Skip to content

Negative Case Analysis

Prime #
1015
Origin domain
Ethnography & Qualitative Methods
Subdomain
grounded theory → Ethnography & Qualitative Methods

Core Idea

Negative case analysis is the structural pattern in which a working account of how something works — a tentative theory, model, rule, classification, design hypothesis, or policy recommendation — is deliberately exposed to the instances that, if real, would force its revision. Rather than letting confirming evidence accumulate, the analyst actively hunts the cases that do not fit: the patient who recovered against the predicted outcome, the firm that thrived in the supposedly hostile environment, the user who succeeded with the configuration the model said would fail.

The pattern has five structural commitments. There is a current account — a model or generalisation treated as serious enough to act on. There is an implied boundary — the claims the account makes, however tacitly, about which cases fall inside its scope and which do not. There is a deliberate search for discordant cases — instances the account did not predict, predicted wrongly, or excluded as anomalous. There is a refinement decision about each discordant case: it can falsify the account, narrow its scope, modify a mechanism, or be argued out of scope with the argument made explicit. And there is iteration — the refined account is itself re-exposed to fresh discordant cases, because the process is a cycle rather than a one-shot test. The prime is simultaneously a method (a procedure for theory-building under uncertainty) and a disposition (a willingness to hunt for what could overturn your account rather than waiting for it to ambush you).

How would you explain it like I'm…

Finding The Grumpy Dog

Imagine you make up a rule like 'all dogs are friendly.' Negative Case Analysis is when, instead of only petting friendly dogs to feel right, you go looking for a grumpy dog on purpose to test your rule. If you find one, you have to fix your rule. You hunt for the example that proves you wrong, instead of waiting for it to surprise you.

Hunting What Proves You Wrong

When you have a working idea about how something works, it's tempting to only notice the examples that agree with you. Negative Case Analysis is the opposite habit: you go out and actively look for the cases that don't fit — the patient who got better when your theory said they wouldn't, or the person who succeeded with the setup you predicted would fail. When you find one, you have to make a decision: maybe your idea is wrong, maybe it just doesn't apply to that case, or maybe you tweak it. Then you do it again with new examples, because it's a loop, not a one-time check. It's both a method and a mindset — going looking for what could prove you wrong rather than waiting to be surprised.

Chasing Discordant Cases

Negative Case Analysis is a pattern where a working account — a tentative theory, model, rule, classification, hypothesis, or policy — is deliberately exposed to the very cases that, if real, would force it to change. Instead of letting confirming evidence pile up, the analyst hunts the discordant cases: the instance the account didn't predict, predicted wrongly, or dismissed as anomalous. It has five commitments: a current account serious enough to act on; an implied boundary, the claims it tacitly makes about what's in or out of scope; a deliberate search for discordant cases; a refinement decision for each one — falsify the account, narrow its scope, modify a mechanism, or rule it out of scope with the argument made explicit; and iteration, re-exposing the refined account to fresh discordant cases. It is at once a method (a procedure for building theory under uncertainty) and a disposition (a willingness to hunt for what could overturn your account rather than wait for it to ambush you).

 

Negative Case Analysis is the structural pattern in which a working account of how something works — a tentative theory, model, rule, classification, design hypothesis, or policy recommendation — is deliberately exposed to the instances that, if real, would force its revision. Rather than letting confirming evidence accumulate, the analyst actively hunts cases that do not fit: the patient who recovered against the predicted outcome, the firm that thrived in a supposedly hostile environment, the user who succeeded with the configuration the model said would fail. The pattern has five structural commitments. First, a current account — a model or generalization treated as serious enough to act on. Second, an implied boundary — the claims the account makes, however tacitly, about which cases fall inside its scope and which do not. Third, a deliberate search for discordant cases — instances the account did not predict, predicted wrongly, or excluded as anomalous. Fourth, a refinement decision about each discordant case: it can falsify the account, narrow its scope, modify a mechanism, or be argued out of scope with the argument made explicit. Fifth, iteration — the refined account is re-exposed to fresh discordant cases, because this is a cycle, not a one-shot test. The prime is simultaneously a method (a procedure for theory-building under uncertainty) and a disposition (a willingness to hunt for what could overturn your account).

Structural Signature

the current account — the implied boundary of scope — the deliberate search for discordant cases — the discordant case itself — the refinement rule (falsify, narrow, revise, or scope-out) — the iteration loop

A practice is negative case analysis when each of the following holds:

  • A current account. There is a working model, theory, rule, classification, or hypothesis treated as serious enough to act on, making predictions however tacitly.
  • An implied boundary. The account makes claims about which cases fall inside its scope and which it excludes as anomalous — a boundary that is often unstated and must be made explicit for the account's predictions to be testable.
  • A deliberate search. The load-bearing commitment: the analyst actively hunts for cases the account did not predict, predicted wrongly, or excluded — rather than letting confirming instances accumulate. The directionality (seek disconfirmation, not confirmation) is what distinguishes the prime from ordinary pattern-matching.
  • A discordant case. An instance is found that the account fails to cover — the patient who recovered against prediction, the adversarial input, the site that failed despite satisfying the preconditions, the counterexample.
  • A refinement rule. Each discordant case forces an explicit choice: falsify the account, narrow its scope, revise a mechanism, or argue it out of scope with the argument made explicit. Honest refinement is distinguished from ad hoc rescue (a content-free immunizing clause).
  • An iteration loop. The refined account is re-exposed to fresh discordant cases; the process is a cycle, not a one-shot test. Clustering of discordant cases diagnoses where the boundary is mis-drawn or whether the mechanism (not the scope) is wrong.

Composed: an account is held against the cases it is supposed to handle, deliberately stressed with the instances that would overturn it, and revised or rescoped by an explicit rule — iterated, so that discordant cases become the steering signal for the next version rather than embarrassments explained away.

What It Is Not

  • Not falsifiability. Falsifiability is the demarcation property — whether a claim could be refuted in principle; negative case analysis is the practical method of where to hunt for the refuters and what to do on finding some. A claim can be falsifiable yet never actually stressed.
  • Not comparative_method. The comparative method (the nearest embedding neighbor) contrasts cases to infer what drives an outcome; negative case analysis hunts the cases that would break a held account. Comparison builds a generalization; this prime stress-tests one already in hand.
  • Not inductive_reasoning. Induction accumulates instances to build a generalization; negative case analysis seeks disconfirming instances to refine or refute one. The directionality is opposite — induction gathers what fits, this prime gathers what does not.
  • Not abductive_reasoning. Abduction generates the best explanatory hypothesis for a surprising observation; negative case analysis tests and revises an existing account against discordant cases. One produces a candidate account; the other pressures it.
  • Not ad hoc rescue. Refining an account against a discordant case must lose content (narrow scope, revise mechanism, exclude with explicit argument); appending a content-free immunizing clause per counterexample is the degenerating opposite of the prime, not an instance of it.
  • Not chasing selection_bias artifacts. Not every off-account case is a theoretical deviant; some are measurement noise or sampling artifacts. Treating every anomaly as a refutation over-fits the account to noise. The prime requires distinguishing a genuine deviant from an outlier.
  • Common misclassification. Amassing confirming instances and calling the account robust because no counterexample is known. Catch it by asking whether disconfirmation was actively sought — an account with no known counterexamples may simply be under-searched, and its clean record reflects little search rather than genuine robustness.

Broad Use

  • Qualitative social science (origin): negative case analysis is named as a trustworthiness criterion in interpretivist research and is constitutive of deviant-case selection in grounded theory.
  • Scientific method: the practice predates its naming — Darwin's notebooks reserved space for "facts that contradict me"; Mendeleev's table was tightened by predicted-then-found elements.
  • Policy evaluation: a programme theory that explains observed successes is hardened by deliberately studying the sites or cohorts where it failed.
  • Product research and UX: systematic study of churned users, failed sales, rejected applications, and abandoned carts is negative case analysis on the implicit product theory.
  • Machine learning: adversarial testing, red-teaming, slice analysis, and out-of-distribution evaluation seek inputs that should break the model under its current account of its domain.
  • Engineering safety: FMEA, HAZOP, and pre-mortems systematically search for cases the current design would mishandle.
  • Investigation and clinical reasoning: ruling out alternative hypotheses, and generating competing diagnoses to seek findings that would not fit the leading one, are the same move at the desk and the bedside.
  • Mathematics: the deliberate hunt for a counterexample that bounds, refines, or refutes a conjecture is the formal instance of the pattern.

Clarity

Naming the pattern separates four things that informal practice routinely conflates. It separates accumulating confirming instances — the default of pattern-matching cognition — from seeking disconfirming ones, the discipline that produces robustness. It separates falsifiability — the demarcation question of whether a claim could be refuted — from negative case analysis — the practical question of where to look for the refuting evidence and what to do on finding some. It separates statistical outliers, which may be measurement noise, from deviant cases, which are theoretical problems in disguise. And it separates honest refinement — using a discordant case to narrow scope or modify mechanism — from ad hoc rescue — inventing a clause that immunises the theory against the case without loss of content. The vocabulary forces the analyst to be explicit about which of these is happening.

That explicitness is the clarifying payoff. A claim like "the programme works" becomes, under the prime, "the programme works for these cases, fails for those, and here is the redrawn boundary or revised mechanism that the failures imply." The discordant cases stop being embarrassments to be explained away and become the steering signal for the next version of the account.

Manages Complexity

The pattern compresses a large family of investigative practices — deviant-case sampling, counterexample hunting, adversarial testing, churn analysis, the pre-mortem, FMEA, ruling out alternative diagnoses — into one shared structural frame with a unified intervention vocabulary: what is the current account, which cases would pressure it, where do we find them, and what is the refinement rule when we do? A practitioner who has internalised this in one substrate can recognise it in another and apply the same discipline without re-deriving it.

It also compresses a recurring and otherwise slippery distinction. When an account produces a wrong call, was the failure a contingent mis-execution — the procedure was right and the case fell on the wrong side of the noise — or a theory failure, the procedure being wrong because the boundary lies somewhere it does not draw? The discipline of taking each discordant case seriously, rather than dismissing it, is what makes this distinction reliably detectable instead of a matter of after-the-fact rationalisation.

Abstract Reasoning

The pattern supports a family of inferences that are otherwise hard to make. Boundary inference: if discordant cases cluster in a particular sub-region — a demographic, a geography, a parameter range — the boundary of the account is mis-drawn rather than the mechanism, and the clustering tells you where to redraw it. Mechanism revision versus scope narrowing: on encountering a counterexample, the analyst can modify the mechanism or narrow the scope, and the choice is itself a diagnosis — many isolated counterexamples favour scope narrowing, while a clustered family favours mechanism revision. Robustness diagnosis: an account with no known deviant cases is not thereby strong, because it may simply not have been searched hard; the age of an account and the rate at which new deviant cases surface per unit of search are themselves signals of its robustness. And ad hoc rescue detection: an account progressively patched with a special clause for each discordant case is a degenerating research programme, and the prime supplies a vocabulary in which to recognise this from inside one's own programme rather than only from outside.

Knowledge Transfer

The pattern travels intact because it operates on the relationship between an account and the cases that account is supposed to handle, and that relationship has the same structure regardless of substrate. The roles map cleanly: the current account is the qualitative theory, the trained model, the programme logic, the working diagnosis, the conjectured bound; the discordant case is the deviant interview, the adversarial input, the failed site, the symptom the leading hypothesis cannot explain, the counterexample; the refinement rule is the explicit choice among mechanism revision, scope narrowing, and principled exclusion.

Concrete transfers are documented across very different media. A trained model is an implicit theory of its data distribution, so adversarial testing and slice analysis are negative case analysis applied to that implicit theory — seek the cases the account fails to cover, then revise or narrow it. A programme's logic model is an implicit theory of for whom and under what conditions a policy works, so the deliberate search for sites that failed despite satisfying the programme theory's preconditions is the qualitative deviant-case move applied to policy. An FMEA on a physical system has the same structure as a churn analysis on a digital product: enumerate the modes by which the current account of value can fail, then search the data for instances of each. The medical training to rule out before ruling in is the same discipline as the detective's deliberate generation of competing suspect-theories and the pursuit of discriminating evidence. And the engineer asking "under what loading regime would this fail?" is doing the analogue of the mathematician's "under what input would this conjecture be false?" — identical structural move, utterly different substrate. The portable intervention library is small and substrate-independent: make the current account explicit enough that its predictions can be tested; generate or sample cases engineered to pressure those predictions; for each discordant case choose explicitly among mechanism revision, scope narrowing, or principled exclusion; and iterate. What transfers is not a domain vocabulary but this four-step loop and the disposition that drives it.

Examples

Formal/abstract

Counterexample hunting against a mathematical conjecture is the prime's sharpest formal instance, because the refinement rules are crisp. Take the conjecture (Euler's, 1769) that no \(n\)th power is the sum of fewer than \(n\) smaller \(n\)th powers — for \(n=4\), that no fourth power equals the sum of three fourth powers. This is the current account; its implied boundary is universal (all \(n\), all such sums). The discipline of the prime is to deliberately search for a discordant case rather than accumulate confirming small examples. The hunt — eventually a computer search — finds one: \(95800^4 + 217519^4 + 414560^4 = 422481^4\), a sum of three fourth powers equal to a fourth power. Now the refinement rule forces an explicit choice, and the formal setting makes each option clean. Falsify: the universal conjecture is dead — a single genuine counterexample refutes a universal claim outright, no rescue available. Narrow the scope: one might instead ask whether the weaker bound (some number of terms) still holds, redrawing the boundary to a defensible region. Revise the mechanism: examine whether the counterexamples cluster — and indeed related counterexamples to Euler's conjecture share number-theoretic structure, which is boundary inference, telling you the failures are not scattered noise but concentrated where a particular construction applies. The prime also names the dishonest move the formal setting throws into relief: an ad hoc rescue would be appending "...except for numbers like 422481" — a content-free immunizing clause that saves the words while abandoning the claim. The iteration loop is the actual history: the refuted conjecture seeds a sharper question (how few terms are really possible?), which is re-exposed to fresh search. The intervention the prime prescribes is exactly the mathematician's craft: state the claim sharply enough to be falsifiable, hunt the input engineered to break it, and on finding one, choose openly among refute, rescope, and revise rather than quietly patching.

Mapped back: Counterexample hunting instantiates the full signature — a universal account, a deliberately sought discordant case, and an explicit refinement choice (here, outright falsification with clustering pointing to a mechanism) — making mathematics the domain where seek-disconfirmation and the refute/narrow/revise rule are at their most rigorous.

Applied/industry

Machine-learning red-teaming and clinical differential diagnosis are the same disconfirmation discipline in software and in medicine. A trained classifier is, structurally, an implicit theory of its data distribution — its current account of which inputs map to which labels, with an implied boundary it rarely states. Ordinary evaluation accumulates confirming cases (high test accuracy); negative case analysis does the opposite — adversarial testing and slice analysis deliberately hunt the inputs the model gets wrong: perturbed images, rare subpopulations, out-of-distribution examples. When the discordant cases cluster — say the model fails systematically on dark-skinned faces or on a rare dialect — that clustering is boundary inference: it localizes where the account is mis-drawn, and distinguishes a mechanism problem (biased training data) from a mere scope problem (a few odd inputs). The refinement rule then chooses openly: narrow the deployment scope (do not use on this population), revise the mechanism (rebalance the data, change the architecture), or principled-exclude with the argument explicit — never the ad hoc rescue of dismissing each failure as a one-off. Clinical reasoning runs the identical loop at the bedside: a leading diagnosis is the current account, and good practice rules out before ruling in — the clinician deliberately generates competing hypotheses and seeks the finding that would not fit the leading one (an ordered lab test, a discriminating symptom). A discordant finding (the patient improves on a treatment the diagnosis says should fail) forces the same explicit choice: discard the diagnosis, narrow it, or revise the mechanism. The shared intervention transfers verbatim across both: make the account's predictions explicit, engineer or sample cases to pressure them, and on each discordant case choose deliberately among falsify, rescope, and revise — treating the failures as the steering signal, not as embarrassments to explain away.

Mapped back: ML red-teaming and differential diagnosis are the same prime as counterexample hunting — an account stressed with deliberately sought discordant cases, with clustering driving boundary inference and an explicit refute/narrow/revise rule — so the four-step disconfirmation loop transfers across the machine-learning, clinical, and mathematical substrates.

Structural Tensions

T1 — Seeking Disconfirmation versus Accumulating Confirmation (sign/direction). The load-bearing commitment is directional — hunt cases that would overturn the account, not cases that fit. The competing default is confirmation-biased pattern-matching. The characteristic failure is amassing confirming instances and calling the account robust, when no one searched for what would break it; an account with no known counterexamples may simply be under-searched. Diagnostic: was disconfirmation actively sought, or has confirming evidence merely been allowed to accumulate?

T2 — Honest Refinement versus Ad Hoc Rescue (scopal). On a discordant case the analyst may legitimately narrow scope or revise mechanism, or illegitimately append a content-free immunizing clause that saves the words while abandoning the claim. The boundary is between principled refinement and degenerating rescue. The characteristic failure is patching the account with a special exception per counterexample, producing an unfalsifiable program. Diagnostic: does each refinement lose content (narrow, sharpen, redirect), or merely immunize against the case without giving anything up?

T3 — Scope Narrowing versus Mechanism Revision (scalar). A counterexample forces a choice the prime itself diagnoses: many isolated counterexamples favor narrowing scope, a clustered family favors revising the mechanism. The tension is between the boundary being mis-drawn and the mechanism being wrong. The characteristic failure is patching the boundary case-by-case when clustering signals a mechanism error, or rebuilding the mechanism when a scope trim would suffice. Diagnostic: do the discordant cases cluster in a sub-region (revise mechanism) or scatter (narrow scope)?

T4 — Statistical Outlier versus Theoretical Deviant (measurement). Not every off-account case is a theory problem; some are measurement noise. The competing concept is the outlier. The characteristic failure is chasing every anomaly as a deviant case (over-fitting the account to noise) or dismissing a genuine deviant as noise (missing the real refutation). Diagnostic: is this discordant case within the expected measurement variation, or does it exceed what noise could produce — making it a theoretical problem in disguise?

T5 — Falsifiability versus Where-to-Look (scopal). Whether a claim could be refuted (demarcation) is distinct from where to hunt for refuting evidence and what to do on finding it (the practical method). The boundary is between the philosophical property and the operational practice. The characteristic failure is conflating them — declaring an account "falsifiable in principle" and stopping, without the deliberate search that actually pressures it. Diagnostic: has the account merely been deemed refutable, or has the search for its refuters been designed and run?

T6 — Robustness versus Search Effort (temporal). An account's apparent strength is confounded with how hard it has been searched; the rate at which new deviant cases surface per unit of search is the real signal, not their current absence. The competing concern is the search budget. The characteristic failure is treating an old, lightly-tested account as strong because no counterexamples are known, when the discovery rate under serious search would be high. Diagnostic: is the account's clean record the product of intense disconfirmation search, or of little search at all?

Structural–Framed Character

Negative case analysis sits on the structural side of the middle of the structural–framed spectrum — mixed-structural, aggregate 0.4. The account-versus-counterexample loop is a genuinely structural skeleton, but the prime presupposes an analyst who holds an account, and three diagnostics read at the half-mark.

The structural core is the load-bearing relation: an account, a deliberate search for discordant cases, and an explicit refine/narrow/refute rule, iterated until residuals stop carrying pattern. That loop recognizes a pattern present across qualitative deviant-case sampling, counterexample hunting in mathematics, ML red-teaming and slice analysis, FMEA and HAZOP, differential diagnosis, and policy effectiveness study — the Euler-conjecture and Neptune-style cases showing it operates wherever there is an account and cases to stress it. The directional commitment (seek disconfirmation, not confirmation) is a structural property of the inference, not a value judgment, which holds evaluative_weight at 0. The three half-framed marks are honest. vocab_travels (0.5): the lexicon — deviant case, negative case, ad hoc rescue, degenerating program — is methodology-coined and travels with that accent. institutional_origin (0.5): the named origin is grounded-theory and qualitative method, an intellectual practice, and the discipline is institutionalized (pre-registration, second-coder agreement). human_practice_bound (0.5): the move presupposes an analyst with an account who can hunt and revise, though the abstract structure (an account stressed by deliberately sought counterexamples) generalizes and even runs in automated form (adversarial search). import_vs_recognize (0.5): invoking it imports the four-step disconfirmation discipline rather than merely spotting a regularity. The account-counterexample skeleton is genuine and broadly portable, which is why this is mixed-structural; the analyst-and-method framing is what keeps it from a clean zero, consistent with 0.4.

Substrate Independence

Negative case analysis is a highly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. Its domain breadth is maximal (5): the loop of holding an account, searching deliberately for the discordant case, and refining on finding one recurs with the same force across qualitative social science (its named home, a trustworthiness criterion and the basis of deviant-case selection), the scientific method (Darwin's notebooks reserving space for "facts that contradict me"; Mendeleev's table tightened by predicted-then-found elements), policy evaluation (hardening a program theory by studying the sites where it failed), product research and UX (systematic study of churned users and abandoned carts), machine learning (adversarial testing, slice analysis, out-of-distribution evaluation), engineering safety (FMEA, HAZOP, pre-mortems), investigation and clinical reasoning (ruling out competing diagnoses), and mathematics (the deliberate hunt for a counterexample that bounds or refutes a conjecture). Structural abstraction is high (4): the account/discordant-search/refine signature is medium-neutral, though it presupposes an inquirer maintaining an account, which keeps it just short of a clean 5. Transfer evidence is heavy (5), with the same move documented from qualitative methods through the natural sciences, engineering safety, and ML adversarial analysis, the mathematical counterexample being its formal limit case. Wide spread and concrete cross-domain transfer lift the composite to a strong 4, just shy of the top because the pattern is an inferential discipline that presupposes a reasoner holding an account to be stress-tested.

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

Relationships to Other Primes

One-hop neighborhood: parents above, mutual partners to the right, children below.NegativeCase Analysiscomposition: FalsifiabilityFalsifiability

Parents (1) — more general patterns this builds on

  • Negative Case Analysis presupposes, typical Falsifiability

    Negative case analysis is the practical method that EXERCISES falsifiability — the deliberate hunt for the refuters a falsifiable claim makes possible (the file: 'falsifiability is the permission slip; negative case analysis is the expedition'). Presupposes a falsifiable account to stress.

Path to root: Negative Case Analysis → Falsifiability

Neighborhood in Abstraction Space

Negative Case Analysis sits in a sparse region of abstraction space (91st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Inference & Evidence (26 primes)

Nearest neighbors

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

Not to Be Confused With

The most precise confusion to dissolve is with falsifiability, because the two are so often run together that "we made it falsifiable" gets treated as if it accomplished the work. Falsifiability is a property of a claim — the demarcation criterion that asks whether the claim could, in principle, be contradicted by some possible observation. It is a yes/no characteristic of the claim's logical form. Negative case analysis is an operational practice — the deliberate hunt for the observations that would actually do the contradicting, plus an explicit rule for what to do when one is found. The distinction is load-bearing because falsifiability is necessary but utterly insufficient: a claim can be perfectly falsifiable in principle and never once subjected to a serious search for its refuters, in which case its clean record is worthless. Negative case analysis is what exercises falsifiability — it is the activity that converts the abstract possibility of refutation into a designed, run search. A practitioner who conflates them declares an account "scientific because falsifiable" and stops, missing that the prime's entire value is in the doing: where to look, how hard, and what the refinement rule is on finding a discordant case. Falsifiability is the permission slip; negative case analysis is the expedition.

A second genuine confusion is with comparative_method, the embedding-nearest neighbor (similarity 0.86), and the difference turns on which direction the inference runs. The comparative method contrasts cases — most-similar or most-different designs — to infer what factor drives an outcome, building a generalization from the pattern of agreements and differences across cases. It is a method of constructing an account. Negative case analysis presupposes an account already in hand and stress-tests it by hunting the cases that would overturn it. The two can chain — comparison builds the theory, negative case analysis then tries to break it — but they are opposite phases. The deviant-case selection at the heart of negative case analysis is precisely not the comparative method's representative or contrastive sampling; it is the deliberately unrepresentative choice of the cases most likely to embarrass the account. Confusing them leads to the error of treating a comparative study's confirmatory pattern as if it had been stress-tested, when in fact no one went looking for the cases that would have broken the inferred generalization — comparison that accumulates agreement is not the same as a search that seeks disconfirmation.

A third confusion worth marking is with inductive_reasoning, since both operate by examining cases and both feed into theory-building. The decisive contrast is the sign of the evidence sought. Induction proceeds by accumulating confirming or representative instances and generalizing from them — more cases that fit strengthen the generalization. Negative case analysis proceeds by deliberately seeking instances that do not fit — its whole directionality is toward the disconfirming, the deviant, the case the account did not predict. This is the prime's foundational commitment, and it is the structural inverse of induction's gather-what-fits logic. The two are complementary in a healthy research program (induction proposes, negative case analysis disposes or refines), but conflating them produces the most common methodological failure the prime names: running an essentially inductive accumulation of confirming cases while believing one has tested the account, when the confirmation-biased default of pattern-matching cognition has simply been allowed to run. An account "supported by many cases" has been induced, not stress-tested — and the prime's question is always whether anyone hunted the cases that would have brought it down.

For a practitioner these distinctions cohere into keeping straight what stage of inquiry each move belongs to and which direction it runs: falsifiability is a static property of the claim, comparative method and induction build accounts (by contrast and by accumulation respectively), and negative case analysis breaks or refines an account already held (by deliberately seeking disconfirmation). The single test that isolates negative case analysis from all three is its directional commitment — it is the only one whose engine is the active hunt for what would overturn the account, with an explicit refute/narrow/revise rule waiting for whatever the hunt turns up.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.