{"id":"c7fec5bc-46d1-4985-a0e4-6d87dfa6b139","arxiv_id":"2506.11660","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Pareto-efficient improvements over the Deferred Acceptance algorithm cannot fix its worst-case inequality, rank inefficiency, or school segregation.","lead":"This economics paper proves that mechanisms designed to fix the inefficiency of the Deferred Acceptance school-choice algorithm cannot fix its inequality, rank inefficiency, or segregation. It gives worst-case bounds and shows that when advantaged students always outrank marginalized ones, every school's demographic mix is locked in exactly as DA left it.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Segregation theorem requires nested group priorities; without them a Pareto-efficient stable-dominating mechanism can change school composition, so the abstract's unconditional claim overreaches.","rationale":"The reader's weakest-assumption identification is correct, and I agree: Theorem 1 as stated is valid only under the extreme priority stratification. My constructed example shows the advertised claim in the abstract is false in general; this is a scope-of-claim concern, not an internal inconsistency of the theorem. Under the stated assumption, the cycle argument in Propositions 3 and 4 goes through, so I would not reject the paper. The presentation should qualify the abstract and conclusion by the nested-priority assumption and note that real priority systems with walk-zone, sibling, or lottery priorities interleave groups, so the segregation-persistence result does not automatically carry over. The reader's other points, including Proposition 1's overreach and undefined null-school ranks, are also valid and reinforce CONDITIONAL, but the priority assumption is the most load-bearing because it gates the paper's headline policy message. No change to the reader's CONDITIONAL verdict is needed.","tokens_in":10844,"tokens_out":30697,"duration_ms":304702,"concrete_test":"Implement the 4-student, 4-school profile above with full preference lists filled consistently, and verify: (i) DA outputs A1-s1, M-s2, A2-s3, X-s4; (ii) the alternative matching A1-s2, M-s1, A2-s3, X-s4 weakly Pareto-dominates DA and admits no Pareto-improving cycle; (iii) the number of advantaged and marginalized students at s1 and s2 differs between the two matchings. If all three hold, the example refutes composition invariance without the Section 4.2 nested-priority assumption.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1's composition-invariance claim is conditional on the Section 4.2 assumption that every marginalized student has lower priority than every advantaged student at every school. The abstract and the policy conclusion ('any policy aimed at reducing school segregation must rely on interventions beyond Pareto improvements') state the result without this caveat. The caveat is indispensable. With interleaved priorities, a Pareto-efficient stable-dominating mechanism can change group counts. Example: advantaged A1,A2; marginalized M,X; schools s1-s4, capacity 1. Preferences: A1: s2>s1>s3>s4; M: s1>s2>s3>s4; A2: s3>...; X: s2>s4>s3>s1. Priorities: s1: A1>M>X>A2; s2: M>X>A1>A2; s3: A2>...; s4: X>... . DA yields A1-s1, M-s2, A2-s3, X-s4. The matching A1-s2, M-s1, A2-s3, X-s4 weakly Pareto-dominates DA and is Pareto-efficient, yet s1 changes from advantaged-only to marginalized-only and s2 changes from marginalized-only to advantaged-only. Thus the exact-composition conclusion of Theorem 1 fails once the nested-priority assumption is dropped.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies mechanisms that weakly Pareto-dominate the student-proposing Deferred Acceptance (DA) outcome, called stable-dominating mechanisms. It makes two main claims. First, even when DA is Pareto-efficient, stable-dominating mechanisms can be a factor n/2 worse than the Rawlsian and rank-minimizing mechanisms in terms of the worst-off student's rank and the average rank (Proposition 1). Second, under the assumption that students are divided into advantaged and marginalized groups with every marginalized student having lower priority than every advantaged student at every school, any stable-dominating mechanism preserves the exact advantaged/marginalized composition of every school, so fully segregated schools under DA remain fully segregated (Theorem 1, Propositions 2-4). The paper concludes that efficiency adjustments cannot fix DA's rank-inefficiency, inequality, or segregation.","tokens_in":11030,"tokens_out":28435,"duration_ms":320013,"significance":"The paper addresses a timely policy question, and the core negative results, if properly qualified, are interesting. The tight n/2 examples are clean and make the point that Pareto efficiency is compatible with severe rank-inequality and inequality. The segregation invariance under nested priorities is a crisp, non-obvious result and complements empirical work on DA segregation. The proof strategy via the envy digraph is promising. However, the current manuscript's abstract and Theorem 1 overreach beyond the model's assumptions, and Proposition 1's proof contains gaps. These issues are fixable without changing the central ideas.","major_comments":[{"comment":"The abstract states that the demographic composition of every school is perfectly preserved under any Pareto-efficient mechanism that dominates DA, without the qualifying assumption introduced in Section 4.2 that every marginalized student has lower priority than every advantaged student at every school. This assumption is indispensable. Consider four unit-capacity schools s1-s4 with students A1,A2 (advantaged) and M,X (marginalized). Preferences: A1: s2≻s1≻s3≻s4; M: s1≻s2≻s3≻s4; A2: s3≻s1≻s2≻s4; X: s2≻s4≻s3≻s1. Priorities: s1: A1≻M≻X≻A2; s2: M≻X≻A1≻A2; s3: A2≻A1≻M≻X; s4: X≻A1≻A2≻M. DA assigns A1-s1, M-s2, A2-s3, X-s4. The matching A1-s2, M-s1, A2-s3, X-s4 weakly Pareto-dominates DA and is Pareto-efficient, yet s1 changes from advantaged-only to marginalized-only and s2 from marginalized-only to advantaged-only. The theorem and the policy conclusion that any segregation-reducing policy must go beyond Pareto improvements must therefore be explicitly conditioned on the nested-priority assumption.","section":"Abstract; §4.2, Theorem 1"},{"comment":"The upper-bound proof for RkI is not valid as written. The sentence 'The sum of ranks in RM(P) cannot be any smaller than n+1' is false for problems with m<n; for example, with m=2 and n=10, both students can receive their first choice, giving a sum of 2. The follow-up 'as otherwise M†(P)=RM(P)' is also not established, since a Pareto-efficient mechanism can have a strictly larger sum of ranks than the rank-minimizing mechanism. Consequently the claimed bound RkI(M′;n)≤n/2 does not follow from the supplied argument. The Rawlsian part also needs a sentence explaining why an arbitrary mechanism M* (not only DA) must assign every student to their top choice when the Rawlsian denominator is 1. Please supply a correct proof or restrict the statement to the case m=n where the construction applies.","section":"§4.1, Proposition 1"},{"comment":"The model allows a null school s∅ for unassigned students, but the rank function rk_i is defined only on the finite set S with values 1..n. Therefore max_i rk_i[M_i(P)] and sum_i rk_i[M_i(P)] are undefined for any problem in which some student is assigned to s∅. Either restrict P_{m,n} to instances with no unassigned students or extend the rank function to include s∅ with a suitably large rank, and re-verify Proposition 1 under that convention.","section":"§3, Eqs. (1)-(2)"},{"comment":"The proof of Proposition 2 contains a non sequitur: after showing that no rejection occurs at school s in round t, it states 'Then school s does not reject any student at time t′≤t.' Earlier rejections at s are not ruled out by the argument, and an extra monotonicity argument is needed to exclude them. Since Theorem 1's proof explicitly invokes Proposition 2, this gap matters. In addition, the proof of Theorem 1 jumps from 'advantaged students at mixed schools are unimprovable' and 'some marginalized student is unimprovable' to exact count preservation; the authors should either provide a formal cycle-decomposition argument (which could use Proposition 3 and Lemma 2 directly) or rewrite the proof.","section":"§4.2, Propositions 2 and Theorem 1"}],"minor_comments":[{"comment":"The text refers to 'Table 2 below' for the six-student motivating example, but the table is labeled TABLE 1; the later eight-student example is labeled TABLE 2.","section":"§1, Table 1"},{"comment":"Footnote 4's statement of school priorities is hard to parse; please write it in explicit quantifier form and specify the remaining preferences and priorities, or state clearly that they are arbitrary.","section":"§4.1, Footnote 4"},{"comment":"Lemma 3 is load-bearing for the segregation result and is attributed in part to the authors' companion paper (Ortega et al., 2025); please include a self-contained proof or ensure the companion manuscript is publicly available.","section":"§3.1, Lemma 3"},{"comment":"The sentence 'No advantaged student can access this school through any cycle (nor would like to)' is unclear; the parenthetical 'nor would like to' should be justified or removed.","section":"§4.2, Proof of Theorem 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is likely correct in its qualified form, and the main ideas are publishable. The most important fixes are to qualify the segregation claim in the abstract and Theorem 1, and to repair Proposition 1's proof. The heavy reliance on the companion paper for Lemma 3 should be checked; if the companion is not yet available, the editor may want a self-contained lemma. The abstract's unconditional phrasing should be toned down before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper's two headline results are real, but the written claims are wider than the proofs. Proposition 1 states that any mechanism's Rawlsian inequality is at most n/2, but the proof only works for Pareto-efficient mechanisms. As stated, it's false: a mechanism that dumps everyone at their worst school while a perfect matching exists gives ratio n. Fix the statement and the proof is fine.\n\nThe segregation theorem is the more interesting piece. With the assumption that marginalized students have lower priority than advantaged students at every school, the proof is convincing: the envy digraph separates along group lines, so no trading cycle crosses the divide, and the occupancy lemmas do the rest. But that assumption is doing all the work. The abstract says 'the demographic composition of every school is perfectly preserved' without the priority caveat. That's not true. The stress-test example is correct: with interleaved priorities, a Pareto-efficient stable-dominating matching can swap an advantaged and marginalized student across schools, flipping two schools' composition. So the theorem is conditional, and the policy conclusion should be stated as conditional too.\n\nWhat's genuinely new: the tight n/2 ratios for the Rawlsian and rank-inefficiency of stable-dominating mechanisms, and the invariance result under nested priorities. The paper is well-grounded in the literature, and the motivating example in Section 4.2 is illuminating: it shows a stable-dominating matching preserves segregation while an alternative Pareto-efficient (but less stable) matching reduces it. That's a real insight for the Flanders EADA discussion.\n\nMinor gaps: the null school is never assigned a rank; the construction in Proposition 1 covers m=n but the definitions take suprema over all m. Both are easy to patch.\n\nBottom line: the core math is sound under the stated assumptions. The paper needs a careful revision of Proposition 1's statement and the abstract's unconditional segregation claim. I'd send it to a serious referee; it deserves a fair shot. If the authors fix these, it's a publishable contribution to the school-choice literature.","headline":"Real results under narrow assumptions, but Proposition 1's statement outruns its proof and the abstract's segregation claim omits the priority condition that makes it true.","tokens_in":11623,"tokens_out":5335,"would_cite":true,"duration_ms":56320,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91B68","91B32"],"pacs":[],"model":"deepseek-v4-flash","headline":"Segregation survives every Pareto-efficiency fix to school choice","keywords":["school choice","deferred acceptance","stable-dominating mechanisms","segregation","rank-efficiency","Rawlsian inequality","unimprovable students","Pareto efficiency"],"falsifier":"Run a stable-dominating mechanism such as EADA on a school-choice problem where a marginalized student has a walk-zone or lottery priority that outranks some advantaged student at a desirable school; if the school's share of marginalized students changes relative to DA, the tiered-priority assumption is the operative condition, not the mechanism class. Alternatively, construct a stable-dominating matching that changes the demographic composition of a school under tiered priorities—this would directly contradict Theorem 1.","tokens_in":10579,"feed_emoji":"🏫","tokens_out":2914,"duration_ms":32745,"temperature":0.7,"pith_summary":"This paper asks whether mechanisms designed to fix the Pareto-inefficiency of the Deferred Acceptance (DA) algorithm also fix its distributional problems, and finds they cannot. The authors prove that any Pareto-efficient mechanism that weakly dominates DA preserves the exact number of advantaged and marginalized students at every school, so a school that is fully segregated under DA stays fully segregated. They also prove tight worst-case bounds: both Rawlsian inequality and rank-inefficiency can be as bad as half the number of schools under these mechanisms, even when DA itself is Pareto-efficient. The practical consequence is that efficiency-adjusted admissions reforms, however useful for student welfare, cannot by themselves reduce school segregation or guarantee equitable outcomes.","feed_headline":"Segregation survives every Pareto-efficiency fix to school choice","feed_subtitle":"Efficiency-adjusted admissions preserve each school's demographic mix, and worst-case rank loss stays at half the number of schools.","key_machinery":"The proof machinery is the envy digraph $G^{DA}(P)$ induced by the DA matching, in which each student points to students whose assignments they prefer. A student is unimprovable exactly when they lie on no cycle in this digraph (Lemma 3). The paper shows that when marginalized students have lower priority than every advantaged student at every school, no trading cycle can contain both groups, because an advantaged student envying a marginalized student's school would form a blocking pair with that school. Advantaged students at mixed schools are shown to be unimprovable, and at least one marginalized student is unimprovable, so all trades occur within groups, preserving each school's group composition.","core_discovery":"The central claim is that stable-dominating mechanisms—mechanisms that produce Pareto-efficient matchings weakly dominating the DA outcome—are structurally incapable of changing the demographic composition of any school when students are partitioned into advantaged and marginalized groups with tiered priorities. Theorem 1 states that the number of advantaged and marginalized students accepted to each school under any stable-dominating mechanism is constant, hence fully segregated schools under DA remain fully segregated. Proposition 1 shows that the worst-case Rawlsian inequality ratio and the worst-case rank-inefficiency ratio for such mechanisms are both exactly $n/2$, where $n$ is the number of schools, and these ratios are tight; the motivating example shows a Pareto-efficient DA allocation where the worst-off student gets rank $n$ while an alternative Pareto-efficient allocation gives everyone rank at most 2.","pith_inferences":["The segregation-preservation result depends on the strict tiered-priority assumption; in real systems with walk-zone, sibling, or lottery priorities that interleave groups, trading cycles could cross the advantaged-marginalized divide and alter school composition, so the invariance is a property of tiered priorities rather than of stable-dominating mechanisms in general.","The same composition-invariance argument extends to any number of priority tiers, implying that efficiency adjustments cannot reduce stratification in multi-tier settings either, and only deliberate priority violations—such as reserved seats—are likely to integrate schools.","A direct empirical test would be to run a stable-dominating mechanism like EADA on a school-choice dataset with interleaved priorities (for example, where some marginalized students live in the walk zone of a desirable school) and check whether the share of marginalized students at mixed schools changes; this would separate the tiered-priority mechanism from the general conclusion.","The $n/2$ worst-case ratios suggest that the cost of stability and Pareto-efficiency is unbounded in the number of schools, so policy debates about DA versus alternatives should focus on average-case behavior and on explicit equity constraints rather than on worst-case guarantees."],"forward_implications":["Any stable-dominating mechanism, including EADA, DA-endowed top trading cycles, and MIDA, leaves the number of advantaged and marginalized students at each school exactly as DA did.","Schools that are fully segregated under DA—admitting only advantaged or only marginalized students—remain fully segregated under every Pareto-efficient mechanism that dominates DA.","The worst-case Rawlsian inequality of stable-dominating mechanisms is $n/2$ times the first-best, where $n$ is the number of schools, and this bound is tight.","The worst-case rank-inefficiency of stable-dominating and Pareto-efficient mechanisms is also $n/2$ times the rank-minimizing benchmark, so efficiency adjustments cannot guarantee good average ranks in the worst case.","Reducing segregation in school choice requires interventions that go beyond Pareto improvements, such as changing priority structures or implementing quota or reserve policies."],"supporting_citations":[{"why":"Introduces stable-dominating matchings and supplies Lemmas 1 and 2 on unimprovable students and constant school occupancy.","marker":"Alva and Manjunath (2019)"},{"why":"Co-authors Lemmas 1 and 2, and Lemma 3, connecting unimprovability to cycles in the DA envy digraph.","marker":"Tang and Yu (2014)"},{"why":"Provides EADA, the main example of a stable-dominating mechanism, and is credited in Lemma 3.","marker":"Kesten (2010)"},{"why":"Companion paper that introduced the DA envy digraph and is cited for Lemma 3; also supplies the result that most students can trade to better schools in large random markets.","marker":"Ortega et al. (2025)"},{"why":"Defines the deferred acceptance algorithm, the baseline mechanism the paper studies.","marker":"Gale and Shapley (1962)"}],"fun_headline_variants":["Pareto fixes can't alter school segregation","Efficient school choice keeps segregation intact","DA's rank waste persists even when efficient","Stable fixes preserve school demographics","Pareto efficiency doesn't cure inequality"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The segregation theorem assumes that every marginalized student has lower priority than every advantaged student at every school; if real priorities allow the groups to interleave, the conclusion that trading cycles never mix groups can fail.","fun_headline_variants_meta":{"raw":{"variants":["Pareto fixes can't alter school segregation","Efficient school choice keeps segregation intact","DA's rank waste persists even when efficient","Stable fixes preserve school demographics","Pareto efficiency doesn't cure inequality"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":9.9e-05,"raw_usage":{"total_tokens":941,"prompt_tokens":799,"completion_tokens":142,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":415,"completion_tokens_details":{"reasoning_tokens":78}},"tokens_in":415,"tokens_out":142,"duration_ms":2297,"temperature":1.0,"reasoning_tokens":78,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T04:05:01.833087+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a stable-dominating mechanism such as EADA on a school-choice problem where a marginalized student has a walk-zone or lottery priority that outranks some advantaged student at a desirable school; if the school's share of marginalized students changes relative to DA, the tiered-priority assumption is the operative condition, not the mechanism class. Alternatively, construct a stable-dominating matching that changes the demographic composition of a school under tiered priorities—this would directly contradict Theorem 1.","supporting_citations":[{"cited_title":"Stable-dominating rules","cited_arxiv_id":null,"evidence_quote":"Introduces stable-dominating matchings and supplies Lemmas 1 and 2 on unimprovable students and constant school occupancy."},{"cited_title":"A new perspective on kesten’ s school choice with consent idea","cited_arxiv_id":null,"evidence_quote":"Co-authors Lemmas 1 and 2, and Lemma 3, connecting unimprovability to cycles in the DA envy digraph."},{"cited_title":"Identifying and quantifying (un)improvable students","cited_arxiv_id":null,"evidence_quote":"Companion paper that introduced the DA envy digraph and is cited for Lemma 3; also supplies the result that most students can trade to better schools in large random markets."},{"cited_title":"College admissions and the stability of mar- riage","cited_arxiv_id":null,"evidence_quote":"Defines the deferred acceptance algorithm, the baseline mechanism the paper studies."}],"review_version":1}