RMP.2:4.5 - Preserve Design-Family Differences

Design family	Characteristic question-to-evidence relation	Criticism that must remain visible	Load-bearing values that commonly hold selection
experimental/quantitative	Deliberately vary or assign an intervention/exposure or comparison condition and estimate a bounded difference or relation.	Rival causes, allocation, baseline imbalance, measurement error, model dependence, missingness, intervention fidelity, and applicability.	Manipulability, admissible comparison, mechanism relation, instrument validity, safety, authority, sample/case feasibility, and effect/uncertainty use.
field/observational	Observe obtaining exposures, states, changes, or contexts without free assignment and construct a comparison that bears on the claim.	Confounding, selection, measurement validity, site/context dependence, seasonality, missingness, reverse relation, and transfer.	Site access, observation admissibility, measure validity, temporal/spatial frame, comparison availability, and domain causal limits.
qualitative/reflexive	Develop a situated account of meaning, experience, practice, process, or mechanism from participants, records, interactions, or cases.	Researcher position, rival interpretation, negative/disconfirming cases, sampling relevance, context, participant voice, and unsupported transfer.	Legitimate participation/access, ethical relation, phenomenon framing, source/record admissibility, interpretive competence, and harm/confidentiality constraints.
historical/source-critical	Form a claim about an occurrence, interpretation, change, or lineage from incomplete and situated records.	Provenance, chronology, source purpose, survival and archival silences, alternative interpretations, anachronism, and corroboration limits.	Source authenticity and access, chronology, rights, archive completeness, admissible relation between record and claim, and missing rival records.
computational/model	Challenge a mathematical, algorithmic, simulation, or code-mediated claim through derivation, implementation, calibration, sensitivity, reproduction, validation, or intended-use comparison.	Code/data/environment dependence, implementation defects, parameter and structural sensitivity, calibration, validation, identifiability, and intended-use fit.	Model-to-phenomenon relation, data and code access, environment identity, validation target, compute feasibility, and domain acceptance criteria.
synthesis	Treat a contested existing evidence corpus as the governed object with a qualified question, eligibility, search, appraisal, integration, and refusal-to-pool logic.	Eligibility bias, search omissions, source dependence, heterogeneity, incompatible measures or meanings, appraisal limits, and publication/reporting bias.	Qualified corpus question, source trace, eligibility rules, appraisal method, comparable result relations, access, and currentness.
mixed	Relate two or more unlike component designs because the question needs distinct evidence relations, not because “more methods” is prestigious.	Each component’s own threats plus the integration claim, disagreement, non-comparability, dominance of one component, and forced convergence.	A reason each component is necessary, component feasibility, a warranted relation among their results, and an honest rule for disagreement or partial failure.

These families are a comparison surface, not a lifecycle or evidence hierarchy. A use may enter with an existing claim, source corpus, model, field anomaly, or proposed intervention. It may stop, repeat, switch family after a changed input, or compose unlike components explicitly.