contest the arbitrary number had closed. At rung three the move is independent audit or narrative review by actors not evaluated on the proxy, predicting that only an outside vantage can re-raise a fact the inside has decided is unimportant. At rung four the prediction is harsher: no metric tweak recovers an erased category, because the instrument was never the problem — the ontology was; recovery requires rebuilding the account of what the institution is for before any measurement fix can land. Each intervention is thus a falsifiable claim that attention and resources will return to the residual, and the failure of a rung-appropriate fix to move them is itself evidence the regime sits a rung lower than diagnosed.

Boundary-drawing — separate the defensible use of proxies from the pathological slide. The fallacy draws a line cleanly: measuring the measurable is not the error, and the presence of a dominant metric is not by itself a McNamara fallacy. The boundary falls at rung three — at the first treatment of the unmeasured as unimportant rather than merely unmeasured. Below that line a metric-heavy regime is healthy; the move is to withhold the diagnosis when the residual is still honestly held open, however crude the proxy. This guards against the over-application that would brand every dashboard pathological and the under-application that excuses erasure as mere data-driven prudence.

Order-of-events — the ratchet predicts direction and irreversibility. The four rungs are not a menu but a sequence, and the staging carries predictive content: a regime at rung two is more likely to advance to three than to retreat to one, because the substituted proxy has by then become the record and the record resists contradiction by what it does not contain. So the move is to forecast the next rung from the current one and to act earlier than the visible symptom warrants — the cheapest place to break the ratchet is before the residual goes silent, since each rung makes the distortion harder to detect and the correction more expensive. The escalation also predicts the asymmetry of repair: descending the ladder costs far more than the local, locally-reasonable step that climbed it.

Knowledge Transfer¶

A first honesty about this entry: its "home domain" is already cross-institutional — decision-making under measurement constraints — so what transfers within the domain is a diagnostic and the four-step ladder, applied to one metric regime after another, rather than a mechanism native to a single substrate. Within that domain the transfer is clean and literal wherever the precondition holds: a decision-relevant objective with both measurable and hard-to-measure components, an available-measurement inventory narrower than the objective, and an institutional reinforcer (dashboard, KPI, evaluation regime) that can lock a proxy in. Given that precondition, the same procedure re-runs identically — name the objective, name the recorded proxy, locate the gap, read the rung — across management (OKRs, KPI cultures), public policy (GDP as a proxy for welfare; the Stiglitz–Sen–Fitoussi critique is essentially this fallacy applied to GDP), education (standardized testing as de facto curriculum), healthcare (process measures standing in for outcomes; pay-for-performance), military strategy (body counts, strike counts — McNamara's own lineage), AI/ML evaluation (benchmark-gaming, leaderboard-chasing, reward misspecification), software engineering (velocity, lines of code, coverage), and academic assessment (citation counts, h-index, impact factor; DORA is this fallacy applied to research metrics). These are not loose analogies but genuine re-applications of the same staged diagnostic to data that genuinely has the required shape; the vocabulary (proxy, residual, the four rungs from defensible to suicidal) and the rung-matched counterweights (honest bounding, qualitative review, independent audit) carry without translation. The cross-domain uses are many; the structural pattern they instantiate is one.

This is the place to be exact about what is and is not distinctive. The deep structural content of the fallacy is not unique to it — it is already carried, across all these same substrates, by a cluster of more general patterns: goodharts_law (a measure that becomes a target ceases to be a good measure — the response that compounds with McNamara's prior substitution), legibility (institutions impose simplifications that make the world manageable but amputate the illegible — McNamara is the decision-making face of this under quantitative bureaucracy), streetlight_effect (search where the light is, not where the keys are), commensurability (rendering heterogeneous values in one metric, with the residue forgotten), and proxy / surrogate_endpoint (the thing substituted, versus the institutional habit of substituting that McNamara names). So when the lesson is needed in a new measurement regime, the portable structural primitives are those parents; what McNamara's fallacy adds, and the only thing that does not reduce to them, is the staged-escalation framing — the specific ratchet measure → arbitrary-value → unimportant → nonexistent, with its prediction that the late rungs go silent rather than loud and that descending the ladder costs more than the locally-reasonable step that climbed it. That ladder is a genuine and useful diagnostic instrument, and it transfers literally wherever a metric regime has an unmeasured residual — but it is a teaching-and-diagnosis structure layered on shared primitives, not an independent mechanism, and the boundary to mark is between the ladder's legitimate diagnostic reach and over-reading it as a distinct causal force where Goodhart, legibility, and proxy already do the explanatory work. The cross-domain lesson should therefore carry those parent primes, with the four-step ladder as the instrument that locates how far a given regime has slid — which is exactly the reasoning developed in Structural Core vs. Domain Accent.

Examples¶

Canonical¶

The namesake case is the enemy body count in the Vietnam War under Defense Secretary Robert McNamara. The true objective was defeating the insurgency's will and establishing South Vietnamese political legitimacy — largely unmeasurable. What could be counted was enemy dead, so body count became the dominant metric of "progress," reported up the chain and used to allocate effort. Because the count was measurable and the objective was not, the ratchet climbed: unmeasured factors (popular support, legitimacy, morale) were first assigned no weight, then presumed unimportant, then dropped from the strategic picture — even as the war was being lost on exactly those dimensions. Daniel Yankelovich named the staged progression a fallacy in 1972.

Mapped back: Defeating insurgent will is the decision-relevant objective; body count is the recorded proxy, and legitimacy and morale are the unmeasured residual. The slide from counting the dead (rung 1) to writing legitimacy out of the strategic vocabulary (rung 4) is the full ratchet, ending in the drift failure mode — optimizing the proxy while the real objective slipped away.

Applied / In Practice¶

The U.S. No Child Left Behind Act (2002) reproduced the ladder in education. It mandated annual standardized testing in reading and mathematics and tied school consequences to those scores, making them the operative measure of school quality. Because only reading and math were tested, many schools reallocated time and budget away from science, social studies, art, music, and physical education — documented "curriculum narrowing" in which untested subjects were reclassified as non-essential. The unmeasured components of a good education did not merely go unmeasured; in many places they were treated as not counting toward what a school is for.

Mapped back: A well-rounded education is the decision-relevant objective; reading and math scores are the recorded proxy and the testing regime is the institutional reinforcer. Art, civics, and creativity form the unmeasured residual, and cutting them as "non-academic" is the rung-4 move — declaring the residual nonexistent — precisely the late-rung erasure that advances by quiet reclassification rather than open argument.

Structural Tensions¶

T1: Defensible proxy use versus pathological slide (a boundary at rung three that each step obscures). The fallacy draws a clean line: measuring the measurable is rung one and healthy, and a metric-heavy regime is not by itself pathological — the error begins only at rung three, the first treatment of the unmeasured as unimportant rather than merely unmeasured. But locating that line in a live institution is genuinely hard, because every step up the ladder looks locally reasonable and no one announces crossing it. The tension is that the concept must simultaneously withhold the diagnosis from honest proxy use (or it brands every dashboard pathological) and apply it before the erasure completes (or it excuses drift as data-driven prudence) — and the very gradualness that makes the ratchet dangerous is what makes the rung-three boundary invisible at the moment of crossing. Diagnostic: Is the unmeasured residual here still being held open as real-but-unmeasured (below the line, healthy), or has it begun to be treated as unimportant (rung three, the slide has started)?

T2: Silence as diagnostic versus silence as genuine irrelevance (the tell that cannot distinguish erasure from correct omission). The fallacy's sharpest observation is that the dangerous late rungs go silent, not loud: a fact treated as nonexistent generates no debate, so silence in the official account — not conspicuous overclaiming — is the signature of erasure. But silence is inherently ambiguous, because a residual that has been wrongly erased and a residual that genuinely does not matter both produce exactly the same institutional quiet. The tension is that the diagnostic tell (nothing is being said about X) cannot by itself separate "an important thing was written out of existence" from "an unimportant thing is correctly ignored," so reading every silence as suppressed importance re-inflates trivia into strategic priorities, while dismissing silence as irrelevance is precisely how rung-four erasure hides. The signal that detects the fallacy is the same signal a well-run institution produces about things that truly don't count. Diagnostic: Is the silence around this residual the quiet of erasure (a decision-relevant fact written out) or the quiet of genuine irrelevance (a fact correctly left aside) — and what evidence distinguishes them beyond the silence itself?

T3: Measurable substitution as error versus functional necessity (institutions must act on what they can see). The fallacy condemns substituting the measurable proxy for the decision-relevant objective — but institutions cannot manage, budget, or coordinate on what they cannot render legible, so some reliance on the measurable is not a mistake but the precondition of acting at scale at all. The tension is that the substitution the concept warns against is continuous with the substitution that makes bureaucratic action possible: an organization that refuses all proxies flies blind, while one that adopts them takes the first, defensible step onto a ladder whose top rung is erasure. The fallacy therefore names a pathology that grows out of a necessity, and cannot simply prescribe "don't substitute" — it must prescribe substituting while keeping the residual visible, a discipline that runs against the whole institutional pull toward the tractable number. Diagnostic: Is the proxy here being used as a necessary handle on the objective with the residual kept explicit, or has managing-on-the-measurable slid into treating the measurable as the whole objective?

T4: Ratchet irreversibility versus early action (the cheapest fix is the least justifiable). The staging carries a hard asymmetry: each rung makes the distortion harder to detect and the correction more expensive, so the cheapest place to break the ratchet is early — before the residual goes silent — and descending the ladder costs far more than the locally-reasonable step that climbed it. But early is exactly when intervention is least justifiable, because nothing has visibly gone wrong: at rung one or two the proxy still tracks the objective well enough that spending resources to protect an as-yet-uncompromised residual looks like waste. The tension is that the evidence needed to justify the fix accumulates precisely as the fix becomes prohibitively expensive, so acting when it is cheap means acting on a forecast rather than a symptom, and waiting for the symptom means acting when recovery may require rebuilding the institution's account of what it is for. Diagnostic: Is intervention here being timed to the visible symptom (late, expensive, near-irreversible) or to the forecast of the next rung (early, cheap, but not yet obviously warranted)?

T5: McNamara versus Goodhart (a prior substitution and a downstream gaming that compound). The fallacy is easily fused with Goodhart's law, but they name different moves: McNamara is the prior substitution — the institution replacing the objective with the proxy and then erasing the residual — while Goodhart is the response, agents gaming the measure once it becomes a target so the measure degrades. The tension is that in real metric failures the two almost always compound (a substituted proxy invites the gaming that further corrupts it), yet they are distinct steps with distinct remedies: the unmeasured can be erased with no one gaming anything (pure McNamara — the residual just vanishes), and a measure can be gamed without any residual being erased (pure Goodhart). Attributing a metric failure wholly to one misses the other, so a Goodhart-only reading fixes the gaming while the erased objective stays gone, and a McNamara-only reading restores the residual while the gaming continues. Diagnostic: Is the failure here the substitution and erasure of the objective (McNamara — restore the residual) or the gaming of a target measure (Goodhart — fix the incentive), or the compound of both requiring both remedies?

T6: Autonomy versus reduction (a staged diagnostic instrument or the composition of its measurement-pathology parents). The McNamara fallacy is a specific, named construct that transfers literally across metric regimes (management, policy, education, healthcare, AI evaluation) as a diagnostic — name the objective, name the proxy, locate the gap, read the rung. But its deep structural content is not its own: it is already carried, across those same substrates, by goodharts_law, legibility, streetlight_effect, commensurability, and proxy/surrogate_endpoint. What McNamara adds, and the only thing that does not reduce to those parents, is the staged-escalation framing — the four-rung ratchet (measure → arbitrary-value → unimportant → nonexistent) with its predictions that the late rungs go silent and that descending costs more than climbing. The tension is that this is a genuine and useful diagnostic instrument layered on shared primitives, not an independent causal mechanism, so over-reading it as a distinct force double-counts explanatory work the parent primes already do. Diagnostic: Resolve toward goodharts_law + legibility + proxy (with streetlight_effect, commensurability) when explaining why a metric regime drifts; toward the McNamara ladder when the task is locating how far along the erasure a given regime has already slid.

Structural–Framed Character¶

The McNamara fallacy sits firmly at the framed pole of the structural–framed spectrum, alongside the other named fallacies, and it is arguably an even purer framed case than most, because by the entry's own admission its distinctive contribution is not a mechanism at all but a diagnostic framing layered on parent primes that do the causal work. On evaluative_weight it scores high: to call a regime's behavior "the McNamara fallacy" is to convict it — the very rungs are graded as a moral descent (defensible → artificial → blind → suicidal), so the label is a normative verdict about a decision process, not a neutral description of a mechanism the way "feedback" or "integration" names something that is neither good nor bad. Human_practice_bound is high: the concept is constituted by the institutional practice of running a quantitative regime against an objective and dissolves the instant that practice is removed — with no institution, no proxy, no dashboard, and no objective being managed, there is nothing for the ratchet to climb; it is not a fact of nature but a pathology of organized human decision-making. Institutional_origin is equally pronounced: the entry is furniture of a specific tradition — named by Yankelovich, anchored in McNamara's Vietnam body-count regime, and formalized as a four-rung ladder that is an invented diagnostic instrument of management and decision theory, not substrate-neutral form. On vocab_travels it scores low (proxy, residual, the rungs, KPI, dashboard are decision-making-under-metrics vocabulary that loses its referents off that substrate) and on import_vs_recognize the deep content is not even proprietary: within metric regimes the ladder re-applies literally, but the explanatory work is recognized as belonging to the parent primes, and the looser "numbers can mislead" proverb is import-by-analogy carried by those parents.

The portable structural skeleton is the measurable displacing the important — proxy substitution in which an institution optimizes what its instruments reach while the decision-relevant residual is silently dropped. But that skeleton is precisely what the McNamara fallacy instantiates from its umbrella primes — proxy/surrogate_endpoint (the substituted stand-in), goodharts_law (the target-corrupted measure), legibility (the illegible amputated to make the world manageable), with streetlight_effect and commensurability — not what makes "McNamara fallacy" itself travel: the cross-domain reach belongs to those parents, which already carry the substitution across every one of the same substrates, while what is proprietary to McNamara — the staged-escalation ratchet (measure → arbitrary-value → unimportant → nonexistent), with its predictions that the late rungs go silent and that descending costs more than climbing — is by the entry's own account a teaching-and-diagnosis framing rather than an independent causal force. That is the framed pole in its clearest form: even the portable structure defers to the parents, and the distinctive residue is a normatively-charged instrument for locating how far a practice has slid. Its character: a practice-constituted, morally-graded fallacy label whose every distinctive feature is decision-theory furniture, structural only in the proxy-substitution skeleton it borrows from its measurement-pathology parents and frames as a four-rung verdict.

Structural Core vs. Domain Accent¶

This section decides why the McNamara fallacy is a domain-specific abstraction and not a prime, and carries the case for its domain-specificity — a case sharpened by the fact that, on the entry's own account, its distinctive contribution is a diagnostic framing rather than a mechanism.

What is skeletal (could lift toward a cross-domain prime). Strip the institutional setting and a thin relational structure survives: an actor optimizes what its instruments can reach while the decision-relevant residual the instruments miss is silently dropped — the measurable displacing the important through proxy substitution. The portable pieces are abstract: an objective with a hard-to-measure component, a measurement inventory endogenously narrower than the objective, a recorded proxy that becomes the working reality, and an unmeasured residual that goes unattended. That skeleton is genuinely substrate-portable — which is exactly why it is already carried, across every one of the same substrates, by a cluster of parents: proxy/surrogate_endpoint (the substituted stand-in), goodharts_law (the target-corrupted measure), legibility (the illegible amputated to make the world manageable), streetlight_effect, and commensurability. It is the substitution core the McNamara fallacy shares, not what is proprietary to it.

What is domain-bound. What makes the concept the McNamara fallacy in particular is not a substrate's worked vocabulary but the staged-escalation framing laid over decision-making-under-metrics: the four-rung ratchet (measure the measurable → arbitrary-value the rest → presume it unimportant → declare it nonexistent), graded as a moral descent from defensible to suicidal, with its predictions that the late rungs go silent rather than loud and that descending the ladder costs more than the locally-reasonable step that climbed it. The operative vocabulary — proxy, residual, the rungs, KPI, dashboard — is pinned to the institutional-metrics substrate. The decisive test: remove the institution running a quantitative regime against a managed objective and there is no ratchet to climb — no proxy, no dashboard, no residual — so the concept dissolves the instant the human decision-making practice is withdrawn.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. The McNamara fallacy's transfer is bimodal, but with a distinctive edge: even the portable structure defers to the parents. Within metric regimes the four-rung ladder re-applies literally across management, policy, education, healthcare, military strategy, AI evaluation, and academic assessment — but as a diagnostic instrument for locating how far a regime has slid, not as a distinct causal force, since the explanatory work is recognized as belonging to goodharts_law, legibility, and proxy. Beyond those regimes the proverbial "numbers can mislead" is import-by-analogy carried by the same parents. So the only thing that does not reduce to the parents is the staged framing itself, which is a teaching-and-diagnosis structure rather than an independent mechanism. The cross-domain reach belongs to the measurement-pathology parents; what stays home is a normatively-charged four-rung verdict for locating how far a practice has already slid.

Relationships to Other Abstractions¶

Current abstraction McNamara fallacy Domain-specific

Parents (4) — more general patterns this builds on

McNamara fallacy is a kind of Legibility Prime

McNamara Fallacy is the quantitative decision-regime species of Legibility in which an institution's simplified record becomes its actionable reality.
McNamara fallacy is a kind of Proxy-Target Divergence Prime

McNamara Fallacy is Proxy-Target Divergence specialized to an institution that lets a measurable stand-in displace and progressively erase its decision-relevant residual.
McNamara fallacy is part of, conditional Commensurability Prime

When the second rung arbitrary-values heterogeneous residuals, McNamara Fallacy contains Commensurability as the common-metric construction that permits the flattening.

Condition / exception Applies when the second rung assigns the residual arbitrary numbers or forces heterogeneous values onto one metric; a regime may instead disregard the residual without commensurating it.
McNamara fallacy is part of Streetlight Effect Prime

McNamara Fallacy contains the Streetlight Effect at its first rung because institutional attention begins by concentrating on what the measurement apparatus illuminates.

Hierarchy paths (9) — routes to 5 parentless roots

McNamara fallacy → Legibility → Abstraction

Not to Be Confused With¶

Goodhart's law. The distinct move in which agents game a measure once it becomes a target, so the measure degrades. McNamara names the prior substitution — the institution replacing the objective with the proxy and then erasing the residual — which can happen with no one gaming anything; the two compound but are separable steps with different remedies. Tell: is the failure the substitution-and-erasure of the objective (McNamara — restore the residual), or the gaming of a target measure (Goodhart — fix the incentive)?
Streetlight effect. Searching where the light is rather than where the keys are — attending to what is easy to observe rather than to what is relevant. It captures rung one's bias toward the measurable but lacks McNamara's staged escalation into presuming-unimportant and declaring-nonexistent. It is a parent primitive McNamara builds on, not the ratchet itself. Tell: is the point only that attention goes to the observable (streetlight), or that the observable progressively erases the unobserved across four rungs (McNamara)?
Legibility. The pattern of institutions imposing simplifications that make the world manageable while amputating the illegible. McNamara is the decision-making face of legibility under a quantitative regime — it carries the same amputation but adds the specific staged ratchet and its silent late rungs. Tell: is the claim the general institutional flattening of the illegible (legibility), or the specific four-step slide from measuring to erasing a residual (McNamara)?
proxy / surrogate_endpoint. The substituted stand-in itself — a measurable quantity used in place of the objective. McNamara names not the proxy but the institutional habit of letting the proxy become the record and then the reality, staged as a ratchet. Tell: is the object the stand-in measure as such (proxy), or the escalating practice of substituting it and erasing what it omits (McNamara)?
Commensurability. Rendering heterogeneous values in a single common metric, with the incommensurable residue forgotten. This is a parent primitive that supplies McNamara's "arbitrary-value the rest" rung, but it names the collapsing-to-one-scale move, not the staged descent to erasure. Tell: is the issue forcing unlike goods onto one scale (commensurability), or the four-rung progression by which the off-scale residual is written out (McNamara)?
"You can't measure everything that matters" (the flat proverb). The bare complaint that metrics are imperfect proxies. This is the fallacy's raw material, not the fallacy: McNamara's distinctive contribution is the staged ratchet that lets you see step one has become step four. Tell: is it a one-line caution that measurement is incomplete (the proverb), or a staged diagnostic that locates which rung a regime occupies (McNamara)?

Neighborhood in Abstraction Space¶

McNamara fallacy sits in a moderately populated region (45^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Proxy Metrics & Venture Adaptation (13 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12