Skip to content

Rival Explanation Discriminator

Discriminating probe — instantiates Theory-Responsive Case Sampling Design

Chooses the one case whose outcome would separate two still-live rival explanations.

A Rival Explanation Discriminator selects the single case whose outcome would come out differently under two still-competing explanations — a case engineered so that, whichever way it falls, it eliminates at least one rival. Its defining feature is indifference to confirmation: it does not care whether the model looks good, it cares that the case adjudicates. A candidate earns its place here by discriminating power alone — how sharply it splits the live explanations — not by being typical, extreme, or convenient. Where other selection moves add evidence, this one removes a contender from the field.

Example

A site-reliability team is chasing intermittent timeouts on a checkout service. Two explanations both fit every incident logged so far. Explanation A: the database connection pool is exhausting under load. Explanation B: a downstream payment API is silently rate-limiting them during peak traffic. More incidents will not help — every past outage is consistent with both.

So the team builds a discriminating case instead of collecting another. They find the observable on which the two explanations diverge: connection-pool exhaustion depends only on their own load, while payment-API rate-limiting requires actual calls to that API. The discriminating case is a load spike that occurs while the payment API is in a scheduled maintenance window, when no external calls are being made. The prediction is written down first: if timeouts still fire → pool exhaustion; if they vanish → the API was the cause. The spike comes, the timeouts persist, and Explanation B is eliminated in a single observation. One case did what a month of confirming incidents could not.

How it works

  • List the live rivals. Enumerate the explanations still standing after current evidence.
  • Find the diverging observable. Identify the one thing the rivals predict differently.
  • Find or engineer the case that realizes it. Choose the case in which that observable will actually show.
  • Predict, then observe. Record each rival's predicted outcome before the case, so the result eliminates whichever prediction fails.

Tuning parameters

  • Rivals discriminated per case — splitting two at a time versus engineering a case that carves several. Fewer is cleaner; more is efficient but harder to interpret when results are mixed.
  • Decisiveness required — insisting on a case both rivals bet on sharply versus accepting a merely suggestive one. Demanding decisiveness makes cases rarer but conclusions firmer.
  • Observable specificity — how tightly the diverging observable is pinned. A crisp observable adjudicates cleanly; a fuzzy one leaves both rivals room to survive.
  • "Both survive" tolerance — how the pass handles a case that eliminates neither. Low tolerance sends you back for a sharper case; high tolerance risks calling a draw a decision.

When it helps, and when it misleads

Its strength mirrors the logic of the crucial case: a single decisive observation can eliminate an explanation that endless confirming cases never touch, because confirmation is cheap and elimination is not.[1] It is the loop's instrument for cutting a tie between accounts that the accumulated evidence cannot break.

Its central failure mode is the false dichotomy — treating two rivals as exhaustive when the truth is a third, unlisted explanation, so the case "confirms" a survivor that was never actually right. A quieter misuse is declaring the surviving rival proven when only one competitor was eliminated; a discriminator narrows the field, it does not crown a winner. The guarding discipline is to keep the rival set explicitly open — always ask what unlisted explanation the discriminating case would also be consistent with — and to treat elimination, never confirmation, as the mechanism's real output.

How it implements the components

A Rival Explanation Discriminator fills the adjudication slice of the archetype, not its disconfirming or revision slices:

  • rival_explanation_matrix — it operates directly on the matrix of live explanations, selecting the case that splits two of them and eliminating the loser.
  • case_learning_question — its learning question is a discriminating one: "which rival does this case's outcome favor?"
  • selection_rationale_record — the recorded reason is the per-rival prediction, logged before the case is run.

It does not hunt a case chosen to disconfirm the whole account — that is Negative Case Sampling Pass via negative_or_deviant_case_trigger; nor does it fold the surviving explanation back into the categories, which is Constant Comparison Matrix via model_revision_register.

Editorial Notes

Form Classification

Form family: Experiment, Test & Rehearsal

Rationale: Rival Explanation Discriminator operates as an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation because it chooses the one case whose outcome would separate two still-live rival explanations.

Independent corroboration: The frozen evidence defines Rival Explanation Discriminator as 'Chooses the one case whose outcome would separate two still-live rival explanations', so its operative form is Experiment, Test & Rehearsal.

Nearest alternative: Analysis, Modeling & Optimization — Rival Explanation Discriminator includes features of an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution, but its defining operation is an active test, trial, simulation, drill, or rehearsal that generates evidence through a deliberate attempt or perturbation.

Review outcome: Independent reviewer agreement; medium confidence.

Origin Attribution

Primary origin: Statistics & Experimental Design

Origin pattern: Cross-disciplinary synthesis

Present-day reach: Universal

Rationale: Choosing a case with different predictions under rival hypotheses is a canonical discriminating-experiment design.

Related originating lineages:

  • Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: chooses the one case whose outcome would separate two still-live rival explanations.
  • Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: chooses the one case whose outcome would separate two still-live rival explanations.
  • Philosophy — Philosophy of science materially articulates severe tests among live explanations.

Review resolution: Both blind reviewers agree that statistics_experimental_design is the primary historical origin. Explicit reconciliation of alternate origin disagreement, origin mode disagreement, domain reach disagreement, encyclopedia synthesis disagreement starts from reviewer_a’s mechanism-specific evidence: Choosing a case with different predictions under rival hypotheses is a canonical discriminating-experiment design. Reviewer A proposed alternates=philosophy, origin_mode=convergent, domain_reach=universal, and encyclopedia_synthesis=false; reviewer B proposed alternates=data_science, mathematics, philosophy, origin_mode=cross_disciplinary_synthesis, domain_reach=multi_domain, and encyclopedia_synthesis=true. The final record retains every independently supported alternate from either review (philosophy, data_science, mathematics) without an arbitrary cap, selects origin_mode=cross_disciplinary_synthesis to represent the combined lineage evidence, and keeps domain_reach=universal and encyclopedia_synthesis=false from the more mechanism-specific assessment. Present-day transfer is recorded as reach and is not treated as proof of historical origin.

Review outcome: Reconciled after independent review; high confidence.

References

[1] The smoking-gun and doubly-decisive evidentiary tests, catalogued in Stephen Van Evera's Guide to Methods for Students of Political Science (1997): observations chosen because rival hypotheses predict them differently, so a single well-placed case can decisively confirm one contender or eliminate another. registry