{"path":"research/what-a-comparison-produces/literature-report-2026-11-09.md","content":"> **Kept as it came back.** The report of a read-only research helper (Claude Sonnet, an `Explore` helper) made on 2026-11-09 for [what-a-comparison-produces.md](../what-a-comparison-produces.md), the all-angles pass on ledger row BL3 (the four products have homes). It was asked for the literature on what an analogical comparison produces and where a difference lands (membership, rule or transfer), with Gentner 2010, Hesse, Norton, Bartha via the Stanford Encyclopedia, Copycat and the corpus's retrieval figures named as already covered. Claude did not open these sources itself in this pass except where the note says so; the read-status tags are the helper's. Nothing else was changed.\n\n# Research report: the products of comparison and where a difference lands\n\nNo files written. Quotes are verbatim from the sources named. Where PDF extraction mangled spacing I cleaned whitespace only. Nothing is quoted from memory. Page numbers are given only where I could see them. Everything else is tagged [page unverified].\n\nRead-status tags: [full text] / [part read] / [abstract only] / [secondary].\n\nHeadline findings:\n- No source gives your five-way partition (matched / unmatched-but-connected / same-role-different-filler / mismatched-but-should-match / in no role). The nearest is Yan, Forbus & Gentner 2003, which does claim completeness, but only for the *ways a match can fail* (§10).\n- Almost every source treats the carrying (transfer of one inference) as where analogical arguments are defeated. Legal theory is the exception: it contests whether distinguishing is a membership/scope move or a reason-balance move (§7).\n\n---\n\n## 1. Gentner & Markman 1997; Markman & Gentner 1993 (\"Splitting the differences\")\n\nSources:\n- Gentner & Markman 1997, \"Structure mapping in analogy and similarity\", American Psychologist 52(1), 45-56. https://groups.psych.northwestern.edu/gentner/papers/GentnerMarkman97.pdf [full text]. The extraction shows page headers but I could not map quotes to page numbers [page unverified].\n- Markman & Gentner 1993, \"Splitting the differences: A structural alignment view of similarity\", Journal of Memory and Language 32, 517-535 (volume and pages from my own knowledge [unverified]). https://groups.psych.northwestern.edu/gentner/papers/MarkmanGentner93b.pdf [part read, abstract plus extracts].\n- A second Markman & Gentner PDF, https://groups.psych.northwestern.edu/gentner/papers/MarkmanGentner93a.pdf [part read]. I take it to be \"Structural alignment during similarity comparisons\" (Cognitive Psychology 25) [title unverified]. It carries the cleanest statement of the split.\n\n**Definition (1993b, abstract):**\n- \"beyond the commonality-difference distinction, there is a further distinction between differences related to the common structure (alignable differences), and differences unrelated to the common structure (nonalignable differences).\"\n- Body text: \"Because this difference occurs within elements connected to the common structure, we refer to it as an alignable difference (AD).\" \"the box of spare parts in the bottom scene does not match any object in the top scene. We call this type of difference a nonalignable difference (NAD).\"\n- The 1993a PDF puts it this way. \"The output of this structural alignment process is a structurally consistent mapping. It is this mapping that is used to determine the commonalities and differences of the pair. Specifically, the matching elements are the commonalities of the pair... differences are split into (1) those differences connected to the matching information (called alignable differences (AD)) and (2) those differences not connected to the matching information (called nonalignable differences (NAD)).\"\n- Its example: the top object is a circle in one configuration and a square in the other. \"an alignable difference, because these items are only placed in correspondence due to the common relational role they play.\" The extra triangle \"is a nonalignable difference, because there is nothing that corresponds to it in the other configuration.\"\n- A later operationalization (Gentner & Gunn 2001, https://groups.psych.northwestern.edu/gentner/papers/GentnerGunn01.pdf [part read]) says alignable differences are \"stated as explicitly differing values on a common dimension or predicate (e.g., squirrels have fluffy tails, mice have thin tails), whereas nonalignable differences are stated by asserting a fact for one item and denying it for the other (squirrels have feet, carpets don't).\"\n\n**Why a three-way split (1993b, discussion):** \"According to this view, comparisons yield commonalities, alignable differences and nonalignable differences. These elements can be used to determine the overall similarity of the objects or can serve as the input to another cognitive process.\"\n\n**Evidential status of an alignable difference.**\n- It is a salience and weighting claim, not a refutation. 1997: \"alignable differences are more salient than nonalignable differences. Intuitively, this focus on alignable differences makes sense, for it leads to a focus on those differences that are relevant to the common causal or goal structure that spans the situations.\"\n- 1997 also reports that alignable differences get more weight in choice: \"alignable differences are given more weight in choice situations than are nonalignable differences.\"\n- Similarity-lowering effect: participants chose the nonalignable-difference option as more similar to the target, \"suggesting that the alignable difference decreased the similarity of the pair more than did the nonalignable difference (Markman & Gentner, 1996).\"\n- Counterpoint (1993b, regression, standardized weights): Similarity = 0.75 (Commonalities) − 0.06 (AD) − 0.22 (NAD). Commonalities dominate, and in that task NADs cost more rating points than ADs.\n- I found no sentence in either paper saying an alignable difference refutes anything. Structure-mapping is a descriptive psychological theory. Validity is not its business (see §2).\n- One sentence goes the other way: a difference that is not connected to the common system does not count against the analogy (1997). \"it does not detract from the analogy that the planet does not look like a boat\". Also: \"The magnet analogy, for example, will stand or fall according to whether the causal relations between two magnets are the same as those between the sun and planet, and not according to the resemblance between a magnet and the sun.\"\n\n**Candidate inferences, 1997 definition:** \"When we have aligned a system in the base domain with a (typically less complete) system in the target domain, then further statements (candidate inferences) connected to the base system in the base can be projected into the target. These candidate inferences are only guesses: Their factual correctness must be checked separately.\"\n\n**A difference that defeats a carrying (1997, on Kepler).** Kepler asked whether the motive power could undergo an eclipse like light. \"he decided not and used this disanalogy to conclude that the motive power cannot be the same thing as the sun's light.\" So a single difference blocked one projection and also blocked an identity claim, while the analogy itself survived as productive.\n\n**Membership link (1997, Figure 5, \"Uses of Comparison in Cognitive Processing\").** The figure lists perceiving similarity (alignment), perceiving differences, categorizing (selecting best match from memory), and discovering new category (abstracting common system). It says alignment \"points the way to a reconciliation of similarity-based and theory-based accounts\" of categorization, with the example that bats are categorized by relational properties (live birth) rather than perceptual similarity. This is the structure-mapping version of membership.\n\n---\n\n## 2. SME (Falkenhainer, Forbus & Gentner 1989); Yan, Forbus & Gentner 2003\n\n### SME\n\nSources:\n- Falkenhainer, Forbus & Gentner 1989, \"The Structure-Mapping Engine: Algorithm and examples\", Artificial Intelligence 41, 1-63. https://groups.psych.northwestern.edu/gentner/papers/FalkenhainerForbusGentner89.pdf [full text, grepped]. Page numbers below come from OCR running headers, so they are approximate.\n- SME4 supplement (Forbus et al.), https://www.qrg.northwestern.edu/software/sme4/sme_complexity.pdf [part read].\n\n**The three outputs of a gmap (§3.2, about p. 13):** \"(1) Correspondences: A set of pairwise matches between the expressions and entities of the two dgroups. (2) Candidate inferences: A set of new expressions which the comparison suggests holds in the target dgroup. (3) Structural evaluation score.\"\n\n**The \"three kinds of criteria\" for a mapping (§2):**\n- \"The structural criteria include the number of similarities and differences, the degree of structural similarity involved, and the amount and type of new knowledge the analogy provides via the candidate inferences. The second criterion concerns the validity of the match and the inferences it sanctions. The inferences must be checked against current world knowledge... The third criterion is relevance, i.e., whether or not the analogy is useful to the reasoner's current purposes. Structure-mapping focuses on structural criteria only.\"\n- §3.2: \"Each candidate inference must be interpreted as a surmise, rather than a logically valid conclusion. The SES reflects the aesthetics of the particular type of comparison, not validity or potential usefulness. Testing the validity of candidate inferences and determining the utility of a match are left to other modules.\"\n\n**Definition of a candidate inference (§3.2.3, about pp. 22-23):**\n- \"Not just any information can be carried over—it must be consistent with the substitutions imposed by the gmap, and it must be structurally grounded in the gmap.\"\n- \"SME begins by examining each root BR in the base dgroup to see if it is an ancestor of any match hypothesis roots in the gmap. If it is, then any elements in Descendants(BR) which are not in BaseItems(GM) are included in the set of candidate inferences.\"\n- \"Each candidate inference is only a surmise, which must be tested by other means. By theoretical assumption, general testing for validity and relevance is the province of other modules which use SME's output.\"\n- Worked case: \"GREATER-THAN(DIAMETER(coffee), DIAMETER(ice-cube)) is not a valid candidate inference for the first gmap because it does not intersect the existing gmap structure.\" A base item unconnected to the mapping is **not** a candidate inference. This is your \"in no role\" slot.\n- SME4 supplement: \"each root expression of the base that intersects the subset of the base that is mapped by M but is not already included in the correspondence gives rise to a candidate inference.\" It states a design choice: \"The most conservative criterion requires overlapping statements, the most liberal criterion requires only overlapping entities... the default mode of operation is the liberal criterion.\"\n- SME4 supplement on the reverse direction: \"Reverse candidate inferences are used when reasoning about differences, for example.\" So target-side unmatched structure that the base lacks is formally the same object as a candidate inference run backwards.\n\n### Yan, Forbus & Gentner 2003\n\nSource: \"A theory of rerepresentation in analogical matching\", Proc. 25th Cognitive Science Society. https://qrg.northwestern.edu/papers/Files/QRG_Dist_Files/QRG_2003/YanForbusGentner03.pdf [full text].\n\n**Definition:** \"Rerepresentation re-construes parts of compared situations in order to improve a match.\"\n\n**Where it sits.** Structure-mapping puts rerepresentation outside the matcher: \"The mapping(s) produced must be examined to see if they yield results that are useful for the current task. If they do not, changes ranging from tweaking the content of the base and target (i.e., rerepresentation) to choosing a new base or even abandoning the current line of effort are options.\" Strict identicality is the default: \"relationships must by default be identical, and only under special circumstances should looser criteria (such as minimal ascension [Falkenhainer 1990]) be used.\"\n\n**Licensing.** Rerepresentation is licensed only at detected opportunities, and by task or context. \"Strategies for controlling the rerepresentation process depend heavily on context and task demands.\" They add a systematicity preference and \"high selectivity: ... very few of the possible opportunities are selected for consideration.\"\n\n**Operations (Table 2).** The table was garbled in extraction. By constraint violated:\n- Identicality violated: Transformation (rewrite into equivalent statements, e.g. greaterThan with the arguments reversed; or cross-dimensional rewriting via a dimension-independent greaterThan) and Decomposition (use the axioms of a relation's meaning to find a shared aspect, e.g. walk and run both decompose to moveTo).\n- 1:1 violated: Entity splitting (one entity plays two roles, so split it into aspects) and Entity collecting (reify equivalent entities into a set).\n- Minimal ascension is named as the earlier, pre-existing criterion for non-identical relations that share \"a close superordinate\" (Falkenhainer 1990). It is not one of the four families in this paper. The other four-family list (transformation, decomposition, entity splitting, entity collecting) is the paper's own.\n\n**Gentner's \"four products\" versus this paper.** Yan et al. treat rerepresentation as a repair step that changes the base or target and then re-matches. Gentner 2010 (already in your corpus) lists it as a product of comparison. That is a real difference in status between the two descriptions of the same operation.\n\n---\n\n## 3. Weitzenfeld 1984, \"Valid reasoning by analogy\", Philosophy of Science 51(1), 137-149\n\nURL: https://doi.org/10.1086/289169 [abstract only; the full text is paywalled and I could not retrieve it]. I could not verify his determining-structure criterion verbatim. Nothing below is a Weitzenfeld quote except the abstract and one SEP-quoted line.\n\n- Abstract: \"I propose that it is apodeictic reasoning from a premise about isomorphic structures that is often uncertain, but which we may have good reasons to believe. I characterize the structures and their isomorphism, describe patterns of reasoning appropriate to them, and discuss some complications not immediately obvious.\"\n- SEP (Bartha, https://plato.stanford.edu/entries/reasoning-analogy/ §4.1) [full text of the SEP; secondary on Weitzenfeld]: \"analogical arguments are deductive arguments with a missing (enthymematic) premise. That premise typically amounts to a determination rule.\" SEP quotes him on surveillance: \"when an array of elements is displayed, either in a visual presentation or to other senses, an isomorphism may be immediately apparent. (1984: 147)\", and \"no one quite knows how it is done.\"\n- Bartha's lecture notes (https://studyres.com/doc/15108392/ [secondary]) give a \"determining structure\" definition \"as for Russell\": \"{a1, …, an, at} constitutes a determining structure if the value of at is determined by the values of a1 through an.\" They add: \"Strategy... Find a source analogue that differs from the target only on variables whose effect can be calculated.\" That is the notes author's paraphrase, not Weitzenfeld's wording.\n- Russell/Davies: \"Determination rule: the value of a variable (or set of variables) Q is completely determined by the values of the variables P1,…,Pm\" (same notes). Davies 1988, https://web.stanford.edu/~davies/Davies-Determination-1988.pdf [search highlight only].\n- Implication for differences: a difference matters iff it lies on a variable in the determining set of the projected predicate. Differences elsewhere are irrelevant. This follows from the determination-rule formalization (Davies/Russell and Bartha's SEP summary), not from a verified Weitzenfeld sentence.\n- Bartha's own critique of this approach, from the same notes: \"To know a determination rule is to have a complete list of which attributes are relevant.\"\n\n---\n\n## 4. Bartha 2010, *By Parallel Reasoning* (OUP)\n\nSource: https://doi.org/10.1093/acprof:oso/9780195325539.001.0001 [abstract only for the book]. I could not read the book. The primary-quality text I got is Bartha 2020, \"Norton's material theory of analogy\", SHPS, https://philsci-archive.pitt.edu/18053/1/Norton_MaterialTheory_preprint_SHPS2020.pdf [full text], and the NDPR review of the book, https://ndpr.nd.edu/reviews/by-parallel-reasoning-the-construction-and-evaluation-of-analogical-arguments/ [secondary].\n\n**Core conditions (Bartha 2020):**\n- \"(i) there is a 'prior association': a clear articulation of the vertical relationship in the source domain, and (ii) there is a 'potential for generalization', i.e., no obstacle to a generalization that extends this relationship to the target domain. (The key issue is whether or not the features that belong to the negative analogy play an essential part in the prior association, potentially blocking transfer to the target.)\"\n\n**The two evaluation criteria (Bartha 2020, §7):**\n- \"Assumption 1: we can identify a set of critical factors X1, …, Xn in the source domain... The critical factors X1, …, Xn are those features that play an essential role in the prior association.\"\n- \"Assumption 2: we can identify corresponding features X1*, …, Xn*, Q* in the target domain. This step once again depends upon context. Analogical arguments require us to defend (and criticize) assertions of correspondence.\"\n- \"Overlap: there is a nonempty positive analogy. This means that some (at least one) analogues of the critical factors, X1*, …, Xn*, are known to hold in the target.\"\n- \"No critical disanalogy: no critical factor belongs to the negative analogy. This means that none of X1*, …, Xn* is known not to hold in the target.\"\n\n**Answer to \"does it attack the association (rule) or the correspondence?\"**\n- Neither, strictly. The prior association is fixed in the source, so the rule is not attacked. A critical disanalogy attacks the *potential for generalization*. That is, it is a reason to think the same kind of connection may not obtain in the target, which blocks transfer (the carrying).\n- The NDPR reviewer's gloss: the argument makes Q* plausible \"iff there is at least one critical property that is known to be shared between the two domains and no critical properties that are known to not be shared; the more critical or secondary properties that are shared, the stronger the analogy, and the more secondary properties that are not shared, the weaker the analogy.\" So only *critical* factors defeat. *Secondary* factors only weaken.\n- The SEP wording (§3.5, in your corpus) is \"there must be no critical disanalogy between the domains.\" I did not re-quote it.\n- Horizontal vs vertical (Bartha 2020, §3): \"The horizontal relationships in this diagram are similarities and differences between the two domains. The vertical relationships are always causal relations in Hesse's work, but I (Bartha, 2010) expand this idea to include predictive, explanatory, functional and statistical relations.\" The vertical relation is the rule. The horizontal relation is the correspondence.\n- Bartha adds that the advocate/critic framing is built in. The argument is \"presented by an enthusiastic advocate to a polite but moderately skeptical interlocutor, the critic\" (quoting his 2010 p. 5). Critical disanalogy is the critic's move.\n- Bartha's own definition of \"critical disanalogy\" in the book: [unverified]. The above is from his 2020 restatement.\n\n---\n\n## 5. Juthe 2005, \"Argument by analogy\", Argumentation 19, 1-27; Juthe 2015\n\nSources:\n- Juthe 2005 https://doi.org/10.1007/s10503-005-2314-9 [abstract and keywords only]. Keywords include \"assigned-predicate\", \"counterpart\", \"one-to-one correspondence\", \"conclusive analogy\", \"inconclusive analogy\".\n- Juthe 2015, \"Analogical argument schemes and complex argument structure\", Informal Logic 35(3), https://doi.org/10.22329/il.v35i3.4211 [part read, extracts].\n- Juthe 2019, https://doi.org/10.12775/llp.2019.025 [part read].\n- Juthe's review of analogy taxonomies, Cogency, https://cogency.udp.cl/index.php/cogency/article/download/281/109 [part read].\n\n**The structure (Juthe 2015):**\n- \"all analogies are composed of four components and two critical relations. The components are the Analogue A, which is the source from which a predicate is transferred to and concluded about the Target-Subject. The Target-Subject TS is that to which one assigns a new predicate; the Assigned-Predicate AP is the new predicate that is transferred, mutatis mutandis, to the Target-Subject, because the element ∈ 'determines' the Assigned-Predicate and because this element is comparable to a counterpart element ∈* in the Analogue A [sic; the Target-Subject].\"\n- \"'Comparability' means that each element in the Analogue that has a vertical determining relation to the Assigned-Predicate also has a horizontal one-to-one correspondence to a counterpart element in the Target-Subject.\"\n- \"the determining relation says neither that the element ∈ of the Analogue always determines the Assigned-Predicate nor that the element ∈ in general determines the Assigned-Predicate. Rather, it says that the element ∈ determines the Assigned-Predicate in the Analogue.\" This makes the vertical relation a particular, not a universal rule.\n- \"The inference proceeds via an insight of resemblance between particulars and not via any general principle.\"\n- He identifies relevance with determination (following Davies 1988): \"In this way, we define relevance.\" The determining relation \"can be any type of relation (including probable, causal, epistemic, normative, evaluative, resultant or supervenient)\" (2019, quoting 2005, p. 10).\n- \"It explains why there is always an issue concerning relevant similarities and relevant differences.\"\n\n**Implication for difference-landing (my inference from his definition):** a difference lands on the horizontal one-to-one correspondence for a determining element, so it attacks comparability, i.e. the carrying. It does not attack the vertical determining relation in the analogue.\n\n**Shelley's four-way taxonomy of counterarguments** (Shelley 2002-2004, as reported in the Juthe/Cogency review [secondary]; I did not read Shelley):\n- False analogy: \"rejects the original analogy by showing relevant differences between the source and the target case... has a destructive effect since it does not replace the criticized conclusion with a new one.\"\n- Misanalogy: like a false analogy, but \"also suggests a revised construal of the analogy which yields a new conclusion.\"\n- Disanalogy and counter-analogy: \"the analogy is accepted as a correct analogy, but the counterarguments operate by overriding the original analogy, through presenting further relevant data that motivate an alternative conclusion.\" Disanalogy uses the same source domain. Counter-analogy uses a different one.\n- Shelley classifies them on orientation (reject the analogy or accept it) and effect (new conclusion or not). This is the only taxonomy I found that explicitly separates \"the comparison itself fails\" from \"the comparison stands but is outweighed\".\n\n---\n\n## 6. Holyoak & Thagard; Hummel & Holyoak\n\n**Holyoak & Thagard 1989**, \"Analogical mapping by constraint satisfaction\", Cognitive Science 13, 295-355. https://doi.org/10.1016/0364-0213(89)90016-5 [part read, extracts]. Summary page: https://jimdavies.org/summaries/holyoak1989a.html [secondary].\n- \"The major assertion of the theory is that mapping is governed by constraints of the three basic types... structural, semantic, and pragmatic. None of these constraints is absolute; rather, they provide 'pressures'...\"\n- Structural: \"The structural constraint of isomorphism favors mappings that satisfy the consistency criterion... and are one-to-one.\"\n- Semantic: \"supports possible correspondences between elements to the degree that they have similar meanings.\"\n- Pragmatic: \"favors correspondences that are pragmatically important to the analogist.\"\n- The paper says nothing about \"products\" as such. Its output is a mapping. Pragmatic centrality is the nearest thing to a relevance filter on which differences matter. It appears at the mapping stage, where SME puts relevance outside the matcher.\n\n**Hummel & Holyoak 2003**, \"A symbolic-connectionist theory of relational inference and generalization\", Psychological Review 110, 220-264. https://labs.psychology.illinois.edu/~jehummel/pubs/lisa2.pdf [part read].\n- Two products from a mapping: inference (\"self-supervised\": \"the decision about whether to license an analogical inference, and if so what inference to make, is based on the system's knowledge about the mapping\") and schema induction (\"The same principles, augmented with the capacity to discover which elements are shared by the source and target, allow LISA to induce general schemas from specific examples\").\n- Schema induction is described as \"a kind of intersection discovery at the level of complete propositions.\"\n- They state the scope: \"we treat aspects of inference that can be viewed as pattern completion, but exclude subsequent evaluation and adaptation of inferences (as the latter processes make open-ended use of knowledge external to the given analogs).\" They also say of their 15 phenomena: \"We make no claim that these phenomena form an exhaustive set.\"\n- Their products (inference, schema) correspond to Gentner's inference-projection and abstraction. Neither treats contrast or rerepresentation as a product.\n\n---\n\n## 7. Legal reasoning on precedent: where \"distinguishing\" lands\n\n**Sources and read-status:**\n- Horty 2011, \"Rules and reasons in the theory of precedent\", Legal Theory 17, 1-33. http://www.horty.umiacs.io/articles/2008-rrp.pdf [full text].\n- Lamond 2005, \"Do precedents create rules?\", Legal Theory 11, 1-26. http://www.horty.umiacs.io/courses/readings/lamond-2005-precedents-rules.pdf [full text].\n- Mullins 2026, \"Distinguishing and reinterpreting in the reason model of precedent\", Legal Theory 32, e1. https://www.cambridge.org/core/journals/legal-theory/article/distinguishing-and-reinterpreting-in-the-reason-model-of-precedent/3E3B7FEB057867DF09250987ABE7728D [part read: abstract, §1-2].\n- Rigoni 2015, \"An improved factor based approach to precedential constraint\", AI & Law 23, 133-160. https://doi.org/10.1007/s10506-015-9166-x [abstract only].\n- Ashley's HYPO/CATO: I did not read Ashley's primary work. Secondary sources only. Prakken 2021, https://link.springer.com/article/10.1007/s10506-021-09284-6 [part read]; Aleven/Ashley-style algorithm text in \"Arguing with cases\", https://jurix.nl/pdf/j97-07.pdf [part read]; Bench-Capon, https://ceur-ws.org/Vol-3614/paper3.pdf [part read]; Bench-Capon, \"HYPO's legacy\", https://www.csc.liv.ac.uk/~tbc/publications/hypoLegacy.pdf [secondary].\n\n### Standard accounts of what distinguishing is\n\n**(a) Rule model, scope reading.**\n- Lamond describes the traditional claim: \"On this approach, distinguishing is really a form of reinterpretation, which seeks to make the ratio more closely match the justification for the rule.\"\n- The ratio's scope is re-drawn so the new case falls outside it: a membership move.\n- Lamond criticizes it: it \"makes the identification of the ratio sound far more mysterious than legal practice warrants.\"\n\n**(b) Reason model, balance reading (Lamond).**\n- \"Distinguishing is the practice whereby later courts cite some difference between the facts of the precedent case and the facts of the later case to explain why they are not following the precedent. A later case may clearly fall within the scope of the earlier ratio, that is, it may be a straightforward case of {J, K, L}, but the later court may decline to reach the result C on the basis that there is some feature in the later case not present in the earlier case that provides a good reason not to reach the result C.\"\n- So on this reading the case is *in* the rule's scope (membership holds), and the rule is not attacked (\"later courts treat earlier cases as correctly decided on their facts\"). A new reason defeats the carrying of the outcome.\n- \"Distinguishing and following are simply two sides of the same coin: case-by-case decision making is the process of determining whether or not a precedent should be followed, given the differences between the case before the court and the precedent case.\"\n- \"Later courts, then, are not bound by rationes—they are bound by precedents; and they are not bound to follow precedents—they are bound to determine whether they should be followed or distinguished.\"\n\n**(c) Horty (formal reason model).**\n- Case = ⟨X, r, s⟩ (fact situation, rule, outcome). A rule is \"applicable\" when its premise ⊆ X. This is membership, defined as a factor-subset test.\n- A rule is **trumped** \"if and only if there is another rule r′ ... that is also applicable in X, but which is such that (1) Premise(r) < Premise(r′) and (2) Conclusion(r′) = Conclusion(r)\" [sic; the extraction dropped a \"≠\"]. A rule is binding if applicable and not trumped (Definition 5-6).\n- \"We can say that the court distinguishes the binding rule r, by contrast, whenever it either follows or formulates another rule r′ supporting the opposite outcome s′... the operation must be performed with some care, guided by the constraints of precedent.\"\n- So in Horty, distinguishing leaves the old rule valid and applicable. It adds a new rule that trumps it, subject to case-base consistency. The old rule is not refuted, and membership is not denied. The effect is on which rule governs the outcome.\n\n**(d) HYPO/CATO.**\n- Prakken's summary: \"In CATO this happens by indicating a factor in the current case that is absent in the cited precedent and that supports the opposite outcome, or a factor in the precedent that is missing in the current case and that supports the outcome of the cited case. In HYPO [it] can happen by pointing at a dimension that more strongly favours the decision in the precedent than in the current case.\"\n- A distinction is then answered by \"downplaying\" or \"emphasizing\". The 3-ply structure is cite, distinguish, rebut-by-downplaying.\n- Bench-Capon's CATO example (Mason v. Jack Daniel Distillery as precedent against MBL): the defendant distinguishes with \"In MBL, the information was known to competitors. This was not so in Mason.\" The plaintiff downplays with \"In Mason, the information was also obtainable, through reverse engineering, and the security measures taken by Mason were sufficient to counter this availability.\" The judge then weighs.\n- Mason citation: 518 So. 2d 130 (Ala. Civ. App. 1987) [volume and page from memory, unverified]. Justia page, https://law.justia.com/cases/alabama/court-of-appeals-civil/1987/civ-6008.html [part read].\n\n**(e) The contested boundary: restrictive reinterpretation.**\n- Mullins separates two moves. \"In the most straightforward cases, courts distinguish by drawing attention to some aspect of the new case that was not present in the facts before the court in the precedent case. In other cases, however, courts avoid the effect of binding precedent by drawing attention to some facts of the precedent case that are absent in the new case. Unlike straightforward cases of distinguishing, this second method of avoiding precedent requires courts to reinterpret the precedent so that it is narrower in scope.\"\n- His conclusion: \"proponents of the reason model should resist assimilating the practice into a form of distinguishing. I argue, instead, that restrictive reinterpretation should be viewed as a form of constrained overruling of precedent.\"\n- Mapped onto your three targets (this mapping is mine, not Mullins's):\n  - Straightforward distinguishing, a new-case fact that is absent from the precedent, is an attack on the **carrying**. The rule and membership both stand.\n  - Restrictive reinterpretation, a precedent-case fact absent from the new case that is now treated as necessary, narrows the rule and denies membership. Mullins treats it as a modification of the **rule** (constrained overruling), not as a membership finding.\n- Mullins says the reason model faces a collapse threat: \"it threatens to lead to a collapse in the distinction between distinguishing and rule-modification that provides one of the principal motivations for the reason model.\"\n- Rigoni: only the abstract was read (rule-based, non-monotonic theories linked to ratio decidendi and obiter dicta). I did not read his treatment of distinguishing.\n\n### Worked legal examples\n\n1. **R v Sharp [1987] QB 853 (CA), distinguishing R v Graham (1982) 74 Cr App R 235 (CA)**, Graham having been endorsed by the House of Lords in R v Howe [1987] AC 417 (Lamond 2005, n. 15 [full text]).\n   - Lamond's description: \"when a former member of a violent criminal organization sought to rely on the defense of duress to a criminal charge under English law, it was held that the defense was not available because the defendant had voluntarily exposed himself to the risk of such threats, despite the earlier formulations of the defense mentioning no such restriction on the availability of the defense.\"\n   - Landing: the case plainly met the duress test (membership). The earlier statement of the test was not held wrong (not an attack on the generalization). A new fact, voluntary exposure, defeated transfer of the \"defence available\" outcome.\n   - Caveat: Lamond's footnote pairs Sharp with Graham. I did not verify in the primary reports which formulation exactly was \"earlier\".\n2. **Donoghue v Stevenson [1932] AC 562 and its progeny, from SEP Example 10** (https://plato.stanford.edu/entries/reasoning-analogy/ [full text]).\n   - The principle was extended in Haseldine v CA Daw & Son Ltd [1941] 2 KB 343 (negligent lift repair).\n   - \"By contrast, the principle was not applicable to a case where a workman was injured by a defective crane, since the workman had opportunity to examine the crane and was even aware of the defects (Farr v. Butters Brothers & Co. 1932 2 KB 606).\"\n   - Landing: a difference on a factor tied to the rule's own condition (\"the consumer had no possibility of intermediate examination\" is part of the Donoghue principle as SEP states it). So Farr is better read as a membership finding (outside the population the rule covers) than as a carrying failure. Under SEP's description this is a rule-scope case, not a trumping case. Compare Sharp, where the difference is outside the rule's stated conditions.\n   - This pair shows the two landing sites in one body of law. I have not read the primary reports.\n\n---\n\n## 8. Walton, Reed & Macagno 2008; Guarini 2004; Govier\n\n**Scheme, Argumentation Schemes (CUP 2008), p. 315.** I could not read the book. I saw the same wording in five separate secondary quotations, all citing p. 315, so I rate it a reliable reproduction [secondary].\n\n- Sources:\n  - https://doi.org/10.2139/ssrn.2185945 (Macagno & Konstantinidou)\n  - https://doi.org/10.47012/jjmll.13.2.7\n  - http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.840.2950 (Dove & Guarini)\n  - https://www.math.uni-hamburg.de/en/personen/loewe/2010-11-II/RFMFS-9.pdf\n  - the Budán et al. CACIC 2013 paper.\n- Premises:\n  - Similarity Premise: \"Generally, case C1 is similar to case C2.\"\n  - Base Premise: \"A is true (false) in case C1.\"\n  - Conclusion: \"A is true (false) in case C2.\"\n- Critical questions:\n  - CQ1: \"Are there differences between C1 and C2 that would tend to undermine the force of the similarity cited?\"\n  - CQ2: \"Is A true (false) in C1?\"\n  - CQ3: \"Is there some other case C3 that is also similar to C1, but in which A is false (true)?\"\n- Walton on CQ1 (via Budán et al., quoting Walton [18]; secondary of secondary): \"In general, the first critical question for the argument from analogy tends to be the most important one to focus on when evaluating arguments from analogy. If one case is similar to another in a certain respect, then that similarity gives a certain weight of plausibility to the argument from analogy. But if the two cases are dissimilar in some other respect, citing this difference tends to undermine the plausibility of the argument.\"\n- Landing: in the scheme, a difference attacks the *similarity premise*, which is the correspondence/comparability. CQ2 challenges the base. CQ3 is a counter-analogy (rival base). No critical question challenges a general rule, because the scheme has none. \"Undermine the force of the similarity cited\" means *weakens*, not *refutes*. The scheme is also silent on which differences count.\n\n**Guarini 2004, \"A defence of non-deductive reconstructions of analogical arguments\", Informal Logic 24(2), 153-168.** https://informallogic.ca/index.php/informal_logic/article/download/2141/1585 [part read: abstract, §15]. He opposes Waller's deductive reconstruction: \"some analogical arguments do not fit Waller's deductive schema, and... such a schema does not allow for an adequate account of the strengths and weaknesses of an analogical argument.\" Govier's a priori schema as he quotes it (Govier 1989, 144): \"1. A has x,y,z. 2. B has x,y,z. 3. A is W. 4′. It is in virtue of x,y,z that A is W. 5. Therefore, B is W.\" Guarini's functional taxonomy distinguishes analogies that support a *classification or treatment* judgment from those that support a *prediction*. The first of those is the nearest philosophy-of-argument analogue to membership.\n\n**Govier** (a priori vs inductive analogies; \"refutation by logical analogy\"): [secondary, via the Juthe/Cogency review only; Govier not read].\n\n---\n\n## 9. Nersessian; Chalmers, French & Hofstadter\n\n**Nersessian, \"The cognitive basis of model-based reasoning in science\"** (in Carruthers, Stich & Siegal (eds), *The Cognitive Basis of Science*, CUP 2002, pp. 133-153). https://mechanism.ucsd.edu/bill/teaching/F12/cs200/Readings/Nersessian.cognitivebasisofmodel-basedreasoninginscience.2002.pdf [part read]. Page 144 appears in the extraction.\n- \"constraints drawn from both the target and source domains are domain-specific and need to be understood in the reasoning process at a sufficient level of abstraction for retrieval, transfer and integration to occur. I call this type of abstraction 'generic'.\"\n- \"In viewing a model generically, one takes it as representing features, such as structure and behaviours, common to members of a class of phenomena... Generality in representation is achieved by interpreting the components of the representation as referring to object, property, relation, or behaviour types rather than tokens of these.\"\n- On analogy: \"Generic abstraction can been seen to be highly significant in analogical modelling, since objects and relations often need to be understood as lacking specificity along certain dimensions in order for retrieval and transfer to occur.\"\n- On evaluation: \"Evaluation of the analogical modelling process is in terms of how well the salient constraints of a model fit the salient constraints of a target problem, with key differences playing a significant role in further model generation (Griffith, Nersessian and Goel, 1996; Griffith, 1999).\"\n- Reading: generic abstraction is the abstraction product, and it is what makes a membership claim possible, because a class is defined by what is held constant. Differences \"play a role in further model generation\" rather than refuting. Compare Maxwell: the generic model \"will not map back\" onto the source domains (summary of Nersessian 1997, https://jimdavies.org/summaries/nersessian1997.html [secondary]).\n\n**Chalmers, French & Hofstadter 1992**, \"High-level perception, representation, and analogy: a critique of artificial intelligence methodology\", JETAI 4, 185-211. https://consc.net/papers/highlevel.pdf [part read: abstract and introduction].\n- \"we argue that perceptual processes cannot be separated from other cognitive processes even in principle, and therefore that traditional artificial-intelligence models cannot be defended by supposing the existence of a 'representation module' that supplies representations ready-made.\"\n- It aims at \"a model of high-level perception and analogical thought in which perceptual processing is integrated with analogical mapping, leading to the flexible build-up of representations appropriate to a given context.\"\n- Chalmers' own summary of the paper's target: using \"frozen\", hand-coded representations in cognitive modeling (BACON, SME) (https://consc.net/ai-and-computation/ [secondary]).\n- Contrast: Yan et al. 2003 keep representation-building outside the matcher and mapping, as a post-hoc repair loop. Chalmers et al. say that separation cannot be drawn. A reply by Forbus, Gentner, Markman and Ferguson exists (listed at consc.net; I did not read it).\n- Yan et al. themselves cite the Hofstadter group's Copycat/Tabletop as \"the most closely related work on rerepresentation\".\n\n---\n\n## 10. Is the set of products exhaustive relative to a mapping?\n\nNo source states the five-way partition you describe. The closest claims:\n\n1. **Yan, Forbus & Gentner 2003: exhaustive, but over failure modes of a match.**\n   - Opportunities for rerepresentation are the ways a mapping can violate structure-mapping's constraints.\n   - \"Are there other opportunities for rerepresentation beyond those listed here? Our analysis suggests not. The only constraint of structure-mapping theory we have not exploited is systematicity. But systematicity is a preference... Since the opportunities described in Table 1 exhaust the constraints of structure-mapping, we conclude that this set is complete.\"\n   - And: \"Because we were able to derive the kinds of opportunities directly from the theoretical constraints of structure mapping, we claim the set we propose here completely characterizes them.\"\n   - Table 1 (garbled in extraction, reconstructed from the surrounding prose). Rows are the constraint violated, columns are whether parallel connectivity is also violated:\n     - **Holes**: identicality fails, and an argument misaligns. Example: walk versus run as antecedents of identical cause relations.\n     - **Gulches**: identicality fails at root statements only. \"Such statements in the base show up as candidate inferences of the match.\"\n     - **Rivals**: 1:1 violated, causing a structural inconsistency.\n     - **Leftovers**: 1:1 conflict leaves a kernel outside the merged mapping.\n   - Rerepresentation methods, by contrast, are \"relatively open\": \"the set of rerepresentation methods is relatively open.\"\n\n2. **Falkenhainer et al. 1989 / SME.**\n   - Every item in the base and target either enters a gmap's correspondences or does not.\n   - A base root that overlaps a gmap but lies outside it yields a candidate inference. A base root that does not overlap yields nothing, and the diameter example in §2 is the worked case.\n   - Reverse candidate inferences run the same procedure target-to-base (SME4 supplement).\n   - That gives a partial partition of unmatched material: connected-to-mapping versus not.\n   - Alignable differences have no separate slot in SME. They appear as matched items with non-identical fillers, via Markman & Gentner.\n\n3. **Markman & Gentner 1993.** Differences are split into AD and NAD, so (matched) + AD + NAD plus candidate inference covers matched/unmatched-connected/unmatched-unconnected. They do not claim exhaustiveness explicitly. I did not find the phrase.\n\n4. **Hummel & Holyoak 2003** explicitly decline exhaustiveness for their own phenomena. They treat evaluation and adaptation as outside scope.\n\nHow your five categories map onto what I found:\n\n| Category | Closest term in the literature |\n|---|---|\n| matched | Correspondences / commonalities |\n| unmatched but connected | Candidate inference (base side); reverse candidate inference (target side); Yan's gulch |\n| matched role, different filler | Alignable difference |\n| mismatched but should match | Yan's hole; leftovers and rivals are the 1:1 analogues |\n| in no role | Nonalignable difference; base items not intersecting the gmap |\n\nCaveats:\n- The first row is exhaustive over a base-side partition only given SME's definition. Others, e.g. LISA, define the mapping differently.\n- \"Mismatched but should match\" is not definable inside a single mapping. It requires a standard external to the mapping (identicality plus minimal ascension, or task demands).\n- An asymmetry worth recording: \"matched-role-different-filler\" is only defined on non-identical but alignable items. Yan et al. note that non-identical functions can be aligned, as can any pair of entities, but non-identical relations cannot unless rerepresented.\n\n---\n\n## What bears on WHERE A DIFFERENCE LANDS\n\n### Summary of what each source implies\n\n| Source | Where a difference lands | Quote |\n|---|---|---|\n| Gentner & Markman 1997; SME (Falkenhainer 1989) | **Carrying.** Differences are salience and weighting effects. The mapping and the candidate inference are the only things evaluated. Validity of a candidate inference is \"left to other modules\". | \"Each candidate inference is only a surmise, which must be tested by other means.\" |\n| Gentner & Markman 1997 (categorization section) | **Membership** is a separate use of comparison, \"categorizing (selecting best match from memory)\" and \"discovering new category (abstracting common system)\". | Figure 5, bats and birds |\n| Kepler example, 1997 | One disanalogy (no eclipse) blocks both a carrying and an identity claim. The analogy survives. | \"used this disanalogy to conclude that the motive power cannot be the same thing as the sun's light\" |\n| Weitzenfeld 1984 / Davies & Russell | **Carrying**, via the determination rule. Only differences on determining variables matter. Differences outside them do not. [criterion unverified verbatim] | \"differs from the target only on variables whose effect can be calculated\" (lecture-note paraphrase, not Weitzenfeld) |\n| Bartha 2010, 2020 | **Carrying.** The rule (prior association) is untouched. A critical disanalogy blocks its generalization to the target. Non-critical differences only weaken. | \"No critical disanalogy: no critical factor belongs to the negative analogy.\" |\n| Juthe 2015 | **Carrying**, as failure of the one-to-one counterpart for a determining element. The vertical relation is particular to the analogue. | \"each element in the Analogue that has a vertical determining relation to the Assigned-Predicate also has a horizontal one-to-one correspondence to a counterpart element\" |\n| Walton, Reed & Macagno 2008 (CQ1) | The *similarity premise* (the correspondence). It weakens rather than refutes. CQ3 adds a rival. No premise states a rule. | \"differences between C1 and C2 that would tend to undermine the force of the similarity cited\" |\n| Lamond 2005 (reason model) | **Carrying.** The case is within the rule's scope (membership holds) and the rule is not refuted. A new reason defeats the outcome. | \"may clearly fall within the scope of the earlier ratio... but the later court may decline to reach the result C\" |\n| Horty 2011 | **Carrying by trumping.** Rule remains applicable (membership as premise-subset); a new rule with opposite outcome outranks it. | \"the court distinguishes the binding rule r... whenever it either follows or formulates another rule r′ supporting the opposite outcome\" |\n| Lamond's rule-model target; Mullins 2026 | **Membership/rule** when a precedent fact absent from the new case is treated as necessary. Mullins calls it \"constrained overruling\", not distinguishing. | \"restrictive reinterpretation should be viewed as a form of constrained overruling of precedent\" |\n| Shelley, via Juthe's review [secondary] | Separates \"comparison fails\" (false analogy, misanalogy) from \"comparison stands but is overridden\" (disanalogy, counter-analogy). | \"the analogy is accepted as a correct analogy, but the counterarguments operate by overriding the original analogy\" |\n| Nersessian | Differences feed model revision. Generic abstraction defines the class. | \"key differences playing a significant role in further model generation\" |\n| Hummel & Holyoak 2003 | Schema induction is intersection discovery, so what is *not* shared is dropped from the schema rather than refuting it. | \"a kind of intersection discovery at the level of complete propositions\" |\n\n### Reading across them\n\n1. **Default landing is the carrying.** Every philosophical and argumentation source (Bartha, Juthe, Weitzenfeld/Davies, Walton) places an effective difference on the horizontal correspondence for a determining element. The source's rule is held fixed. Only the transfer is blocked.\n2. **The rule is attacked by a different move.** Nothing in this literature has a difference attack the shared generalization directly. In the legal literature, narrowing the rule (restrictive reinterpretation) is the only move that does, and Mullins classes it as overruling. Counter-analogy (Walton CQ3, Shelley) attacks via a rival case, not via a difference.\n3. **Membership is separately theorised.** The cognitive literature supplies it only as an application of comparison (categorization, schema abstraction). Argumentation theory supplies it as Guarini's classificatory analogies. Legal theory splits on it: for the rule model, distinguishing is a scope (membership) move. For the reason model, membership is conceded and a competing reason defeats the outcome.\n4. **Relevance is the pivot.** Gentner's account (via SME) puts relevance and validity outside the structural mapping. The determination approaches (Weitzenfeld, Juthe) and Bartha's \"critical factors\" put relevance inside the argument, as a vertical relation. A difference only matters if it touches a factor that plays a role in the vertical relation (Bartha: \"essential role in the prior association\"). Otherwise it is \"secondary\" (weakening) or irrelevant. Holyoak & Thagard's pragmatic centrality is the cognitive counterpart, a pressure rather than an absolute.\n5. **Alignable versus nonalignable is not the same cut as relevant versus irrelevant.** Alignable differences are salient because they sit inside the common system, but nothing verified says they refute. Whether one is a *critical* disanalogy is decided by the vertical relation, not by alignability.\n\n### Not verified, for your corpus notes\n- Weitzenfeld's determining-structure criterion, verbatim. Paywalled.\n- Bartha's own wording of \"critical disanalogy\" in the 2010 book. I have only his 2020 restatement and the NDPR review.\n- Walton et al. 2008 p. 315, directly. I have five consistent secondary reproductions.\n- Ashley's HYPO/CATO primary sources, and Rigoni's treatment of distinguishing.\n- Page numbers for Gentner & Markman 1997 quotes, and Markman & Gentner 1993 page range.\n- Govier primary text.\n- Whether the Lamond/Sharp/Graham facts in the primary law reports match Lamond's description.\n- The Mason v. Jack Daniel Distillery reporter citation (from memory)."}