Skip to content

Relevance Substitution

Prime #
1134
Origin domain
Philosophy Argumentation
Subdomain
informal logic and inference → Philosophy Argumentation

Core Idea

Relevance substitution is the structural pattern in which a psychologically active but epistemically irrelevant signal is supplied in place of an evidentially relevant one, and the recipient updates their disposition on the substituted signal as if it bore on the question at issue. A claim or judgement is at stake; the recipient is positioned to evaluate it; the evidence channel that should carry information about the claim's truth is replaced by a different channel that reliably moves the recipient's disposition through a non-evidential mechanism — affective response, social-pressure response, authority deference, fatigue, payoff salience, novelty attention. The substitution is the load-bearing move; the specific substituted channel is a surface taxonomy.

Three commitments fix the shape. First, an evaluative target — a claim, judgement, decision, or assessment whose truth or merit is the proper object of update. Second, a relevance gap — the supplied signal is not informative about the evaluative target through the inference channel the recipient takes themselves to be using. Third, a psychological activity asymmetry — the substituted signal moves the recipient's disposition more reliably than the relevant signal would, often because it exploits a heuristic, an affective response, or a deference disposition that operates faster and more reliably than the careful assessment the relevant signal would require. The pattern is not "people use heuristics" (too generic, and heuristics are sometimes truth-tracking) and not "irrational behaviour" (the recipient may behave entirely rationally given their cognitive economics — the failure is in the channel's relevance, not in the recipient's use of it). It is the specific claim that a substitution occurs at the inference stage, that the substituted channel is epistemically irrelevant to the target, and that the recipient's blindness to the substitution is the distinctive content the pattern names.

How would you explain it like I'm…

Candy Instead of Reasons

Imagine someone wants you to believe their toy is the best, so instead of showing you why, they just smile really big and give you candy. The candy makes you feel good and say yes — but candy has nothing to do with whether the toy is actually good. They swapped a real reason for something that just tugs at your feelings.

The Wrong Reason Trick

Relevance substitution is when someone swaps in a signal that pushes you to agree but has nothing to do with whether the thing is actually true. You're supposed to be judging a claim based on real evidence, but instead you get something that just moves your feelings or pressures you — a confident tone, a famous name, social pressure, a flashy reward. The tricky part is you update your opinion as if that signal mattered, when it doesn't. And usually you don't even notice the swap happened — that blindness is the heart of it.

Swapping In a Fake Reason

Relevance substitution is the pattern in which a psychologically active but epistemically irrelevant signal is supplied in place of a relevant one, and the recipient updates their view on the substituted signal as if it bore on the question. A claim is at stake; the recipient is positioned to evaluate it; but the evidence channel that should carry information about its truth gets replaced by a different channel that reliably moves the recipient through a non-evidential mechanism — emotion, social pressure, authority deference, fatigue, payoff salience, novelty. Three commitments fix it: an evaluative target (a claim or decision whose truth is the proper object of update); a relevance gap (the supplied signal isn't actually informative about the target through the channel the recipient thinks they're using); and a psychological-activity asymmetry (the substituted signal moves the recipient more reliably than the relevant one would). It's not just 'people use heuristics' — heuristics are sometimes truth-tracking — and not just 'irrationality'; the failure is specifically in the channel's relevance, plus the recipient's blindness to the swap.

 

Relevance substitution is the structural pattern in which a psychologically active but epistemically irrelevant signal is supplied in place of an evidentially relevant one, and the recipient updates their disposition on the substituted signal as if it bore on the question at issue. A claim or judgment is at stake; the recipient is positioned to evaluate it; the evidence channel that should carry information about the claim's truth is replaced by a different channel that reliably moves the recipient's disposition through a non-evidential mechanism — affective response, social-pressure response, authority deference, fatigue, payoff salience, novelty attention. The substitution is the load-bearing move; the specific substituted channel is a surface taxonomy. Three commitments fix the shape. An evaluative target — a claim, judgment, decision, or assessment whose truth or merit is the proper object of update. A relevance gap — the supplied signal is not informative about the target through the inference channel the recipient takes themselves to be using. And a psychological-activity asymmetry — the substituted signal moves the recipient's disposition more reliably than the relevant signal would, often by exploiting a heuristic, affective response, or deference disposition that operates faster than careful assessment. The pattern is not 'people use heuristics' (too generic, and heuristics are sometimes truth-tracking) and not 'irrational behavior' (the recipient may be rational given their cognitive economics — the failure is in the channel's relevance, not the recipient's use of it). It is the specific claim that a substitution occurs at the inference stage, that the substituted channel is epistemically irrelevant to the target, and that the recipient's blindness to the substitution is the distinctive content the pattern names.

Structural Signature

the evaluative target — the relevant inference channel — the substituted (psychologically active but irrelevant) channel — the relevance gap between substituted channel and target — the psychological-activity asymmetry favouring the substituted channel — the recipient's blindness to the substitution

A situation is relevance substitution when each of the following holds:

  • An evaluative target. There is a claim, judgement, decision, or assessment whose truth or merit is the proper object of update — the thing the recipient is positioned to evaluate.
  • A relevant inference channel. There is a channel that should carry information bearing on the target's truth or merit, and the recipient takes themselves to be reasoning through it.
  • A substituted channel. A different signal is supplied in its place — affective response, social pressure, authority deference, repetition, novelty, payoff salience, ease-of-processing — that reliably moves the recipient's disposition through a non-evidential mechanism.
  • A relevance gap. The substituted signal is not informative about the evaluative target through the inference channel the recipient believes they are using; the inferential distance between signal and target is real and unbridged.
  • A psychological-activity asymmetry. The substituted channel moves the recipient's disposition more reliably or faster than the relevant channel would, typically by exploiting a heuristic, affect, or deference disposition that operates below careful assessment.
  • A blindness invariant. The recipient does not recognise the substitution: they take themselves to be evaluating the target through the relevant channel while in fact responding to the substituted one. (Openly chosen heuristic use is not the pattern.)

The substitution is the load-bearing move and the specific substituted channel is a surface taxonomy; the blindness is the distinctive content. The signature yields a channel-indexed family of phenomena (relevance fallacies, heuristic substitution, shortcut learning, surrogate endpoints) and a structure-targeted intervention set (channel separation, substitution detection, friction, channel re-pricing, target-side audit).

What It Is Not

  • Not a heuristic. A heuristic is sometimes truth-tracking and is often used openly and rationally. This prime is the specific failure where a heuristic is used without awareness while the relevant signal is available but unaccessed; open, chosen shortcutting is not the pattern.
  • Not bias in general. Bias is a systematic deviation from a benchmark — a family of outcomes. Relevance substitution is the upstream mechanism that produces many such biases (authority bias, affect bias), one structure indexed by channel.
  • Not shortcut_learning. Shortcut learning is the machine-learning child (a model's inference running through a spurious feature); the prime is the umbrella also covering relevance fallacies, attribute substitution, and surrogate endpoints in human and organisational evaluators.
  • Not confounding. Confounding is a spurious statistical association from an unmeasured variable; relevance substitution is an evaluator updating on an irrelevant active channel at the inference stage, blind to it.
  • Not proxy_target_divergence. A cousin, but that prime names measurement-and-action substitution in an apparatus whose basis decoupled; this names inference-stage substitution in an evaluator who takes themselves to be reasoning on the merits.
  • Common misclassification. Treating a reliably-moving signal as therefore informative — mistaking the force with which a channel shifts disposition for evidence that it bears on the target. Psychological activity and epistemic relevance are orthogonal, and the pattern exploits the gap.

Broad Use

The inference-stage substitution of an irrelevant active channel for a relevant one recurs across evaluation pipelines, though it is tied to evaluation and inference. In informal logic, the pattern was named in essence by the taxonomy of relevance fallacies — threat, character of speaker, pity, repetition, perceived majority, authority deference, novelty, tradition — each the same substitution move with a different substituted channel. In cognitive psychology, heuristic substitution names a mechanism in which an easy-to-assess attribute is substituted for the hard-to-assess attribute the decision actually requires, systematically and below conscious access. In machine learning, shortcut learning substitutes spurious features (image background, scanner artefact, watermark) for the task-defining features, the model's inference running through the irrelevant channel. The pattern recurs in diagnostic medicine (surrogate-endpoint substitution of an easy-to-measure proxy for a hard-to-measure clinical endpoint), in marketing (brand-affect, celebrity-endorsement, social-proof, and urgency substitution), in political communication (likeability, partisan-tribal, fear-salience, and patriotism substitution), in legal reasoning (character-evidence rules existing precisely to prevent juror relevance substitution), in recommender systems (engagement signal substituted for user-value signal), in scientific evaluation (citation count and author-prestige substituted for research quality), and in audit and inspection (surface-feature audit substituted for substantive-compliance assessment). The cross-substrate fit is wide because the substitution mechanism is cognitive — a heuristic substitution of an easy attribute for a hard one — but operates in any evaluation pipeline (human cognitive, organisational, judicial, machine-learning, market) that processes signals about hard-to-assess targets.

Clarity

Relevance substitution clarifies by separating the evaluative target from the inference channel actually being used, and naming the substitution as a structural object with its own diagnostic signature. Without the prime, the various named fallacies and heuristic-substitution phenomena read as a catalogue of unrelated cognitive quirks — "people are biased by authority," "people are biased by emotion," "models learn spurious features." With the prime, they read as instances of one structural pattern with a shared mechanism (substitution at the inference stage) and a shared intervention family, so a sprawling catalogue collapses into one structure indexed by which channel was substituted.

The clarifying force is sharpest in identifying the recipient's blindness as the distinctive failure mode. A recipient who knows they are using authority deference rather than argument-on-merits, and chooses to do so for tractability, is not in the pattern — they are using a heuristic openly. The pattern specifically names cases where the recipient takes themselves to be evaluating the target through the relevant channel but is in fact responding to the substituted one; the blindness is structural and the substitution is below conscious detection. The prime also distinguishes itself carefully from neighbours. It is not the generic concept of a heuristic, since some heuristics are truth-tracking and not instances of the pattern; it is the specific failure mode where a heuristic is used without awareness while the relevant signal is available but unaccessed. It is not bias in general, a systematic deviation from a benchmark; it is the upstream mechanism that produces a family of biases. It is the umbrella over shortcut learning (the ML instance) and surrogate-endpoint substitution (the clinical instance). It is distinct from statistical confounding (a spurious association from an unmeasured variable) and from signalling (conveying private information through costly action, where relevance substitution is receiving a signal that does not carry the information it appears to). And it is a cousin of proxy-target divergence — both involve substitution at a critical step, but relevance substitution names inference-stage substitution in evaluators while proxy-target divergence names measurement-and-action substitution in apparatus. Holding these apart keeps the prime from being mistaken for an open heuristic, a bias, a single instance, or an apparatus-level pattern.

Manages Complexity

The prime compresses a large family of named phenomena — relevance fallacies, heuristic substitution, shortcut learning, surrogate-endpoint failure, marketing affect-channels, political symbolism, jury character-evidence prejudice, citation-count surrogation, audit-society failures — under a single three-piece structural diagnosis: evaluative target, relevance gap, psychological-activity asymmetry. Each named phenomenon corresponds to a different substitution channel, and the umbrella organises them, replacing a catalogue of disconnected quirks with one structure indexed by channel.

The compression is operational because the intervention catalogue compresses in the same way and each family targets a piece of the structure. Channel separation forces the relevant signal into the inference channel and the substituted signal out — judicial admissibility rules, blind peer review, regulatory limits on advertising claims. Substitution detection runs explicit checks for the relevance gap — cognitive-bias training, trial registration to detect surrogate-endpoint substitution, model interpretability to surface shortcut learning. Friction slows the inference long enough for the substituted channel to lose its first-mover advantage — waiting periods, deliberation protocols, explicit reasoning steps. Channel re-pricing makes the relevant signal cheaper to process or the substituted signal costlier — decision aids, structured argument templates. And target-side audit independently assesses the evaluative target through a channel that does not pass through the substituted signal — mortality auditing alongside surrogate-endpoint reporting, replication studies, user-research audits alongside engagement metrics. Because each intervention attaches to a piece of the structure — the channel, the gap, the inference's speed, the relative cost, the target itself — the prime turns the open-ended problem of "why does this evaluation go wrong" into a bounded set of structure-targeted moves.

Abstract Reasoning

Relevance substitution trains a reasoner to interrogate any evaluation pipeline through the three-piece structure. The reasoner asks: what is the evaluative target, what is the relevant inference channel, what substitution channels could be exploited, is the recipient blind to the substitution, and what intervention family fits? Because these questions reference only the abstract roles — evaluative target, relevance gap, activity asymmetry — they apply to an argument, a clinical decision, an ML model, or a hiring committee without translation.

Several reusable moves follow. The substitution-channel inventory catalogues, for a given setting, the channels that could supplant the relevant signal — affect, authority, popularity, repetition, novelty, payoff, ease-of-processing — surfacing the vulnerabilities before they are exploited. The relevance-gap diagnosis tests whether the inference channel actually carries information about the target or whether a substituted channel is doing the work, the signature being that the inference moves with the substituted-channel signal even when the relevant-channel signal is held constant. The blindness assessment estimates whether the recipient is aware of using a heuristic (open use, not in the pattern) or takes themselves to be using the relevant channel (in the pattern), distinguishing substitution-aware shortcut from substitution-blind error. The channel-strength estimation characterises the psychological-activity strength of the substituted channel relative to the relevant one, calibrating expected failure rates. And the adversarial-substitution move recognises that when an active adversary shapes the substituted channel — advertisers, propagandists, reward-hackers — the intervention must outpace the adversary's learning rate, treating substitution as a game-theoretic rather than static-cognition problem. The same reasoning that tells a judge to exclude inflammatory-but-irrelevant evidence tells an ML engineer to ablate spurious features, because both are reasoning about an irrelevant channel substituting for a relevant one at the inference stage.

Knowledge Transfer

The transferable content of the prime is a three-piece structure and a five-family intervention catalogue that port identically across substrates. A practitioner who has internalised relevance substitution looks at any evaluation pipeline and asks: what is the evaluative target, what is the relevant channel, what substitution channels could be exploited, is the recipient blind, and what intervention fits? The intervention families — channel separation, substitution detection, friction, channel re-pricing, target-side audit — transfer with minor local adaptation, because each targets a piece of the structure shared by all the substrates, and the cross-substrate borrowing is concrete: a logic-textbook fallacy treatment can borrow shortcut-learning examples from ML, a clinical-trial designer can borrow heuristic-substitution findings from cognitive psychology, an ML researcher can borrow fallacy-taxonomy distinctions from informal logic.

The transfer is deep because the three-piece structure is the same object in each case, making the transfer non-metaphorical. The authority-deference fallacy makes this concrete: a recipient is presented with a contested claim and an authority endorsement, the evaluative target is the truth of the claim, and the relevant inference channel runs through the evidence that bears on it, while the authority's endorsement is, at most, evidence about the speaker's belief rather than the claim's truth unless the authority operates as an information-aggregator. The substitution occurs when the recipient updates on the endorsement as if it bore on the claim's truth through the evidence channel — without checking what the authority endorses on the basis of, whether their expertise covers the claim, or how their endorsement compares to the broader expert distribution. The relevance gap is the inferential distance between the endorsement and the truth; the activity asymmetry is the speed with which deference moves the recipient compared to the slower evidence-evaluation channel; and the structural failure is not the deference itself (often rational) but that it was done blindly, the recipient taking themselves to have evaluated the claim's truth when the authority channel was doing the work. The same pattern recurs in heuristic substitution (a complex statistical question answered as "does this make me feel comfortable?"), shortcut learning (a model "taking itself" to detect pneumonia while its inference runs through a scanner artefact), surrogate-endpoint substitution (a physician updating on LDL reduction as if it were evidence about cardiovascular mortality), and engagement-substitution (a designer taking themselves to optimise user value while the engagement metric captures something different). Because the three-piece diagnostic and the five-family catalogue are substrate-neutral, a practitioner who has detected and intervened on substitution in one domain can do so in another on first contact, and the strip-the-jargon form ("people and systems sometimes update their assessment of one thing based on a signal that doesn't actually bear on it but reliably moves them, without realising they are doing so") does load-bearing work across informal logic, cognitive psychology, machine learning, clinical medicine, marketing, political communication, judicial procedure, peer review, platform design, audit, and education.

Examples

Formal/abstract

Attribute substitution in the psychology of judgment is the pattern at its most precisely characterised, because the mechanism is specifiable as an operation on the question itself. The evaluative target is a hard question — "what is the probability that this described person is a librarian rather than a farmer?" — whose correct evaluation runs through a relevant inference channel (base rates for the two occupations, combined with diagnostic evidence). The substituted channel is an easier question that the mind answers in its place — "how similar is this description to my stereotype of a librarian?" — i.e., representativeness. The relevance gap is exact and demonstrable: similarity-to-stereotype carries no information about the base-rate-weighted probability the target requires, so the substituted signal is epistemically irrelevant to the target through the channel the reasoner believes they are using. The psychological-activity asymmetry is the engine — the similarity judgment is fast, automatic, and below conscious access, while the base-rate computation is slow and effortful, so the easy attribute reliably wins the race. The blindness invariant is the distinctive content and is experimentally isolable: subjects report having assessed probability (they take themselves to be using the relevant channel) while their answers track similarity and ignore base rates, the signature being that judgments move with the substituted-channel signal (stereotype match) even when the relevant-channel signal (base rate) is varied. This is what separates the pattern from open heuristic use: a statistician who knowingly uses similarity as a tractable shortcut, aware of the substitution, is not in the pattern. The interventions follow the structure — friction (force an explicit base-rate step), channel separation (present the base rate before the description), substitution detection (cognitive-bias training that names representativeness).

Mapped back: The probability question is the evaluative target, base-rate reasoning the relevant channel, representativeness the substituted channel, the stereotype-versus-base-rate gap the relevance gap, the automaticity of similarity the activity asymmetry, and the subjects' false belief that they assessed probability the blindness invariant — relevance substitution rendered as experimentally isolable attribute substitution.

Applied/industry

Shortcut learning in machine vision and engagement substitution in recommender systems run the identical three-piece structure in unrelated technical substrates. A pneumonia classifier's evaluative target is the presence of pneumonia in a chest X-ray; the relevant inference channel is the lung pathology in the image. During training the model discovers a substituted channel: a metal token or scanner artefact that hospitals with sicker populations happen to stamp on their films correlates with the label, so the model's inference runs through the artefact. The relevance gap is that the token carries no pathological information about this patient's lungs; the activity asymmetry is that the artefact is an easier, lower-loss feature to fit than diffuse lung texture; and the blindness invariant is mechanical and exact — the model (and the team reading only accuracy) "takes itself" to detect pneumonia while inference runs through the artefact, exposed only when deployment to a hospital using different equipment collapses performance. The signature is the prime's: the prediction moves with the substituted-channel signal (artefact present) even when the relevant signal (lung pathology) is held constant. A recommender system instantiates the same structure at the objective level: the target is user value, the relevant channel is whether the user is actually served, and engagement (clicks, watch-time) is the substituted channel that reliably moves the optimiser while being epistemically distinct from value — the designer "takes themselves" to optimise value while the metric captures something else, blind to the substitution. The intervention families port directly across both: substitution detection (interpretability tools that surface which feature drives the prediction; held-out user-research audits), target-side audit (test on data where the artefact is decorrelated from the label; measure user-reported value alongside engagement), channel re-pricing (penalise reliance on spurious features; multi-objective reward), and, because reward-hacking adversaries adapt, update the apparatus faster than the substituted channel can be gamed.

Mapped back: Pneumonia presence and user value are evaluative targets; lung pathology and genuine service are the relevant channels; the scanner artefact and the engagement metric are substituted channels with real relevance gaps; the model's and the designer's blindness to which channel drives the output is the blindness invariant — the same prime in machine learning and platform design.

Structural Tensions

T1 — Blind Substitution versus Open Heuristic Use (sign/direction). The distinctive content is blindness: the pattern holds only when the recipient takes themselves to be using the relevant channel while in fact responding to the substituted one. The characteristic failure is misclassifying — condemning a knowingly-chosen tractable heuristic as a fallacy, or excusing a blind substitution as legitimate shortcutting. The diagnostic is to assess awareness: a statistician who knowingly uses similarity as a shortcut, aware of the substitution, is not in the pattern, while a subject who reports assessing probability but tracks stereotype is. Only substitution-blind error is the prime; substitution-aware shortcut is something the recipient can audit and the prime does not target.

T2 — Relevant Channel versus Substituted Channel (measurement). The core test is whether the inference actually runs through the channel the recipient believes, or through a substituted one — and the signature is exact: the judgment moves with the substituted-channel signal even when the relevant-channel signal is held constant. The failure is inferring channel from self-report, trusting that an evaluator reasoning "on the merits" actually is. The diagnostic is the held-constant test: vary the substituted signal while fixing the relevant one (swap the scanner artefact while holding pathology, vary stereotype while holding base rate); if the output moves, the substituted channel is doing the work regardless of what the recipient reports using.

T3 — Psychological Activity versus Epistemic Relevance (scopal). The substituted channel wins because it is psychologically more active — faster, automatic, below conscious access — while being epistemically irrelevant to the target; activity and relevance are orthogonal and the pattern exploits the gap. The failure is treating a reliably-moving signal as therefore informative, mistaking the force with which a channel shifts disposition for evidence that it bears on the truth. The diagnostic is to separate "how reliably does this move me?" from "does this bear on the target?": a channel can be maximally activating and zero-relevant, and the strength of the pull is not evidence of relevance but, under this pattern, often a warning of its absence.

T4 — Substitution Mechanism versus Surface Channel Taxonomy (scopal). The load-bearing move is substitution at the inference stage; the specific substituted channel (affect, authority, repetition, novelty, payoff) is a surface taxonomy. The failure is cataloguing channels as unrelated quirks — "biased by authority," "biased by emotion," "learns spurious features" — and missing that one mechanism underlies them, so each is fought ad hoc. The diagnostic is to look past which channel was substituted to the substitution itself: the fallacy taxonomy, heuristic substitution, and shortcut learning are one structure indexed by channel, and an intervention aimed at the mechanism (channel separation, target-side audit) generalises where a channel-specific patch does not.

T5 — Inference-Stage Substitution versus Apparatus-Stage Divergence (scopal). Relevance substitution is a cousin of proxy-target divergence: both involve substitution at a critical step, but this prime names inference-stage substitution in an evaluator (blind to it) while proxy-target divergence names measurement-and-action substitution in an apparatus whose basis decoupled. The failure is applying the wrong fix — treating an evaluator's blind heuristic with apparatus-level basis-monitoring, or a decoupled measurement instrument with cognitive-bias training. The diagnostic is to locate the substitution: is a reasoner updating on an irrelevant active channel (this prime), or is a calibrated apparatus running on a proxy whose relationship to the target broke (proxy-target divergence)? The locus selects the intervention.

T6 — Static Cognition versus Adversarial Shaping (temporal). Substitution can be a static cognitive slip or an adversarially-shaped attack — advertisers, propagandists, and reward-hackers actively cultivate the substituted channel, so the relevance gap is being widened on purpose and faster than static defences adapt. The failure is treating an adversarial substitution as a fixed bias, training evaluators once against a channel an adversary continuously re-engineers. The diagnostic is to ask whether an active agent benefits from the substitution: where one does, the countermeasure must outpace the adversary's learning rate (refresh interpretability probes, rotate audits), treating substitution as a game rather than a one-time cognitive correction.

Structural–Framed Character

Relevance substitution sits just onto the framed side of the structural–framed spectrum — the balanced-hybrid case where its framed label and aggregate of 0.5 reflect every diagnostic reading exactly mid. There is a clean structural core: an evaluative target, a relevant inference channel, a substituted-but-irrelevant channel, a relevance gap, a psychological-activity asymmetry, and the recipient's blindness to the swap. But the prime is stated through irreducibly epistemic notions of relevance and evidence, and that pulls each criterion halfway toward framed.

The decisive load comes from the epistemic vocabulary itself. "Relevant," "evidentially," and "epistemically irrelevant" are normative-epistemic terms: whether a channel is relevant to a target is a judgement about the proper inference relation, not a bare physical fact, so the prime imports an evaluative standard about what should bear on the question (evaluative_weight 0.5). The structural skeleton is real and partially substrate-independent — shortcut learning in ML, surrogate endpoints in medicine, fallacies in informal logic all instantiate it — which keeps it from sliding fully to framed (vocab_travels 0.5). Its origin is in informal logic and argumentation rather than a single named institution, sitting between formal and human-practice origins (institutional_origin 0.5). It is partly human-practice-bound: an ML pipeline exploiting a spurious feature instantiates the substitution without a person, but the home cases presuppose a recipient who takes themselves to be using one channel while updating on another (human_practice_bound 0.5). And invoking it imports an interpretive frame — judging that a psychologically active channel is epistemically irrelevant to the target and that the recipient is blind to it — while still recognising a genuine channel-substitution that is really present in the inference pipeline (import_vs_recognize 0.5). Because every criterion lands at the midpoint, the prime is a true hybrid that the rubric places just on the framed side of center: a real substitution structure wrapped in the normative vocabulary of relevance and evidence.

Substrate Independence

Relevance substitution is a moderately substrate-independent prime — composite 3 / 5 on the substrate-independence scale. Its domain breadth is real but bounded to evaluation and inference pipelines: the inference-stage substitution of an irrelevant active channel for a relevant one recurs across informal logic (the taxonomy of relevance fallacies — authority, pity, popularity, novelty, tradition), cognitive psychology (heuristic substitution), machine learning (shortcut learning on backgrounds, watermarks, scanner artefacts), diagnostic medicine (surrogate-endpoint substitution), marketing, political communication, legal reasoning (character-evidence rules exist to prevent it), recommender systems, scientific evaluation (citation count for quality), and audit. The breadth is wide within the class of systems that process signals about hard-to-assess targets but does not reach into physical or biological substrates that do not evaluate. Its structural abstraction is mid because the prime turns on an irreducibly normative-epistemic judgement — whether a psychologically active channel is relevant to the target is a claim about the proper inference relation, not a bare physical fact — so the signature imports an evaluative standard about what should bear on the question and presupposes, in its home cases, a recipient who takes themselves to use one channel while updating on another. Its transfer evidence earns a 4: the same substitution mechanism — an easy attribute swapped for the hard one the decision requires, below awareness — is documented across the substrates with concrete named instances (heuristic substitution, shortcut learning, surrogate endpoints, fallacy taxonomies), and an ML pipeline exploiting a spurious feature instantiates the very move a juror does under a character-evidence appeal. The composite sits at 3 because the evaluation-pipeline ceiling and the normative-relevance framing pin breadth and abstraction while cross-pipeline transfer is concretely evidenced.

  • Composite substrate independence — 3 / 5
  • Domain breadth — 3 / 5
  • Structural abstraction — 3 / 5
  • Transfer evidence — 4 / 5

Relationships to Other Primes

One-hop neighborhood: parents above, mutual partners to the right, children below.RelevanceSubstitutioncomposition: BiasBias

Parents (1) — more general patterns this builds on

  • Relevance Substitution presupposes, typical Bias

    The file: this prime is the upstream MECHANISM that produces many biases (authority bias, affect bias) — one structure indexed by channel. Tentative mechanism-to-outcome: bias is the family of outcomes, this is the producing mechanism; recorded as a soft presupposes-adjacency, owner may prefer no parent.

Path to root: Relevance Substitution → Bias

Neighborhood in Abstraction Space

Relevance Substitution sits in a sparse region of abstraction space (72nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Logical Moves & Precondition Gating (10 primes)

Nearest neighbors

Computed from structural-signature embeddings · 2026-06-14

Not to Be Confused With

The first confusion to clear is with the generic heuristic, because relevance substitution operates through heuristics and is easily collapsed into them. A heuristic is a fast, frugal rule for reaching a judgement, and crucially many heuristics are truth-tracking (recognition often correlates with the right answer) and many are used openly — the reasoner knows they are taking a shortcut and chooses it for tractability. Relevance substitution is the specific failure mode where a heuristic is used blindly: the reasoner takes themselves to be evaluating the target through the relevant channel while in fact responding to an epistemically irrelevant but psychologically active one, with the relevant signal available but unaccessed. The discriminating content is the blindness. A statistician who knowingly uses similarity-to-stereotype as a tractable shortcut, aware of the substitution, is not in the pattern; a subject who reports assessing probability while their answers track stereotype and ignore base rates is. Treating the prime as "people use heuristics" both over-generalises (sweeping in truth-tracking and openly-chosen heuristics that are fine) and misses the point (the failure is the unawareness, not the shortcut), and it loses the intervention that follows — surface the relevant channel so the substitution becomes visible.

A second genuine confusion is with bias as such. Bias names a systematic deviation from a benchmark — a family of outcomes, observed at the level of results: authority bias, affect bias, anchoring. Relevance substitution is the upstream mechanism that produces many of these biases: one structural move (substituting an irrelevant active channel for the relevant one at the inference stage) instantiated with different substituted channels. The relationship is mechanism-to-outcome. "Biased by authority," "biased by emotion," and "learns spurious features" look like a catalogue of unrelated quirks, but they are one substitution indexed by which channel was substituted. Reading the prime as just another bias keeps the phenomena disconnected and prescribes channel-specific patches, whereas recognising the shared mechanism licenses interventions aimed at the substitution itself (channel separation, target-side audit) that generalise across the whole family.

A third confusion, sharpest at the conceptual edge, is with proxy_target_divergence, a genuine cousin. Both involve substitution at a critical step, and both can present as a stand-in doing the work the real thing should. The distinction is locus. Relevance substitution is inference-stage substitution in an evaluator who is blind to it — a reasoner updating on an irrelevant channel while believing they reason on the merits. Proxy-target divergence is measurement-and-action substitution in a calibrated apparatus whose proxy-target basis decoupled — a metric, hedge, or control loop running on a proxy that no longer tracks the target. The loci differ — a reasoner's cognition versus an apparatus's basis — and the fixes differ accordingly: relevance substitution is addressed by channel separation and cognitive-bias detection in the evaluator, proxy-target divergence by basis-monitoring and target-side audit on the apparatus. Conflating them treats an evaluator's blind heuristic with apparatus-level basis-monitoring, or a decoupled instrument with cognitive-bias training — in each case aiming the remedy at the wrong locus. The discriminating question is whether a reasoner is updating on an irrelevant active channel (this prime) or a calibrated apparatus is running on a proxy whose relationship to the target broke (proxy-target divergence).

These distinctions matter because each mis-framing prescribes the wrong fix: a heuristic framing excuses the blind substitution as legitimate shortcutting, a bias framing patches each channel ad hoc, and a proxy-divergence framing monitors an apparatus basis where the defect is a reasoner's cognition — whereas the prime's diagnostic (name the target, the relevant channel, and the substituted channel; test whether the judgment moves with the substituted signal while the relevant one is held constant) routes to channel separation, friction, and target-side audit aimed at the inference-stage substitution itself.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.