Maximum-Variation Case Sampling¶
Procedure — instantiates Universality Extraction
Selects cases that maximize relevant variation so a proposed invariant is tested against strong differences rather than easy repetitions.
Most people gather examples that resemble each other, because resemblance is what caught their eye in the first place. Maximum-Variation Case Sampling does the opposite on purpose: it chooses cases for their differences, stocking the evidence base with the strongest contrasts and hardest near-misses a topic can offer, so that any pattern claimed later has to survive genuine variation instead of a flattering echo chamber. Its defining move is that it curates the case set before the invariant is even fixed — it is a sampling discipline, not a comparison or a test — and it treats independence and provenance as first-class selection criteria, so ten retellings of one event never masquerade as ten cases. The point is to engineer the toughest fair jury the evidence can afford.
Example¶
An ecologist suspects that critical slowing down — a system taking longer to recover from small shocks as it nears a tipping point — is a universal early-warning signal of regime shift. Rather than assembling a dozen shallow lakes (all the same kind of case), she samples for maximum contrast: a eutrophic pond, a semi-arid rangeland facing desertification, a coral reef under bleaching stress, and a kelp forest with urchin outbreaks. Each differs in medium, timescale, and driving mechanism, so a recovery-time signal that persists across all four is far harder to dismiss as an artifact of ponds.
She then deliberately plants hard negatives: a system known to shift smoothly and reversibly with no hysteresis, where the early-warning signal should be absent. If her signal lights up there too, it is measuring something other than an approaching tipping point. Finally she records source independence — two of her candidate reef datasets turn out to derive from the same monitoring program and shared instrumentation, so she counts them as one case, not two. The output is not a conclusion but a stocked shelf: a traceable ensemble built to make the eventual invariant work for its universality.
How it works¶
- Map the variation dimensions. List the axes along which cases could differ that would threaten the eventual claim — medium, scale, mechanism, population, era — rather than axes that merely add color.
- Select for high contrast and independence. Fill the corners of that space, preferring cases that share as little upstream history as possible; a common dataset, vendor, or diffusion lineage collapses several apparent cases into one.
- Seed hard negatives and near-misses. Include cases that look similar on the surface but should fall outside the class, so the sample already contains its own disconfirmers.
- Log the empty cells. Record which regions of the variation space you could not populate; those gaps become honest limits on any later claim.
Tuning parameters¶
- Variation targeting — whether the sampled differences actually stress the favored explanation. Aim the contrast at the explanation; variation on irrelevant axes buys breadth without evidence.
- Independence threshold — how aggressively you collapse cases that share provenance. Strict independence shrinks the sample but kills pseudoreplication; loose settings inflate apparent support.
- Near-miss dose — how many surface-similar hard negatives to plant. More sharpens discrimination but can starve the positive side of the ensemble.
- Breadth vs. depth — many shallow cases or fewer deeply characterized ones. Breadth locates the question; depth is what later settles causal identity.
- Empty-cell tolerance — how many unpopulated regions you accept before declaring the sample too thin to support a cross-domain claim.
When it helps, and when it misleads¶
Its strength is that it front-loads the hardest possible test: a pattern that survives a maximum-variation sample has already faced the differences most likely to break it, so downstream induction and perturbation start from evidence that is earned rather than cherry-picked.[1] It is the cheapest place to kill a false universal, before anyone has invested in naming it.
Its failure mode is breadth theater — adding domains for rhetorical sweep rather than causal diversity, which lowers evidence quality while looking more impressive. The classic misuse is counting many reports that all trace to one incident, one dataset, or one diffused policy as independent corroboration; the resemblance is then pseudoreplication, not universality. The discipline that guards against this is to select variation that targets the explanation and to track provenance ruthlessly, so the ensemble's apparent breadth reflects real independence rather than shared ancestry.
How it implements the components¶
comparison_case_ensemble— it is the ensemble builder: it produces the high-contrast, independent, provenance-traced case set, complete with logged empty cells, that every other mechanism operates on.adversarial_counterexample_set— it seeds the ensemble with hard negatives and surface-similar near-misses at selection time, so the class is born with its disconfirmers already attached rather than discovered late.
It does not align cases into a common relational language (cross_case_comparison_frame, done by Relational Case Normalization), state the pattern (candidate_macro_invariant, done by Invariant Signature Induction), or plan detail-level manipulations (microdetail_perturbation_plan, done by Microdetail Ablation Suite); this procedure stocks the shelf, it does not read, name, or perturb what is on it.
Related¶
- Instantiates: Universality Extraction — it supplies the evidentiary foundation the whole extraction depends on.
- Sibling mechanisms: Relational Case Normalization · Invariant Signature Induction · Microdetail Ablation Suite · Regime-Boundary Sweep · Red-Team Case Search · Equivalence-Class Refinement
Draft — one mechanism instantiating part of the Universality Extraction archetype; templated operating steps and generic inputs live on the archetype page.
Editorial Notes¶
Form Classification¶
Form family: Experiment, Test & Rehearsal
Rationale: The mechanism deliberately selects high-contrast cases, hard negatives, and near-misses to test whether a proposed invariant survives meaningful variation.
Nearest alternative: Decision, Gate & Allocation — Case selection is a bounded choice, but it is subordinate to the experimental exposure the selected cases create.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Ethnography & Qualitative Methods
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Selecting maximally diverse cases to test what persists is canonical qualitative sampling practice.
Related originating lineages:
- Sociology & Anthropology — For Maximum-Variation Case Sampling, institutions, membership, social scale, norms, and collective meaning materially shaped the mechanism's characteristic form.
Review resolution: Both independent reviews place the primary provenance in ethnography_qualitative_methods. The queued differences (alternate_origin_disagreement) concern secondary metadata, not primary lineage. The final retains sociology_anthropology only where a reviewer supplied a formative-lineage rationale; downstream use or broad applicability by itself is not treated as origin. origin_mode=single_lineage because one disciplinary lineage remains dominant and application breadth alone does not create another origin. domain_reach=multi_domain records established application breadth separately from provenance. confidence=high preserves the more cautious evidence assessment. encyclopedia_synthesis=false records whether either reviewer identified deliberate corpus-level composition.
Review outcome: Reconciled after independent review; high confidence.
References¶
[1] Patton, M. Q. Qualitative Research & Evaluation Methods. 3rd ed., SAGE Publications (2002). Uses maximum-variation sampling to identify patterns that recur across deliberately diverse cases. registry ↩