Transfer Assumption Review¶
Assumption audit — instantiates Domain-Specificity of Confidence
Makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
A Transfer Assumption Review is a prospective analysis run before acting: it surfaces the specific premises that would have to hold for confidence earned in a source domain to apply to a new target, and checks each one. Its defining move is to treat transfer as a hypothesis with named assumptions — do the same mechanisms operate? are the cues equally reliable? is feedback still available? are the failure modes comparable? — and to grade those assumptions one at a time, forward, before any outcome exists. It is deliberate rather than real-time, and analytic rather than a binary gate; its product is a per-assumption verdict, not a yes/no stamp.
Example¶
A team wants to redeploy a sepsis-prediction model, validated at Hospital A, onto Hospital B. Rather than assume "medicine is medicine," the review lists the assumptions the transfer rides on: the same lab assays reported in the same units, a similar sepsis base rate, comparable nurse charting cadence, and a similar patient mix. Checking each, they find Hospital B records certain vitals on a different schedule and codes one lab differently — two assumptions fail. The transfer is not abandoned, but confidence to deploy drops from "ready" to "pilot and re-validate first." The review converted a vague sense that the model "should generalize" into named, falsifiable premises — the discipline external-validity failures usually punish for being skipped.[n1]
How it works¶
- Define the target precisely. Name the actual population, context, incentives, time horizon, and operating conditions the claim will face — not "the new market," but its concrete mechanisms.
- Inventory the source warrant. List what actually supports source-domain confidence: which measured outcomes, which causal understanding, which feedback loops.
- For each warrant, ask if its generator still runs. Does the mechanism that made that evidence valid in the source still operate in the target? This per-assumption comparison is the core.
- Grade transfer assumption-by-assumption. Output a verdict on each premise, so the decision attaches to which assumptions failed rather than to a single headline confidence.
Tuning parameters¶
- Assumption granularity — how finely premises are decomposed; finer decomposition catches subtle transfer failures but multiplies the checklist.
- Evidence bar per assumption — whether a premise passes on argument or requires data; a higher bar is more rigorous and slower.
- Distance conservatism — how readily "adjacent" is judged "structurally different"; conservative framing catches hidden gaps but can veto useful transfer.
- Stopping rule — how many failed assumptions kill the transfer vs. merely downgrade it.
- Reviewer independence — self-audit vs. an outside challenger; an independent challenger resists motivated reasoning at the cost of coordination.
When it helps, and when it misleads¶
Its strength is converting a hand-wave — "it should be fine over there" — into named, checkable premises, so a target that is structurally different but superficially adjacent gets caught before commitment rather than after failure.
Its failure is analysis theater: assumptions listed and then rubber-stamped, especially when the review is built to justify a decision already made. It is also blind to the unknown-unknown assumption nobody thought to write down — the review is only as good as its premise list. The classic misuse is running it backward, assembling a reassuring set of "checks" to license a foregone conclusion. The guarding discipline is to demand disconfirming evidence for the highest-stakes assumptions and to seat an independent challenger whose job is to find the premise the proponents omitted.
How it implements the components¶
The Transfer Assumption Review fills the forward-looking transfer side of the archetype:
target_domain_definition— it specifies the concrete target context, population, and conditions the claim will actually face, refusing "merely adjacent" as a description.warrant_inventory— it enumerates the warrants behind source-domain confidence precisely so each can be tested for whether it survives the move.transfer_validity_check— its core: the assumption-by-assumption comparison of source against target that decides whether confidence may travel.
The review checks a transfer before the fact and never looks at outcomes, because none exist yet. Comparing prior confidence labels against how they actually resolved, and repairing the scope map from the misses — feedback_update_loop and boundary_exception_register — belongs to the Confidence Retrospective, its backward-looking twin.
Related¶
- Instantiates: Domain-Specificity of Confidence — the review is the transfer-validity gate the pattern applies before exporting confidence.
- Consumes: Track Record by Domain Scorecard supplies measured warrant strength in the source; Expertise Scope Matrix supplies the source and target domain definitions to compare.
- Sibling mechanisms: Expertise Scope Matrix · Claim Confidence Labeling · Out-of-Domain Prompt · Track Record by Domain Scorecard · Referral or Collaboration Protocol · Confidence Retrospective
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: Transfer Assumption Review operates as a bounded evaluation of existing evidence or work that produces a finding or disposition because it makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
Independent corroboration: The frozen evidence defines Transfer Assumption Review as 'Makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries', so its operative form is Assessment, Review & Assurance.
Nearest alternative: Analysis, Modeling & Optimization — Transfer Assumption Review includes features of an analytical, modeling, inference, comparison, or optimization procedure that derives insight or a solution, but its defining operation is a bounded evaluation of existing evidence or work that produces a finding or disposition.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Historically ambiguous
Present-day reach: Universal
Rationale: Bareinboim and Pearl, External Validity: From Do-Calculus to Transportability Across Populations makes cross-population transfer depend on explicit, testable invariance assumptions rather than treating source evidence as automatically portable. This directly supports statistics experimental design as the best-evidenced historical home of the operation—Makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.—while the alternates record adjacent lineages rather than mere domains of later use.
Related originating lineages:
- Data Science & Analytics — Data science, analytics, and operational monitoring supplies a parallel or contributing lineage for the mechanism's defining operation: makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
- Education & Pedagogy — Education, assessment, and instructional practice supplies a parallel or contributing lineage for the mechanism's defining operation: makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
- Ethnography & Qualitative Methods — Ethnography and qualitative comparative inquiry supplies a parallel or contributing lineage for the mechanism's defining operation: makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
- Logistics & Supply Chain Management — Logistics supply chain supplies a historically relevant adjacent lineage or formative practice for the operation—Makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.—but the researched evidence more directly locates the defining lineage in statistics experimental design.
- Mathematics — Mathematical modeling, proof, and abstract-structure practice supplies a parallel or contributing lineage for the mechanism's defining operation: makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries.
- Systems Thinking & Cybernetics — Feedback, system boundaries, stocks, flows, and regulation supplies a distinct formative lineage for the mechanism's transfer assumption review logic.
Review resolution: The blind reviewers disagree on primary lineage (logistics_supply_chain versus statistics_experimental_design). The defining operation is: Makes the leap from a source domain to a target domain explicit and tests, assumption by assumption, whether the warrant actually carries. The researched Bareinboim and Pearl, External Validity: From Do-Calculus to Transportability Across Populations makes cross-population transfer depend on explicit, testable invariance assumptions rather than treating source evidence as automatically portable. That is mechanism-specific evidence for statistics experimental design as the historical origin. Logistics supply chain remains represented among the uncapped alternates where it contributes a genuine formative practice, but broad deployment or governance of the operation is not by itself evidence that the mechanism originated there. origin_mode=historically_ambiguous records lineage; domain_reach=universal separately records later applicability.
Attribution caveat: Transfer review spans statistical external validity, learning transfer, and qualitative domain comparison.
Encyclopedia synthesis: The exact catalogued form synthesizes established practice rather than reproducing a single standard historical label.
Review outcome: Researched adjudication after independent review; medium confidence.
Sources consulted:
Notes¶
[n1] External validity is the degree to which a finding holds beyond the specific sample, setting, and conditions in which it was established. Transfer failures — a model or result that works in one population and breaks in another under distribution shift — are external-validity failures; the review's premise checklist is a structured way to interrogate them in advance. ↩