Vanity-Metric Addiction¶

A team locks onto a metric chosen for how impressive it looks rather than its causal link to the outcome, then keeps it after the disconnect is known because dropping it carries social cost.

Core Idea¶

Vanity-metric addiction is the pattern in which a team selects and optimizes a metric that is visible, easy to move, and flattering — but causally disconnected from the outcome the organization actually needs (retention, profitability, mission delivery) — and then becomes locked into that metric through social and psychological reinforcement loops that persist even after the disconnect is recognized. The structural mechanism has three coupled layers. First, the metric is selected by the wrong criterion: it is picked for its reportability — it rises, looks impressive in a deck, and can be attributed to team action — rather than for its causal fidelity to the underlying outcome. Registered users, pageviews, app downloads, and monthly-active-user counts defined permissively all share this profile: they are real quantities that genuinely measure something, but what they measure does not sit on the causal path from team action to the outcome the business depends on. Second, effort flows to whatever moves the metric: strategy, roadmap, OKRs, and sprint planning all optimize for the visible number, producing moves that grow the dashboard while the underlying outcome stagnates or degrades. A promotion floods the MAU count with low-intent users; a weekly digest email inflates engagement without producing the retention or conversion the company needs. Third — and this is what the "addiction" framing captures — the metric becomes socially load-bearing: the board expects it, the CEO quotes it, performance reviews reference it, and investor decks display it in the hero position. Dropping it carries social cost even when everyone in the room privately knows it is wrong. The reward loop is psychological and organizational, not analytical, which is exactly why ordinary "this metric isn't working" feedback fails to dislodge it: the metric is retained not because it is believed but because no one bears the cost of replacing it with an outcome metric that is harder to move, harder to explain, and slower to vindicate team effort. Eric Ries's lean-startup framing of this pattern (the distinction between vanity metrics and actionable metrics) identifies the pre-registration cure: before adopting a metric, articulate in writing the causal chain from team action to metric movement and from metric movement to outcome change; vanity metrics systematically fail at one of those two links and cannot survive the requirement.

Structural Signature¶

Sig role-phrases:

the outcome objective — the result the organisation actually needs (retention, profitability, mission delivery), the thing the metric is nominally a window onto
the vanity metric — a real, visible, easy-to-move, flattering number (registered users, pageviews, downloads, permissive MAU) selected for reportability rather than causal fidelity
the two-link causal chain — the (often unstated) path: team action → metric movement → outcome change, which a vanity metric fails at one link of
the selection-by-reportability — the metric is picked because it rises, looks impressive in a deck, and can be attributed to team action, not because it sits on the causal path
the optimization loop — strategy, roadmap, OKRs, and sprint planning all bend toward the visible number, producing moves that grow the dashboard while the outcome stalls or degrades (low-intent users, engagement theater)
the social-reinforcement lock-in — the "addiction" layer: the board expects it, the CEO quotes it, reviews reference it, the investor deck features it, so dropping it carries social cost independent of its truth
the persistence-after-refutation — the tell of the second layer: a cohort study lands, everyone agrees in principle, and the metric reappears unchanged in the next board deck because no one bears the cost of replacing it
the scissors signature — the surface diagnostic: the headline number climbing steeply while the outcome it indexes sits flat or drifts down, widening the longer the loop runs
the pre-registration and incentive-surgery cures — articulate the two-link chain in writing before adoption (vanity metrics fail it), demote the proxy off the default view, and — the binding move — tie status and compensation to the outcome rather than the proxy

What It Is Not¶

Not a measurement or instrumentation problem. A dashboard climbing while retention and revenue sit flat is not a sign the metric is noisy, the cohort immature, or the model under-fed; the number's problem is that it was never on the causal path to the outcome. Better instrumentation and cleaner attribution cannot redeem a metric that fails the two-link test — no amount of analytic polish installs a causal relation that was never there.
Not a false or fake number. Vanity metrics are real quantities that genuinely measure something — registered users, pageviews, downloads, permissive MAU all rise and can be honestly reported. The defect is not falsity but causal disconnect: what they measure does not sit on the path from team action to the outcome the business depends on. Auditing them for accuracy misses the point, because they are usually accurate.
Not "the metric is inaccurate." The question the concept sharpens is causal fidelity versus reportability, not precision. A perfectly accurate, well-instrumented number is still a vanity metric if it was selected because it rises, looks impressive, and can be attributed to team action rather than because it indexes the outcome. Accuracy and relevance are separately accountable, and improving the former does nothing for the latter.
Not curable by teaching better metrics. The stickiness lives in the social-reinforcement layer — the board expects it, the CEO quotes it, reviews and decks reward it — so it survives even after everyone privately agrees it is wrong. Education attacks a problem the team does not have; the binding intervention is incentive surgery (tie status and compensation to the outcome), because until someone's reward no longer depends on the flattering number, every analytic fix reverts at the next board meeting.
Not Goodhart's law proper. Goodhart is about a proxy drifting from its target under optimization pressure; the "addiction" layer is about retaining a proxy already known to be hollow because its load-bearing function is social, not epistemic. The distinctive content here is the persistence-after-refutation — a metric kept not because it is believed but because no one bears the cost of replacing it — which Goodhart's drift dynamic does not capture.

Scope of Application¶

Vanity-metric addiction lives within product, marketing, and innovation analytics; its reach is within that domain, across every setting that runs on a dashboard where a flattering, easy-to-move number can be selected over a causally faithful one and then socially locked in. The genuinely distant cousins (Campbell's law, teaching-to-the-test, RL reward hacking) belong to the broader proxy-target-divergence / Goodhart parent and the separate social-lock-in pattern, not to "vanity-metric addiction" by name.

Software startups — permissively-defined MAU, registered users, pageviews, and downloads climbing while paying-user retention and contribution margin stagnate.
Marketing — impressions, reach, follower count, and brand-mention volume moving while lead conversion and attributed revenue do not.
Educational technology — minutes-of-use and lessons-completed rising while skill assessments do not.
Public-sector innovation — pilots-launched and participant counts reported while employment, health, or cost-avoided outcomes go unmeasured.
Health-tech — devices distributed and accounts activated reported as success while clinical-outcome change is unmeasured.
Internal team metrics — story points completed, PRs merged, and tickets closed moving while customer-facing reliability and delivered functionality do not.

Clarity¶

Naming vanity-metric addiction makes legible a failure that product teams routinely misread as a measurement problem. Without the label, a dashboard climbing while retention and revenue sit flat invites the wrong diagnosis — the metric is "noisy," the cohort is "still maturing," the model "needs more data" — and the proposed fixes (better instrumentation, cleaner attribution) cannot touch a number whose problem is that it was never on the causal path. With it, the analyst asks a sharper question: was this metric chosen for causal fidelity to the outcome or for reportability? That single split separates a real leading indicator from a flattering proxy, and it tells a team that no amount of analytic polish will redeem a metric that fails the pre-registration test — articulating, in writing, how team action moves the number and how the number moves the outcome — at one of its two links.

The concept also sharpens a distinction ordinary metric critiques blur: why the metric is wrong versus why it persists. Vanity-metric addiction insists these are two separate layers — a causal disconnect (the analytic defect) and a social-reinforcement loop (the reason it survives detection). Localizing the stickiness to the second layer is what the "addiction" framing buys: it explains why "this metric isn't working" feedback reliably fails, reframing the problem from one of education (teach the team better metrics) to one of incentive and status (no one bears the cost of replacing a number the board expects and the deck rewards). The practitioner can then stop asking "is this metric accurate?" and start asking the load-bearing question — "who would pay the social cost of dropping it, and what outcome metric, harder to move and slower to vindicate, would have to take its place?"

Manages Complexity¶

Product analytics generates a steady catalog of named pathologies — top-of-funnel inflation, MAU theater, growth-hacking that lifts sign-ups but not lifetime value, NPS gaming, attribution confusion, the digest email that manufactures engagement — and a team can spend its diagnostic energy treating each as its own puzzle with its own cohort study and its own remedy. Vanity-metric addiction collapses that list into one structure: a number chosen for reportability rather than causal fidelity, sustained by a social-reinforcement loop after the disconnect is known. Once a team holds that frame, it stops re-deriving each case and instead runs every metric through two scalars — does team action move the number through a real causal link, and does the number move the outcome through one — plus a third question about who pays the social cost of dropping it. The qualitative reading follows from those: a metric that passes both causal links is a real leading indicator; one that fails either is a vanity metric whose stickiness lives entirely in the social layer, where no analytic fix can reach it. A sprawling, case-by-case audit of "which dashboard numbers are lying and why won't they die" becomes a short, repeatable test over a two-link causal chain and one incentive question — the same small parameter set predicting both whether a metric is hollow and why it survives detection.

Abstract Reasoning¶

Vanity-metric addiction licenses a tight set of inferences, each running off the two-layer split — the causal disconnect and the social-reinforcement loop — rather than re-applying the compression.

Diagnostic (infer the disconnect from its surface signature). The tell is a divergence shape: a headline number climbing steeply while the outcome it supposedly indexes sits flat or drifts down. From that scissors pattern — MAU up 20% month-over-month, net-new paying customers flat for six months, revenue-per-customer declining — the analyst infers not noise or cohort immaturity but a broken causal link, and predicts the corroborating evidence: a cohort decomposition will show the growth concentrated in a low-intent segment (a promotion's free accounts, a digest email's one-click logins) whose downstream conversion is near zero. The direction of inference is fixed: from number rises but outcome doesn't follow to the number was selected for reportability, and the bulk of its movement traces to actions that touch the metric without touching the outcome. A second, sharper diagnostic distinguishes which of the two links is severed — does team action fail to move the number through a real mechanism, or does the number fail to move the outcome? — because the answer dictates whether the metric is merely useless (no action-to-metric link, so optimizing it is wheel-spinning) or actively misleading (action-to-metric link intact but metric-to-outcome link absent, so optimizing it manufactures the illusion of progress).

Diagnostic (infer the addiction from the response to disconfirmation). The "addiction" layer has its own signature, read not off the dashboard but off the organization's reaction when the disconnect is demonstrated. When a cohort analysis lands, everyone agrees in principle, and the metric reappears in the next board deck unchanged, the analyst infers the stickiness is social, not epistemic — the number is retained not because it is believed but because no one will absorb the cost of dropping it. The reasoning move is to treat persistence after refutation as positive evidence of the reinforcement loop: ordinary error gets corrected once shown; a vanity metric survives precisely because its load-bearing function (board expectation, CEO talking point, review reference, investor-deck hero position) is independent of its truth. Continued display in the face of an accepted refutation is therefore diagnostic of the second layer, the way a flat valuation curve is diagnostic of substitution.

Interventionist (what to change, and what each change should do). Because the defect is layered, interventions sort by which layer they target, and the concept predicts which will move the outcome. Analytic fixes aimed at the first layer — better instrumentation, cleaner attribution, more data — are predicted to fail outright when the link severed is metric-to-outcome, since polishing a number's accuracy cannot install a causal relation it never had; this is a falsifiable prediction (the outcome stays flat after the instrumentation improves). The pre-registration intervention attacks selection: requiring, in writing and before adoption, an articulated chain from team action to metric movement and from metric movement to outcome change is predicted to filter out vanity metrics at one of the two links, because they cannot survive an explicit statement of a chain they do not complete. The structural-demotion intervention (moving the proxy off the default board view, pairing every flattering metric with a harder actionable one) targets the optimization loop, redirecting effort by changing what is visible. And the incentive intervention — tying status and compensation to the outcome rather than the proxy — is the only one aimed at the second layer, and the concept predicts it is the binding one: until someone's reward no longer depends on the flattering number, education and better metrics leave the loop intact. The ordering is itself a prediction: move the incentive and the metric becomes droppable; leave it and every other fix is reversible at the next board meeting.

Boundary-drawing (which regime, and the gate on what the number licenses). The concept draws a line between a real leading indicator and a vanity metric by the two-link test, and the line gates a downstream inference: only a number that passes both links may be read as a proxy for the outcome and fed forward into a decision — a forecast, an OKR, a cost-benefit aggregation. A metric that fails either link is barred from that role; multiplying it out (extrapolating MAU growth into a revenue projection) yields a figure with no defensible relation to the outcome, an artifact of the broken link rather than a measurement of progress. The boundary also delimits where the "addiction" reading applies at all: the social-lock-in story is licensed only once the causal disconnect is known to the organization. A team that has simply not yet noticed its metric is hollow has a detection problem, not an addiction; the addiction frame attaches specifically to the regime where the disconnect has been recognized and the metric is retained anyway. Misapplying it to the pre-recognition case mistakes ignorance for incentive-capture and prescribes the wrong remedy (education rather than incentive surgery).

Predictive / order-of-events. The mechanism implies a characteristic life cycle the analyst can read forward. A metric is adopted for reportability; effort reorganizes around it (strategy, roadmap, OKRs, sprint planning all bend toward the visible number); the metric climbs while the outcome stalls; the gap eventually surfaces, often through an outsider's cohort study; agreement-in-principle follows; and — absent an incentive change — reversion follows the agreement, because the social layer was never touched. Knowing the sequence lets the analyst predict, at the agreement-in-principle stage, that reversion is the default outcome unless the conversation has moved from "is this metric accurate?" to "who pays the social cost of dropping it, and what slower-to-vindicate outcome metric replaces it?" The order of events also predicts a degradation signature: because effort flows to whatever moves the number, the underlying outcome does not merely stagnate but can actively worsen (low-intent users crowding the funnel, engagement theater displacing retention work), so the longer the loop runs unbroken, the wider the scissors between dashboard and outcome grows.

Knowledge Transfer¶

Within product, marketing, and innovation work the pattern transfers as mechanism across every setting that runs on a dashboard, because the two-layer structure it isolates — a number chosen for reportability over causal fidelity, sustained by a social-reinforcement loop after the disconnect is known — is the same regardless of which number it is. The diagnostics (the scissors signature of a headline climbing while the outcome sits flat; the two-link causal test; persistence-after-refutation as the tell of the social layer) and the layered intervention menu (pair every flattering metric with an actionable one; structurally demote the proxy off the default board view; pre-register the causal chain from team action to metric to outcome; apply the "so what changed in customer outcomes?" test; and — the binding one — tie incentives to outcome rather than proxy) carry intact across software startups (permissively-defined MAU while paying-user retention stalls), marketing (impressions and follower count while attributed revenue does not move), educational technology (minutes-of-use rising while skill assessments do not), public-sector innovation (pilots-launched and participant counts reported while employment or health outcomes go unmeasured), health-tech (devices distributed while clinical-outcome change is unmeasured), and internal team metrics (story points and tickets closed while customer-facing reliability does not improve). These are not analogies between separate problems; they are one structure with the flattering number swapped, so the pre-registration and incentive-surgery remedies proven against MAU theater apply unchanged to pilots-launched theater.

The reach of the named concept stops at the edge of that product/marketing/innovation domain, and honesty requires marking why its apparent breadth is one substrate replayed: the value-add that makes the entry distinctive — the addiction framing, the flattering-and-visible selection criterion, the product-analytics operational vocabulary, the investor-deck signaling dynamics — is domain idiom, and the settings above transfer because they share that reporting substrate.

What genuinely travels to distinct substrates is the pattern beneath vanity-metric addiction, and that — not the named concept — is what should carry any cross-domain lesson (case B), and here it is unusually clear because the entry carries two portable shapes. The first is the first layer: a measured proxy diverging from the outcome it was meant to track, especially under optimization pressure — proxy-target divergence, the Goodhart's-Law family — which recurs as genuine co-instances across domains that share no product-analytics machinery: Campbell's law and teaching-to-the-test in education, the Lucas critique in macroeconomics, target-fixation in policy, and reward hacking in reinforcement learning. The second is the "addiction" layer itself: organizational lock-in to a known-bad indicator through social-reinforcement loops — an escalation-of-commitment-to-flattering-measures shape — which is distinct from Goodhart proper (Goodhart is about a proxy drifting under optimization; the addiction layer is about retaining a proxy already known to be hollow because its load-bearing function is social, not epistemic) and which recurs wherever status, expectation, or sunk reputation keep a discredited measure in place. The cross-domain insight belongs to those parents (the proxy-target-divergence / Goodhart pattern, partially adjacent to existing performativity, signaling, and goal_congruence_alignment; and the social-lock-in / commitment-escalation pattern), not to "vanity-metric addiction" with its dashboard clothing. The honest report is: across the product/marketing/innovation domain the diagnosis transfers as mechanism with only the number swapped; for genuinely distant settings, carry the general proxy-target-divergence-under-optimization pattern (of which Campbell's law, teaching-to-the-test, and RL reward hacking are co-instances) plus the separate social-lock-in-to-a-known-bad-measure pattern, while the addiction framing and product-analytics apparatus stay home as the domain accent. (See Structural Core vs. Domain Accent.)

Examples¶

Canonical¶

Eric Ries's The Lean Startup (2011) crystallised the vanity-versus-actionable distinction with the cohort test. Take a team that reports total registered users climbing from 100,000 in January to 500,000 in June — a five-fold rise that looks like triumph in a board deck. A cohort decomposition tells a different story: split each month's new signups into their own group and track how many convert to a paying subscription within 30 days. If January's cohort converts at 2.0% and June's converts at 2.0%, the product has not improved at all — the flattering headline grew only because more top-of-funnel traffic was poured in (2% of 400,000 new users is 8,000 conversions, exactly the same rate as before). The registered-user total moves with spend, not with product quality; the flat cohort conversion is the actionable metric the vanity number concealed.

Mapped back: Total registered users is the vanity metric, chosen for reportability; paying conversion is the outcome objective. The cohort test exposes the broken second link of the two-link causal chain — team action moves signups but not conversion — and the identical 2.0% across cohorts is the scissors signature (headline up, outcome flat). Ries's write-it-down cohort discipline is the pre-registration cure.

Applied / In Practice¶

Digital news publishers spent the 2000s–2010s optimising pageviews and unique visitors — numbers that rose with clickbait headlines and slideshow pagination and looked impressive to advertisers, but that did not predict the subscription revenue the business increasingly needed. The Financial Times responded by building an engagement score it calls RFV (Recency, Frequency, Volume): rather than counting raw pageviews, it scored each reader on how recently, how often, and how much they read, because those dimensions actually predicted subscription retention. The FT then steered its newsroom and product decisions by RFV instead of pageviews. This was a deliberate swap of a flattering, socially-entrenched proxy for a harder-to-move metric on the causal path to the outcome (paid retention) the organisation depended on.

Mapped back: Pageviews are the vanity metric, entrenched by advertiser expectation — the social-reinforcement lock-in — while subscription retention is the outcome objective. The pageview-chasing newsroom is the optimization loop producing clickbait that grows the dashboard without the outcome. Building and steering by RFV is the structural demotion and metric replacement the concept's cures prescribe: install a metric that completes the two-link causal chain.

Structural Tensions¶

T1: Reportability versus causal fidelity (the communicable number is the corruptible one). The two selection criteria the concept pries apart are not simply good and bad — each is a genuine virtue, and they pull against each other. A metric chosen for reportability is useful: it rises legibly, fits in a deck, and can be attributed to a team's action, which is exactly what a board, an investor, and a coordinating org need to align around. A metric chosen for causal fidelity is often the opposite: harder to move, harder to explain, and slower to vindicate the effort that moved it. The tension is that the properties that make a number communicable — visibility, easy movement, clean attribution — are the same properties that let it float free of the outcome, so an organisation's legitimate need for a legible headline is itself the pressure that selects vanity metrics. Diagnostic: Was this metric chosen because it sits on the causal path to the outcome, or because it is the number that communicates and coordinates most cleanly?

T2: The analytic layer versus the social layer (the fix that feels productive reverts). The concept's signature move is to split why the metric is wrong (causal disconnect) from why it survives (social-reinforcement loop), and the two layers reward opposite interventions. Analytic fixes — better instrumentation, cohort studies, cleaner attribution — are cheap, satisfying, and squarely within an analyst's control, but they touch only the first layer and revert at the next board meeting because the second is untouched. Incentive surgery — tying status and compensation to the outcome — is the binding fix but is costly, political, and outside the analyst's authority. The tension is that effort naturally flows to the tractable layer while the binding constraint sits in the intractable one, so a team can run endless honest analytics and never dislodge a metric whose stickiness was never epistemic. Diagnostic: Does the proposed remedy change what the dashboard shows, or change whose reward depends on the flattering number — and which layer is actually holding the metric in place?

T3: A real quantity versus a hollow proxy (accuracy is the metric's armour). Vanity metrics are not fake; registered users, pageviews, downloads, and permissive MAU are all real, honestly reportable quantities that genuinely measure something. That reality is exactly what protects them: an accuracy audit passes them, because they are usually accurate, and "the number is correct" is repeatedly mistaken for "the number is relevant." The tension is that the concept must separate two things a well-instrumented, truthful metric fuses — precision and causal relevance — and the metric's very truthfulness disarms the ordinary critique. A false number can be caught by verification; a true-but-disconnected number survives verification and must be defeated on the wholly separate ground of the causal chain, which is harder to argue and easier to wave away. Diagnostic: Is the objection to this metric that it is inaccurate (auditable) or that an accurate number does not sit on the path to the outcome (a different, harder charge)?

T4: Useless versus misleading (controllability is the double edge). The two-link chain severs in two ways with opposite danger profiles. When the action-to-metric link is absent, the metric is merely useless — optimising it is wheel-spinning, and the harm is wasted effort. When the action-to-metric link is intact but the metric-to-outcome link is absent, the metric is actively misleading — the team can move it, so optimising it manufactures a convincing illusion of progress while the outcome stalls or degrades. The tension is that controllability, ordinarily a virtue in a metric (a number you can move is a number you can manage to), is precisely what turns a disconnected proxy from harmless to dangerous: the easier a hollow number is to move, the more effort it will attract and the wider the scissors it will open. Diagnostic: Can the team actually move this number through its own action — and if so, is that movement traceable to the outcome, or is easy movability manufacturing false progress?

T5: Persistence-after-refutation versus not-yet-noticed (the precondition on the addiction frame). The "addiction" reading is licensed only once the causal disconnect is known to the organisation and the metric is retained anyway — persistence-after-refutation is its diagnostic tell. A team that simply has not yet run the cohort study has a detection problem, not an addiction, and the two demand opposite remedies: education and better analytics for the pre-recognition case, incentive surgery for the post-recognition one. The tension is that from the outside both look identical — a hollow metric still on the board deck — so the analyst must establish whether the refutation has landed before choosing the cure, and misreading ignorance as incentive-capture prescribes political surgery where a cohort chart would have sufficed, while misreading capture as ignorance prescribes education that reverts. Diagnostic: Has the disconnect been demonstrated and accepted, with the metric surviving anyway (addiction) — or has no one yet shown the metric is hollow (a detection problem)?

T6: Autonomy versus reduction (dashboard idiom or the instance of two parents). "Vanity-metric addiction" is a named product-analytics pattern with its own operational vocabulary — the addiction framing, the flattering-and-visible selection criterion, the investor-deck signaling dynamics — and across product, marketing, and innovation work it transfers as mechanism with only the number swapped. But its cross-domain cargo is carried by two separable parents, not one: the first layer is proxy-target divergence under optimization (the Goodhart family — Campbell's law, teaching-to-the-test, RL reward hacking as co-instances), and the second layer is social lock-in to a known-bad measure (an escalation-of-commitment shape), which is distinct from Goodhart proper because Goodhart describes a proxy drifting under pressure while the addiction layer describes retaining a proxy already known hollow. The tension is between a domain concept that fuses the two into one vivid diagnosis and the recognition that its portable structure is two parents that travel separately. Diagnostic: Resolve toward the parents (proxy-target divergence for the analytic layer, social lock-in / commitment-escalation for the addiction layer) when carrying the lesson to a non-dashboard setting; toward named vanity-metric addiction when diagnosing a product team's entrenched proxy in situ.

Structural–Framed Character¶

Vanity-metric addiction sits in the framed-leaning region of the spectrum, closer to the framed pole than most entries here because it is a diagnosis of a dysfunction rather than a neutral mechanism. On evaluative_weight it points strongly framed: "vanity-metric addiction" is a problem-word that convicts — both "vanity" and "addiction" render a verdict that a team's practice is defective — so the label carries a normative charge closer to ad_hominem's than to a value-free mechanism, even though it also names a real causal-plus-social structure. Human_practice_bound points framed decisively: the pattern is constituted by organizational reporting practice and dissolves without a team, a dashboard, and the social apparatus of boards, CEOs, performance reviews, and investor decks — the "addiction" layer is literally a fact about human status and expectation, not about any observer-free process. Institutional_origin is framed: the addiction framing, the flattering-and-visible selection criterion, and the product-analytics operational vocabulary are artifacts of a specific tradition (Ries's lean-startup vanity-versus-actionable distinction), not facts of nature. On vocab_travels it fails: MAU, pageviews, cohort studies, and investor-deck signaling are dashboard idiom that renames every component off-substrate. And import_vs_recognize points framed — the distant cousins (Campbell's law, teaching-to-the-test, RL reward hacking) are independent co-instances of the underlying parents, recognition of the shared pattern rather than imports of the named concept.

Unusually, the portable content is carried by two separable skeletons, and both are genuinely needed because the entry fuses two distinct mechanisms: (1) proxy-target divergence under optimization — the Goodhart/performativity family, of which Campbell's law and RL reward hacking are co-instances — carries the first (analytic) layer; and (2) social lock-in to a known-bad indicator — an escalation-of-commitment shape distinct from Goodhart's drift, since here the proxy is retained after being known hollow — carries the second ("addiction") layer. These two parents are what vanity-metric addiction instantiates and fuses, not what makes "vanity-metric addiction" travel: the cross-domain reach belongs to the proxy-divergence pattern and the social-lock-in pattern separately, while the addiction framing and the product-analytics apparatus stay home. Its character: a normatively charged, practice-constituted product-analytics diagnosis that vividly fuses two portable parents — proxy-target divergence and social lock-in to a known-bad measure — into one dashboard-clothed pathology, structural in those two skeletons but framed in everything that makes it "vanity-metric addiction."

Structural Core vs. Domain Accent¶

This section decides why vanity-metric addiction is a domain-specific abstraction and not a prime, and carries the case for its domain-specificity too.

What is skeletal (could lift toward a cross-domain prime). Strip away the dashboard and — unusually — two thin relational structures survive, because the entry fuses two distinct mechanisms. The first: a proxy chosen to stand in for an outcome diverges from that outcome under optimization pressure, so effort that moves the proxy fails to move the outcome. That is proxy-target divergence, the Goodhart family, portable through the catalog's performativity, signaling, and goal_congruence_alignment. The second: an indicator already known to be hollow is retained because its load-bearing function is social — status, expectation, sunk reputation — rather than epistemic. That is social lock-in, an escalation-of-commitment shape, distinct from Goodhart's drift because here the proxy is kept after being known bad. Both skeletons are genuinely substrate-portable, and both are needed — but they are the cores vanity-metric addiction shares and fuses, not what makes it vanity-metric addiction.

What is domain-bound. Everything that gives the concept its vivid bite is product-analytics furniture that does not survive extraction: the "addiction" framing; the flattering, visible, easy-to-move selection criterion; the operational vocabulary (MAU, pageviews, downloads, cohort studies, the scissors signature); the investor-deck / board / CEO signaling apparatus that supplies the lock-in; and Ries's vanity-versus-actionable codification with its pre-registration and incentive-surgery cures. The decisive test: rename the components — drop the dashboard, the board deck, the flattering headline number — and neither the causal-disconnect story nor the retention-of-a-known-bad-measure story is any longer vanity-metric addiction; each becomes one of its two looser, more general parents. The entry marks the boundary itself by noting that the genuinely distant cousins — Campbell's law, teaching-to-the-test, RL reward hacking — are not imports of this concept but independent co-instances of the shared parents, which had no product-analytics machinery to borrow.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy or co-instantiation. Vanity-metric addiction's transfer is bimodal. Within product, marketing, and innovation work — software MAU, marketing impressions, edtech minutes-of-use, public-sector pilot counts, health-tech device distribution, internal story points — the diagnosis moves intact with only the number swapped, because these all share one reporting substrate. Beyond the dashboard it does not travel as the named concept at all: the distant cousins are separately re-derived instances of the two parents, recognition of the shared patterns rather than of "vanity-metric addiction." So when the bare lesson is genuinely wanted cross-domain, it is already carried, in more general form, by the two parents the concept fuses — proxy-target divergence (performativity / signaling / goal_congruence_alignment) for the analytic layer, and social lock-in / commitment-escalation for the addiction layer. The cross-domain reach belongs to those parents; "vanity-metric addiction," as named, carries the addiction framing and the product-analytics apparatus as baggage that should stay home.

Relationships to Other Abstractions¶

Current abstraction Vanity-Metric Addiction Domain-specific

Parents (2) — more general patterns this builds on

Vanity-Metric Addiction is part of Lock-In Prime

Lock-In is the addiction layer: social and reputational switching costs keep a known-hollow metric in place after its proxy failure is understood.
Vanity-Metric Addiction is part of Proxy-Target Divergence Prime

Proxy-Target Divergence is the analytic first layer inside the two-layer vanity-metric pathology; the social persistence layer does not replace it.

Not to Be Confused With¶

A vanity metric (the number itself). A single flattering, easy-to-move, causally-disconnected quantity — registered users, pageviews, permissive MAU. That is the object; vanity-metric addiction is the two-layer pattern — selecting such a number for reportability and then socially locking into it after the disconnect is known. A team can report a vanity metric it would drop the moment a cohort study lands (no addiction) or cling to one everyone knows is hollow (addiction). Tell: is the topic a single disconnected number (vanity metric) or the organizational refusal to let it go once refuted (the addiction pattern)?
Goodhart's law. The principle that a measure ceases to be a good measure once it becomes a target, because optimization pressure makes the proxy drift from the goal it once tracked. It is the parent of the entry's first (analytic) layer, but it describes drift-under-optimization, not the entry's distinctive second layer — retaining a proxy already known to be hollow because its function is social. Tell: is the claim that pressure corrupted a once-valid proxy over time (Goodhart) or that a known-bad proxy is kept anyway because dropping it costs status (the addiction layer)?
Campbell's law and teaching-to-the-test. Social-science and education co-instances of proxy-target divergence: high-stakes indicators (crime stats, standardized test scores) get corrupted or narrowly gamed under pressure. They are independent instances of the same first-layer parent, sharing no product-analytics machinery, not applications of vanity-metric addiction. Tell: does the case involve a public-policy or classroom proxy corrupted under incentive (Campbell / teaching-to-the-test), or a product team's dashboard number entrenched by board and investor expectation (vanity-metric addiction)?
Reward hacking (reinforcement learning). An agent maximizing a specified reward signal in ways that satisfy the proxy while subverting the intended objective. Another co-instance of the proxy-target-divergence parent, in an ML rather than organizational substrate — and lacking the human social-reinforcement lock-in entirely. Tell: is a learning agent exploiting a reward specification (reward hacking) or a human organization socially retaining a discredited measure (vanity-metric addiction)?
Escalation of commitment / sunk-cost lock-in. The tendency to keep pouring resources into a failing course because of what's already invested (reputation, prior public statements, spent effort). This is the parent of the entry's second ("addiction") layer, treated more fully as its social-lock-in cargo — but on its own it is substrate-general and carries no proxy-target-divergence content. Tell: is it persistence in a discredited measure specifically because of social/reputational stakes (the addiction layer, an instance of this parent), or persistence in any failing course of action? Vanity-metric addiction is the metric-specific fusion of this with proxy divergence.
Metric gaming / deliberate manipulation. Actors consciously juking a number they know is watched (padding stats, sandbagging). Vanity-metric addiction need not involve any deception: the metric is honestly reported and genuinely accurate — its defect is causal disconnect and social retention, not manipulation. Tell: is someone deliberately falsifying or juicing a monitored number (gaming), or is an accurate number being honestly reported and clung to despite a known causal disconnect (vanity-metric addiction)?

Neighborhood in Abstraction Space¶

Vanity-Metric Addiction sits in a crowded region of the domain-specific corpus (22^nd percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Proxy Metrics & Venture Adaptation (13 abstractions)

Nearest neighbors

Build Trap — 0.87
Feature Factory — 0.86
Progress Illusion — 0.86
Surrogation — 0.86
Innovation Accounting — 0.85

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