Cherry Picking¶

Selectively presenting confirming evidence while suppressing disconfirming evidence from the same available population, so the offered sample gives an impression the full distribution would not support — locating the dishonesty in the selection process, not the individual data points.

Core Idea¶

Cherry picking — also "the fallacy of incomplete evidence" or "suppressed evidence" — is the argumentative practice of selectively presenting confirming evidence while suppressing disconfirming evidence from the same available population, such that the offered sample gives an audience an impression that the full distribution of evidence would not support. The structural commitment is a specific locus of dishonesty: not at the level of individual data points, each of which may be accurate, but at the level of the selection process that chose which data points to display. The arguer samples from a distribution of available evidence in a way correlated with their preferred conclusion, then presents the sample as if it were representative. The fallacy is therefore undetectable by local fact-checking of individual claims; it becomes visible only when the audience can compare the offered sample against the population from which it was drawn or ask what the complementary subset would show. In scientific communication the mechanism appears as selective outcome reporting: a clinical trial pre-registers five outcomes, measures all five, and reports only the two that reached significance — each reported finding is real, but the inferential claim is unwarranted because the reporting decision was correlated with the result. Structural remedies follow from the diagnosis: pre-registration of outcomes (eliminates selective reporting), systematic review (fixes selective citation), rules of evidence requiring disclosure of contrary precedent (fixes selective legal argument), and adversarial presentation formats (require engagement with counter-evidence).

Structural Signature¶

Sig role-phrases:

the available evidence population — the full body of evidence relevant to a claim, only some of which is shown
the deliberate presenter — a selector who chooses which evidence to display (the cargo that makes this cherry picking rather than mechanical selection)
the conclusion-correlated selection rule — the displayed subset drawn in a way correlated with the arguer's preferred conclusion
the locus of dishonesty — the fault sits in the selection process, not the individual data points, each of which may be accurate
the rule-blind audience — receivers who see the subset without knowing the selection rule, and so misread it as representative
the unwarranted impression — an inference from the shown subset that the full distribution would not support, undetectable by local fact-checking
the complementary-subset probe — the diagnostic counterfactual: what would the un-shown evidence show? — which reverses or weakens the inference when the bias is real
the selection-attacking remedies — structural fixes aimed at the rule, not the claims: pre-registration, systematic review, disclosure-of-contrary-precedent rules, adversarial formats

What It Is Not¶

Not a collection of false claims. The locus of dishonesty is the selection process, not the individual data points — each offered datum may be perfectly accurate. The fault is that the displayed subset was drawn from an available population in a way correlated with the conclusion, so the impression is unwarranted even when every shown item is true.
Not catchable by fact-checking individual claims. Local truthfulness is the wrong instrument: verifying each datum cannot detect the error, because the manipulation lives in what was omitted. The diagnostic question is not "is this claim true?" but "was the displayed subset drawn correlated with the conclusion?" — answered by the counterfactual probe, what would the complementary subset show?
Not confirmation bias. Confirmation bias is the cognitive tendency to seek and weight confirming evidence — internal, often unconscious. Cherry picking is its argumentative externalization: a deliberate presentation choice about what to display. One is a disposition; the other is a move made in an argument.
Not passive or mechanical selection bias. Selection bias can arise without any presenter — a survey medium that excludes a demographic, an instrument with a detection threshold. Cherry picking's defining cargo is a deliberate presenter making a display choice; calling a mechanical sampling artifact "cherry picking" smuggles in an intent that is not there. Where there is no selector, the right name is selection bias.
Not the Texas sharpshooter fallacy. The sharpshooter fallacy adds post-hoc target-drawing — framing a conclusion around a cluster of hits after the fact. Pure cherry picking is selective presentation of pre-existing evidence, without the after-the-fact target. Likewise, p-hacking and the garden of forking paths are the same family at analysis time rather than presentation time.
Not applicable where there is no presenter. The fault is locatable only where a selector exists who displays a subset — a planet's orbit cannot cherry-pick. So the concept does not reach into natural or formal substrates; there the portable structure is the selection_bias parent, and importing the rhetorical term mislabels a process with no presenter.

Scope of Application¶

Cherry picking lives across reasoning, argumentation, and the fields that present evidence to persuade — wherever a deliberate presenter samples an available evidence population in a conclusion-correlated way and displays the subset as representative — recurring with the same complementary-subset probe and the same selection-attacking remedies; its reach is bounded to settings with a selector who displays (substrates with no presenter — a survey medium, a measurement threshold — are selection_bias, not cherry picking, and a planet's orbit cannot cherry-pick).

Reasoning and argumentation — the named fallacy in informal logic and critical-thinking pedagogy ("incomplete evidence," "suppressed evidence").
Scientific communication and meta-research — selective outcome reporting in clinical trials and the popular framing of publication bias, foreclosed structurally by pre-registration and systematic review.
Journalism and political rhetoric — citing supporting anecdotes, statistics, or expert quotes while omitting countervailing ones to manufacture a one-sided impression.
Marketing claims — "9 out of 10 dentists" subgroup framings and selectively quoted reviews, displaying a favorable slice of an available population.
Legal argumentation — hostile selective use of precedent or expert testimony, against which rules of evidence force disclosure of contrary precedent.

Clarity¶

Naming cherry picking relocates the locus of dishonesty from the individual claim to the selection process, and that relocation is what makes the error catchable at all. The intuitive picture of a misleading argument is one containing false statements, so the natural defense is to verify each datum — and cherry picking defeats exactly that defense, because every offered point can be accurate while the impression is still unwarranted. The concept names the gap between what an arguer presents and what was available to present, and in doing so tells the evaluator that local truthfulness checks are the wrong instrument: the question is not "is this claim true?" but "was the displayed subset drawn in a way correlated with the conclusion?" That single reframing converts an undetectable manipulation into a diagnosable one, because it directs attention to the population the sample came from rather than to the sample itself.

The sharper question the concept licenses is a counterfactual one: what would the complementary subset show? — and the corresponding move, to ask of any evidence base what its selection rule was. This sharpens several distinctions the surface blurs. It separates cherry picking from confirmation bias (the cognitive tendency to seek confirming evidence, versus its argumentative externalization in a presentation choice), from passive selection bias (a mechanical, often unintentional sampling artifact, versus a deliberate display decision), and from the Texas sharpshooter fallacy (which adds post-hoc target-drawing absent from pure selective presentation). Most usefully, it makes the remedies legible as attacks on the selection process rather than on the claims: pre-registration forecloses selective outcome reporting, systematic review forecloses selective citation, rules of evidence force disclosure of contrary precedent, and adversarial formats compel engagement with the complementary subset — each a structural fix aimed precisely where the concept locates the fault.

Manages Complexity¶

An evaluator of arguments otherwise faces a long, miscellaneous list of misleading moves that each seem to demand their own scrutiny: a pundit citing three favorable cities and omitting four unfavorable ones, a "9 out of 10 dentists" subgroup framing, a trial reporting only its two significant outcomes out of five, a brief quoting only the precedents that help its side, a review article citing only confirming studies, a marketer quoting only five-star notices. Treated case by case, each invites the wrong instrument — verifying the individual claims — and each defeats it, since every offered datum can be accurate while the impression is unwarranted. Cherry picking compresses the whole list to one regularity by relocating the fault from the data points to the selection process: the offered subset was drawn from an available population in a way correlated with the arguer's conclusion. That single relocation reduces an open-ended audit of claims to one tracked relationship — the correlation between the selection rule and the favored conclusion — and from it the qualitative verdict reads off directly. The evaluator stops asking "is this claim true?" (which cannot catch the error) and asks "was the displayed subset drawn correlated with the conclusion?" (which always can), so the diagnosis turns on a single counterfactual probe applied uniformly across every instance: what would the complementary subset show? The branch structure is correspondingly clean. Either the selection rule is independent of the conclusion (the sample is representative and the inference stands) or it is correlated (the inference is unwarranted however accurate each point), and the same probe resolves the branch in rhetoric, science, law, journalism, and marketing alike. The relocation also organizes the remedies, which would otherwise be a scattered toolkit, into one family — structural attacks on the selection process rather than on the claims: pre-registration forecloses selective outcome reporting, systematic review forecloses selective citation, rules of evidence force disclosure of contrary precedent, adversarial formats compel engagement with the complementary subset. Each fixes the same locus the concept identifies. And the same frame draws the boundaries that keep the diagnosis precise — separating the deliberate presentation choice from confirmation bias (its cognitive precursor), from passive selection bias (a mechanical sampling artifact), and from the Texas sharpshooter fallacy (which adds post-hoc target-drawing) — so the evaluator knows which probe and which remedy apply. So instead of holding a catalogue of distinct rhetorical tricks in mind and fact-checking each, the analyst tracks one quantity, the selection rule's correlation with the conclusion, runs one complementary-subset probe, and reads off both the verdict and the structural fix — the move from a high-dimensional list of misleading-argument cases to a single selection-process diagnostic with a binary branch and a matched remedy family.

Abstract Reasoning¶

Cherry picking licenses inferences that all relocate scrutiny from the individual claim to the selection process that chose which claims to display. Diagnostic: confronting an argument whose every datum survives fact-checking yet leaves a one-sided impression, the analyst infers that the fault lies not in the points but in the rule that selected them — that the displayed subset was drawn from an available population in a way correlated with the arguer's conclusion. So a presentation in which all the cited evidence happens to favor one side is read as a candidate selection bias to be tested, and local truthfulness is treated as the wrong instrument: passing a fact-check is uninformative about the error, not exonerating. The signature diagnostic probe is a counterfactual — what would the complementary subset show? — applied uniformly: ask what the un-shown cases (the omitted cities, the unreported trial outcomes, the contrary precedents) would do to the inference, and if surfacing them would reverse or weaken it, the original presentation is diagnosed as cherry-picked however accurate each shown item. Interventionist: because the fault is the selection rule, the remedies attack the selection rule rather than the claims, and each is predicted to foreclose one channel of suppression — pre-registration forecloses selective outcome reporting (all measured outcomes must be reported), systematic review forecloses selective citation (the full study population must be surveyed), rules of evidence force disclosure of contrary precedent, and adversarial presentation formats compel engagement with the complementary subset. Choosing among them is a matter of which selection channel is open in the setting, and an intervention aimed at the truthfulness of individual claims is predicted to fail because it leaves the selection process untouched. Boundary-drawing: the concept resolves to a clean binary the analyst applies to any evidence base — either the selection rule is independent of the conclusion (the sample is representative and the inference stands) or it is correlated (the inference is unwarranted regardless of each point's accuracy) — and it marks the borders that keep the diagnosis precise: separating the deliberate presentation choice from confirmation bias (its cognitive precursor, internal rather than externalized in a display), from passive selection bias (a mechanical, often unintentional sampling artifact with no presenter), and from the Texas sharpshooter fallacy (which adds post-hoc target-drawing absent from pure selective presentation), so the analyst reaches for the matching probe and remedy. It also bounds the concept out of substrates with no presenter — a planet's orbit cannot cherry-pick — which is part of why its fault is locatable only where a selector exists. Order-of-events reasoning fixes when the bias enters: the population of relevant evidence exists, a selection step draws a subset, and the audience receives the subset without the selection rule — so the analyst traces the unwarranted impression back to the selection step rather than to the data or the audience, and reconstructs the missing rule to recover what was suppressed.

Knowledge Transfer¶

Within reasoning, argumentation, and the fields that present evidence to persuade, cherry picking transfers as mechanism, because every instance is the same relocation of the fault from the data points to the selection process that chose which to display, and the same single probe resolves it. It carries intact across informal-logic pedagogy (the named fallacy), scientific communication and meta-research (selective outcome reporting, the popular framing of publication bias), journalism and political rhetoric (supporting anecdotes and statistics cited while countervailing ones are omitted), marketing ("9 out of 10 dentists" subgroup framings, selectively quoted reviews), and legal argumentation (hostile selective use of precedent or testimony). Across all of them the diagnostic is identical: stop fact-checking individual claims (the wrong instrument, since each shown datum can be accurate) and ask whether the displayed subset was drawn correlated with the conclusion, then run the one counterfactual probe — what would the complementary subset show? And the remedies form one family because they all attack the selection rule rather than the claims: pre-registration forecloses selective outcome reporting, systematic review forecloses selective citation, rules of evidence force disclosure of contrary precedent, and adversarial formats compel engagement with the complementary subset; which to reach for depends only on which selection channel is open in the setting. The vocabulary travels because the locus of the fault — a presenter sampling an available population in a conclusion-correlated way — is constant beneath the changing arena.

Beyond evidence-presentation contexts the honest report is that what genuinely recurs across substrates is not "cherry picking" but the broader mechanism it specializes — biased sampling that distorts an inference from a subset to a population — which is the prime selection_bias. That parent recurs everywhere data are drawn from a population, including in places with no arguer at all: a survey medium that excludes a demographic, a measurement apparatus that registers only surviving cases, an instrument with a detection threshold. Those are selection bias as co-instances, but they are not cherry picking, because cherry picking's defining cargo is exactly what they lack — a deliberate presenter making a display choice. The concept is locatable only where a selector exists; a planet's orbit cannot cherry-pick. So when the cross-domain lesson is wanted in a substrate without a presenter, it should be carried by selection_bias (and, for the cognitive precursor — the internal tendency to seek and weight confirming evidence rather than its externalization in a presentation — by confirmation_bias), not by the rhetorical concept. Invoking "cherry picking" for a mechanical sampling artifact renames a passive process as a deliberate one and smuggles in an intent that is not there; that is the over-application to watch, and the correct move is to name it selection bias instead.

The boundary against neighbors keeps the cut precise and is itself part of the transfer discipline: cherry picking is the deliberate-presentation specialization of selection bias (versus the mechanical, often unintentional artifact); it is the argumentative externalization of confirmation bias (versus the cognitive tendency); survivorship bias is a specific selection mechanism rather than this general selective-presentation pattern; p-hacking and the garden of forking paths are the same family at analysis time rather than presentation time; and the Texas sharpshooter fallacy adds post-hoc target-drawing absent from pure selective presentation. The clean summary: within argumentation and adjacent evidence-presenting fields it transfers as mechanism and remedy across rhetoric, science, law, journalism, and marketing; beyond it, the portable structure is the selection_bias parent (with confirmation_bias for the cognitive substrate), which recurs wherever a subset is drawn from a population, while the deliberate-presenter, rhetorical-honesty framing that makes this cherry picking stays bound to settings with a selector who displays. See Structural Core vs. Domain Accent.

Examples¶

Canonical¶

The cleanest worked instance is selective outcome reporting in a clinical trial. Suppose a trial pre-specifies five outcome measures, collects data on all five, but the published paper reports only the two that reached statistical significance and quietly omits the three that did not. Every number in the paper is accurate; nothing has been fabricated. Yet the inference the reader draws — "the intervention works" — is unwarranted, because the decision about what to report was correlated with the results. The COMPare project (led by Ben Goldacre's team, 2015–2016) audited trials against their registered protocols and found that reported outcomes were routinely added or silently dropped relative to what had been pre-specified, exactly this selection operating in the wild. The fault is invisible to fact-checking the reported figures; it surfaces only against the registered population of outcomes.

Mapped back: The five measured outcomes are the available evidence population; the authors deciding what to publish are the deliberate presenter, and reporting only significant results is the conclusion-correlated selection rule. Because each reported number is true, the locus of dishonesty sits in the selection, and the registered protocol is what enables the complementary-subset probe — what would the unreported outcomes show?

Applied / In Practice¶

Evidence-based medicine deploys structural remedies aimed precisely at this selection locus. Turner and colleagues (2008, New England Journal of Medicine) showed that among FDA-registered antidepressant trials, studies with positive results were far more likely to be published than those with negative or questionable results, inflating the drugs' apparent efficacy in the literature even though each published trial was sound. The field's response attacks the selection rule rather than the individual studies: mandatory trial pre-registration (ClinicalTrials.gov registration became required for many trials in the United States in 2007) forecloses selective outcome reporting, and Cochrane-style systematic reviews forecloses selective citation by surveying the entire population of trials — published and unpublished — rather than an author's favorable slice.

Mapped back: The full set of antidepressant trials is the available evidence population; publishing mainly the positive ones is the conclusion-correlated selection rule producing the unwarranted impression of efficacy. Pre-registration and systematic review are the selection-attacking remedies — they fix the rule, not the claims, which is where the concept locates the fault.

Structural Tensions¶

T1: All argument selects versus the fallacy of selection (where emphasis becomes cherry-picking). Every act of persuasion, and indeed every honest exposition, chooses which evidence to present — no one shows the full population, and a presenter always argues toward some conclusion, so their selection is always correlated with it to a degree. The construct locates the fault in "conclusion-correlated selection," but that description, taken literally, indicts all argument. What distinguishes legitimate emphasis (leading with the strongest relevant evidence, omitting the irrelevant) from the fallacy (suppressing disconfirming evidence the audience would want) is a matter of degree and of what a fair-minded evaluator would deem material — a line the crisp "correlated or not" binary papers over. The concept's power to catch suppression is bought against a genuine difficulty in saying when selection crosses from advocacy into dishonesty. Diagnostic: Is the omitted evidence merely less rhetorically useful, or is it material enough that a fair evaluator would need it to draw the inference — and who decides which?

T2: The complementary-subset probe versus the contestable population (defining what was available). The signature diagnostic — what would the complementary subset show? — presumes a well-defined population from which the sample was drawn, against which omissions are visible. But that reference population is itself often contestable: what counts as the "available" evidence, the "relevant" precedents, the outcomes that "should" have been reported? A presenter can gerrymander the reference class as easily as the sample, and an accuser can inflate it to manufacture a suppression that was really just scope. The probe that makes cherry-picking detectable depends on agreement about the population boundary, which is exactly what a sophisticated arguer will dispute. Pre-registration works precisely because it fixes the population in advance; where no such anchor exists, the probe's verdict rests on a contested denominator. Diagnostic: Is the population against which omission is being judged fixed independently (a protocol, a docket), or is the reference class itself being drawn to produce the verdict?

T3: The intent requirement versus its unobservability (imputing deliberateness from a pattern). The concept's defining cargo — what makes it cherry picking rather than mechanical selection bias — is a deliberate presenter making a display choice. But intent is not observable; all the evaluator sees is a one-sided sample, and the diagnostic (correlation with the conclusion) cannot by itself distinguish deliberate suppression from innocent selection, honest disagreement about relevance, or ignorance of the omitted evidence. So the very feature that separates cherry-picking from its neutral parent is the feature that cannot be verified from the evidence, and the charge routinely imputes deliberateness from a pattern that also has non-culpable explanations. The construct requires intent to exist and offers no way to observe it. Diagnostic: Does the evidence show the presenter knew of and chose to suppress the complementary subset, or is deliberateness being inferred from a one-sidedness that ignorance or honest scoping could equally produce?

T4: Structural remedies versus the cost of forced completeness (disclosing everything has a price). The remedy family attacks the selection rule: pre-register all outcomes, survey the full study population, disclose contrary precedent, engage the complementary subset. These reliably foreclose suppression, but forced completeness is not free — reporting all five outcomes can bury the two that matter in noise, mandatory disclosure raises the cost of every honest communication, systematic review is slow and expensive, and adversarial formats assume symmetric resources that many settings lack. Pushed hard, "show everything" degrades the signal that selective emphasis exists to provide, and burdens the scrupulous more than the manipulator, who can comply in letter while misleading in structure. The cure for suppression trades against the legitimate economy of presentation. Diagnostic: Does the completeness remedy foreclose the suppression channel at issue, or does it impose disclosure costs and signal-dilution disproportionate to the manipulation it prevents?

T5: Autonomy versus reduction (a rhetorical specialization of selection bias). Cherry picking has genuine home cargo — the deliberate presenter, the rhetorical-honesty framing, the selection-attacking remedies — and transfers as mechanism across every evidence-presenting field (rhetoric, science, law, journalism, marketing). But the pattern that recurs beyond arguers is its parent selection_bias: biased sampling distorting an inference from subset to population, which appears wherever data are drawn — a survey medium excluding a demographic, an instrument with a detection threshold — with no presenter at all. Those are selection bias, not cherry picking, because they lack the deliberate display choice; a planet's orbit cannot cherry-pick. Its cognitive precursor is confirmation_bias. The tension is that the rhetorical concept is a presenter-bound specialization whose portable structure belongs to selection bias, and importing "cherry picking" into a presenter-less substrate smuggles in an intent that is not there. Diagnostic: Resolve toward selection_bias (or confirmation_bias for the cognitive tendency) wherever there is no deliberate presenter; toward "cherry picking" only where a selector displays a conclusion-correlated subset as representative.

Structural–Framed Character¶

Cherry picking sits at the framed pole of the spectrum — closely paralleling ad hominem: a normatively charged, practice-constituted fallacy whose only substrate-spanning content is carried by a parent it instantiates. On evaluative_weight it scores high: to call a presentation "cherry picking" is to convict it — the concept locates a locus of dishonesty, a charge that an inference is unwarranted and the arguer's selection deceptive, not a neutral description of a sampling process. Human_practice_bound is high in the strongest sense: the concept is constituted by the practice of argumentation and evidence-presentation and dissolves the instant that practice is removed — its defining cargo is a deliberate presenter making a display choice, so where there is no selector who displays (a survey medium, a detection threshold, a planet's orbit) there is no cherry picking at all, only the presenter-less parent. Institutional_origin points framed: it is a named fallacy of informal logic and critical-thinking pedagogy ("incomplete evidence," "suppressed evidence"), and its remedies — pre-registration, systematic review, disclosure-of-contrary-precedent rules, adversarial formats — are argumentation-and-institutional furniture. Vocab_travels is low: that deliberate-presenter, rhetorical-honesty framing is pinned to evidence-presentation, so importing "cherry picking" into a presenter-less substrate smuggles in an intent that is not there. And import_vs_recognize patterns as import beyond the home: off the argumentation substrate the operative concept is the parent, not this fallacy.

The one structural-looking feature is selection bias — biased sampling that distorts an inference from a subset to the population it was drawn from. That skeleton is genuinely portable and recurs as real co-instances wherever data are drawn (survey media excluding a demographic, instruments registering only surviving cases), which is what tempts a structural reading. But it does not pull cherry picking off the framed pole, because that portable structure is precisely what cherry picking instantiates from its parent (selection_bias, with confirmation_bias for the cognitive precursor), not what makes "cherry picking" itself travel: the cross-domain reach belongs to selection bias — which recurs with no arguer at all — while cherry picking's distinctive content, the deliberate presenter and the rhetorical-dishonesty framing, is exactly the part that stays bound to settings with a selector who displays. Its character: a normatively charged, presenter-constituted fallacy label whose every distinctive feature is argumentation-honesty furniture, structural only in the selection-bias skeleton it specializes from its parent and frames as a charge of dishonest selection.

Structural Core vs. Domain Accent¶

This section decides why cherry picking is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity — there is no separate section for that.

What is skeletal (could lift toward a cross-domain prime). Strip the argumentation and a thin relational structure survives: a subset is drawn from an available population in a way correlated with a favored conclusion, and then treated as representative, so an inference the full distribution would not support is drawn from the biased sample. The pieces that travel are abstract — a full population, a selection step whose rule is correlated with an outcome, a subset passed off as representative, and an unwarranted subset-to-population inference detectable only by asking what the complementary subset would show. That skeleton is genuinely substrate-portable, which is exactly why it is already housed in the catalog as selection_bias (biased sampling distorting a subset-to-population inference), recurring wherever data are drawn — even with no arguer at all — with confirmation_bias supplying the cognitive precursor; these are the parent primes the entry instantiates. But it is the core it shares, not what makes cherry picking distinctive.

What is domain-bound. Everything that makes it cherry picking in particular is argumentation-and-evidence-presentation furniture and none of it survives extraction intact: the deliberate presenter making a display choice (the intent-cargo that separates it from mechanical sampling); the rhetorical locus of dishonesty framing (the fault is a deceptive selection, a breach of argumentative honesty, not a neutral artifact); the named-fallacy vocabulary ("incomplete evidence," "suppressed evidence"); the selection-attacking remedy family (pre-registration, systematic review, disclosure-of-contrary-precedent rules, adversarial formats); and the boundary jurisprudence separating it from confirmation bias, survivorship bias, p-hacking/forking paths, and the Texas sharpshooter fallacy. These are the worked apparatus and empirical cases (selective outcome reporting, the COMPare audit, the Turner antidepressant-publication study) that argumentation and meta-research actually study. The decisive test: the fault is locatable only where a selector exists who displays — a survey medium that excludes a demographic, an instrument with a detection threshold, a planet's orbit cannot cherry-pick — so strip the deliberate-presenter substrate and the concept does not reach a new domain, it loses its intent-cargo and its dishonesty framing and collapses into the presenter-less parent selection_bias. Importing "cherry picking" into a presenter-less process renames a passive artifact as a deliberate one and smuggles in an intent that is not there.

Why this does not clear the prime bar. A prime is a relational structure whose vocabulary travels and whose cross-domain transfer is recognition of the same mechanism, not analogy. Cherry picking's transfer is bimodal. Within reasoning, argumentation, and the evidence-presenting fields it travels intact — informal logic, scientific communication and meta-research, journalism and political rhetoric, marketing, legal argumentation — because every instance is the same relocation of the fault to a presenter's conclusion-correlated selection, so the complementary-subset probe, the binary verdict, and the selection-attacking remedy family carry without translation; that is recognition, not analogy. Beyond evidence-presentation the mechanism genuinely recurs, but as selection_bias, not as cherry picking: survey media, detection thresholds, and survivorship filters are co-instances of the parent with no presenter, and calling them "cherry picking" mislabels them. And when the bare structural lesson is needed in a presenter-less substrate — a subset biased in its drawing distorts the inference to the whole — it is already supplied in more general form by the parent the entry instantiates: selection_bias (with confirmation_bias for the cognitive tendency). The cross-domain reach belongs to those parents; "cherry picking," as named, carries its deliberate presenter, its rhetorical-dishonesty framing, and its selection-attacking remedies as baggage that stays bound to settings with a selector who displays.

Relationships to Other Abstractions¶

Current abstraction Cherry Picking Domain-specific

Parents (1) — more general patterns this builds on

Cherry Picking is a kind of Selection Bias Prime

Cherry picking is selection bias specialized to a deliberate presenter whose conclusion-correlated evidence selection makes the displayed subset misrepresent its source population.

Not to Be Confused With¶

Confirmation bias. The internal cognitive tendency to seek, notice, and overweight evidence that fits a prior belief — often unconscious, operating in a single reasoner's own mind. Cherry picking is its argumentative externalization: a deliberate choice about what to display to an audience. Tell: is the bias a disposition shaping which evidence a person attends to (confirmation bias), or a presentation move made to persuade others (cherry picking)? The former can produce the latter, but one is a mental habit and the other an act in an argument.
Selection bias (the parent). The general, presenter-neutral pattern — biased sampling that distorts an inference from a subset to the population it was drawn from — which arises with no arguer at all: a survey medium that excludes a demographic, an instrument with a detection threshold. Cherry picking is the deliberate-presentation specialization of this parent, adding an intending selector who displays. Tell: is there a selector who chose to display a conclusion-correlated subset (cherry picking), or a mechanical sampling artifact with no presenter (selection bias)? A planet's orbit cannot cherry-pick. (Treated fully in a later section.)
Survivorship bias. A specific selection mechanism in which only the cases that passed some filter (surviving firms, returning aircraft, still-standing buildings) are visible, so the sample systematically excludes the failures. This is one particular way a subset gets drawn, typically without any deliberate presenter, whereas cherry picking is the general selective-presentation pattern driven by a selector's display choice. Tell: are the omitted cases missing because a filter destroyed or hid them (survivorship bias), or because a presenter chose not to show them (cherry picking)?
Texas sharpshooter fallacy. The error of drawing the target after seeing where the shots landed — framing a conclusion around a cluster of data points identified post hoc, treating chance patterns as significant. It adds after-the-fact hypothesis-fitting that pure cherry picking lacks: cherry picking is the selective presentation of pre-existing evidence toward a conclusion already held. Tell: was a pattern found first and a conclusion drawn around it afterward (sharpshooter), or was a conclusion held and confirming evidence selectively shown (cherry picking)?
P-hacking / the garden of forking paths. The same suppression family, but operating at analysis time rather than presentation time: running many analytic specifications, subgroups, or tests and reporting only those that reach significance. Cherry picking selects among evidence to display; p-hacking selects among analyses to run and keep. Tell: is the conclusion-correlated selection happening over which findings to present (cherry picking) or over which statistical procedures and comparisons to report (p-hacking)? Both are foreclosed by pre-registration, which is why they are easily conflated.

Neighborhood in Abstraction Space¶

Cherry Picking sits in a crowded region of the domain-specific corpus (13^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Unclustered & Miscellaneous (309 abstractions)

Nearest neighbors

Suppressed Evidence — 0.90
Null Ritual — 0.86
Two-Sided Message — 0.86
Quote Laundering — 0.86
Spotlight Fallacy — 0.86

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