{"id":"7896b992-90b0-40c6-99b1-982ac466fc9d","arxiv_id":"1908.08406","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"SCF2, a new abstract argumentation semantics, is defined and proven to satisfy Directionality, Irrelevance of Necessarily Rejected Arguments, and Strong Completeness Outside Odd Cycles, with at least one extension for every argumentation framework.","lead":"This paper proposes a new semantics, SCF2, for deciding which arguments in an argumentation graph humans would accept, and proves it satisfies two new principles that no existing standard semantics satisfies together. A generalist should read it to see how formal argumentation theory tries to match human judgment by combining normative principles with the authors' own cognitive experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3(b), on which Theorem 8's INRA claim entirely rests, is only sketched and its adaptation from part (a) changes the antecedent from 'attacks a' to 'a not in E'; if the omitted induction is wrong, SCF2's central combination of principles is unproved.","rationale":"The paper's central formal claim is that SCF2 uniquely combines Directionality, INRA, and SCOOC. Directionality and SCOOC have at least sketched direct proofs (Theorems 6 and 7), but INRA is not proved directly: Theorem 8 is a one-line appeal to Lemma 3(b), which itself is delegated to a proof sketch with an omitted induction. This is exactly the property that distinguishes SCF2 from CF2, which fails INRA. The reader's verdict of CONDITIONAL is appropriate, and my concern is the same gap, so I recommend UNCHANGED, subject to the missing proof being supplied. The negative theorems (2 and 4) are supported by explicit counterexamples, and the definition of SCF2 is precise enough that the gap is checkable. The empirical overclaim and the 'no semantics from the literature' wording are presentation issues, but they are not load-bearing for the formal claim. Credit is due for the precise definitions and for the explicit counterexample-based separation of the seven related semantics in Table 1.","tokens_in":15860,"tokens_out":8128,"duration_ms":83856,"concrete_test":"Run an exhaustive computational check over all AFs up to n=6 or n=7 arguments verifying Theorem 8 directly: for every AF F and every argument a that is attacked by every SCF2 extension of F, the set of SCF2 extensions of F must equal the set of SCF2 extensions of F−a. The recursive definitions in Definitions 5, 7, and 17 are decidable by brute force, so this reveals any small counterexample. Independently, complete the missing part of Lemma 3(b) in a proof assistant or in a written companion proof that handles the splitting of the SCC containing a; if the 'continue as in part (a)' step cannot be discharged, the concern is confirmed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"SCF2's satisfaction of INRA (Theorem 8) is delegated to Lemma 3(b), and Lemma 3(b)'s proof is not actually the promised 'continue as in part (a)' step; it contains an unproved adaptation. First, the proof starts from 'for every E∈nsa(scc(σ))(F), a∉E', but INRA's antecedent is 'every E attacks a'. These are not equivalent in general, and the part-(a) argument uses SCC-semi-richness to rule out the single-SCC case by producing an extension that does not attack a; that step requires the stronger antecedent. Second, after setting F′=NSA(F), the proof says to continue with F′ in place of F and stops. The hard part of part (a) is precisely the induction over the SCC partition after a is deleted, because a's SCC can split; that case analysis is what must be re-verified for nsa(scc(σ)), where extensions of the reduced framework are compared at the scc level. The paper itself calls the SCC-splitting complication 'the main difficulty' (Section 3, Lemma 3 proof), so this is not a minor ellipsis. If Lemma 3(b) fails, Theorem 8 fails, and SCF2's claimed conjunction of Directionality+INRA+SCOOC is not established.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a new abstract argumentation semantics, SCF2, defined as nsa(scc(SCOOC-naive)), and claims it satisfies three principles: Directionality, Irrelevance of Necessarily Rejected Arguments (INRA), and Strong Completeness Outside Odd Cycles (SCOOC). The authors introduce INRA and SCOOC, show which of several existing semantics satisfy them, and prove that SCF2 satisfies all three, that every argumentation framework has at least one SCF2 extension, and that among seven related naive-based semantics only SCF2 satisfies all three principles (Table 1, Theorems 5-8). They also discuss two cognitive studies by Cramer and Guillaume that they interpret as providing preliminary empirical support for SCF2. The theoretical part is largely detailed, but a load-bearing proof step in Lemma 3(b) is only sketched, and the abstract's claim that no semantics from the literature satisfies the two new principles overreaches the considered list of semantics.","tokens_in":16189,"tokens_out":7610,"duration_ms":70275,"significance":"If the central theorems are correct, SCF2 is a genuine new semantics that provably combines Directionality, INRA, and SCOOC, and it comes with a systematic comparison of seven related semantics in Table 1. The paper also makes a falsifiable empirical prediction that SCF2 matches human judgments better than other semantics, which is a strength given the paper's stated goal of combining normative and descriptive approaches. The theoretical contribution is significant for the principle-based analysis of argumentation semantics, though its scope is narrower than the abstract suggests. The main weakness is that the proof of Lemma 3(b), on which Theorem 8 (SCF2 satisfies INRA) entirely depends, is not fully carried out; this is a fixable but load-bearing gap.","major_comments":[{"comment":"The proof of Lemma 3(b) is incomplete and begins with the wrong antecedent. The proof states 'let a ∈ Ar be an argument such that for every E ∈ nsa(scc(σ))(F), a /∈ E', but the INRA principle (Definition 11) requires 'every E attacks a'; these are not equivalent, and the later use of SCC-semi-richness to rule out the single-SCC case relies on the stronger 'does not attack a' condition. After defining F′ = NSA(F), the proof says only 'we continue as in the proof of part (a), just with F′ in place of F' and stops. Part (a)'s hard step is the induction over the SCC partition after deleting a, because a's SCC can split; the manuscript itself calls this 'the main difficulty' in the proof of Lemma 3. Since Lemma 3(b) is the sole support for Theorem 8, this omitted argument is load-bearing; the proof must be completed (or Theorem 8 proven directly) before the claimed conjunction of Directionality, INRA and SCOOC is established.","section":"§3, Lemma 3(b)"},{"comment":"The abstract claims that 'no semantics from the literature satisfies both' of the two new principles, but the body only shows this for the nine semantics listed in Section 1 plus nsa(CF2), and Table 1 covers seven variants of naive semantics. Many other extension-based semantics from the literature—including ideal, eager, and semantics definable in SESAME [6]—are not considered. The novelty claim should be restricted to the considered semantics, or the survey must be broadened; as written, the claim overreaches the evidence.","section":"Abstract and §1"},{"comment":"The empirical support for SCF2 over CF2 is not statistically significant after 12 of 61 participants are excluded, and the decisive comparison rests on a single six-cycle framework in which SCF2, stage2, preferred and semi-stable coincide. The conclusion appropriately calls the findings preliminary, but the abstract's statement that the results 'suggest that SCF2 is more in line with the judgments of participants than any existing semantics' is not supported by the reported data. In addition, the cognitive studies are conducted by the paper's first author; the potential conflict should be disclosed in a journal version.","section":"§5"}],"minor_comments":[{"comment":"The notation 'a /∈ E' in the proof should read 'E attacks a' to match Definition 11; the current wording makes the proof appear to prove a different statement.","section":"§3, Lemma 3(b)"},{"comment":"The line 'Since F′ = NSA(F′)' should be 'Since F′ = NSA(F)', and the expression 'F′ Ar\\U \\ DF′(E1)' should use the restriction notation F′|(...) to avoid ambiguity.","section":"§4, Theorem 6 proof"},{"comment":"The expression '|SCCs(F) = 1|' should be '|SCCs(F)| = 1', and 'attakers' should be 'attackers'.","section":"§4, Lemma 4 proof"},{"comment":"The phrase 'they disconsider them' should be 'they exclude them' or 'they discard them', and 'can by found' should be 'can be found'.","section":"§5"},{"comment":"The text cites 'Rahwan et al. [18]' for the behavioral experiments on reinstatement, but reference [18] is the edited volume 'Argumentation in Artificial Intelligence'; the relevant cognitive-science study appears to be reference [17], so the citation should be corrected.","section":"§5 and References"},{"comment":"The paper should mention that the binomial tests are not corrected for multiple comparisons, and it would be helpful to report effect sizes or confidence intervals for the non-significant SCF2-versus-CF2 comparison.","section":"§5"}],"recommendation":"major_revision","confidential_remarks":"The central theoretical idea is attractive, but the proof of Lemma 3(b) must be completed before the paper's main claim is fully supported; the current sketch is not sufficient for a journal publication. The abstract's 'no semantics from the literature' claim should be narrowed to the considered semantics. The empirical section is preliminary, and the fact that the studies were conducted by the first author should be disclosed in the published version. These issues are fixable within the scope of the manuscript, so major revision rather than rejection is appropriate."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, here's my read.\n\nThe paper does two things: it introduces two new semantics principles (INRA, SCOOC) and a new semantics SCF2 = nsa(scc(SCOOC-naive)) that provably satisfies both plus Directionality. The principles are well-motivated from the debate about normative vs descriptive evaluation, and the systematic 7-semantics comparison in Table 1 is a clean way to show why SCF2's specific combination matters. The proofs of Directionality and SCOOC for SCF2 look fine, and Lemma 4 (existence of extensions) is a real piece of work. So there is genuine formal content here.\n\nThe soft spot is exactly the one you'd worry about. Theorem 8, which says SCF2 satisfies INRA, is delegated to Lemma 3(b). And Lemma 3(b) is not proved. The proof sketch says 'we continue as in the proof of part (a), just with F′ in place of F,' but the antecedent changes: part (a) assumes every extension attacks a; part (b) only assumes a is not in any extension. Those are not equivalent, and the part (a) argument relies on the stronger condition to show a must be in a later SCC. Without that, the omitted induction over SCC partitions after deleting a is genuinely load-bearing. The paper even tells you the SCC-splitting complication is 'the main difficulty.' This is not a minor ellipsis. As far as I can tell, the gap is real.\n\nThere are two smaller issues. The 'no semantics from the literature satisfies both' claim only checks nine or ten named semantics, so it should be phrased relative to that list. And the empirical support is from the first author's own prior studies; for the key SCF2-vs-CF2 difference the results are not significant, and after excluding 12 of 61 participants the remaining analysis is underpowered. It's honest about that, but it shouldn't be sold as strong evidence.\n\nOverall: the central idea is worth taking seriously. The paper is for people working on SCC-recursive argumentation semantics and principle-based evaluation. It deserves a real peer-review slot — the referee should demand a complete proof of Lemma 3(b) and a more careful empirical framing. I'd read it as a strong technical report in need of one solid revision.","headline":"SCF2 is a genuinely new semantics with a nice principle-based motivation, but the proof that it satisfies INRA rests on a sketched Lemma 3(b) that needs to be fully worked out before the central claim is safe.","tokens_in":16641,"tokens_out":3381,"would_cite":false,"duration_ms":31888,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper defines SCF2, an argumentation semantics that provably satisfies two newly introduced principles — Irrelevance of Necessarily Rejected Arguments and Strong Completeness Outside Odd Cycles — plus the established Directionality…","keywords":["abstract argumentation","argumentation semantics","SCF2","INRA","SCOOC","Directionality","SCC-recursive semantics","cognitive study"],"falsifier":"Run a brute-force search over all finite argumentation frameworks with up to, say, five arguments, looking for a framework in which some argument is attacked by every SCF2 extension yet deleting that argument changes the set of extensions. A single such framework refutes Theorem 8; if none exists, the open gap in Lemma 3(b) becomes a clear target for a full rewrite of the proof.","tokens_in":15623,"feed_emoji":"🧠","tokens_out":8581,"duration_ms":76548,"temperature":0.7,"pith_summary":"The paper sets out to find an argumentation semantics that captures what humans consider a rational judgment about which arguments to accept. It introduces two normative principles — INRA, meaning a universally rejected argument can be deleted without changing the verdict, and SCOOC, meaning an argument whose attackers are all rejected must be accepted unless it sits inside an odd cycle — and shows that none of the examined existing semantics satisfies both. It then defines SCF2, a variant of CF2 that first removes self-attacking arguments and then recursively applies a strengthened naive semantics inside each strongly connected component, and proves SCF2 satisfies INRA, SCOOC, and Directionality while every framework still has at least one extension. The paper closes with two cognitive studies reporting that SCF2 tracks human acceptability judgments at least as well as, and on some frameworks better than, existing semantics.","feed_headline":"Only SCF2 passes three key tests of rational argument acceptance","feed_subtitle":"A CF2 variant that deletes self-attacking arguments and blocks odd-cycle paradoxes also matches human judgments.","key_machinery":"The central object is the composed semantic operator nsa(scc(SCOOC-naive)). 'nsa' deletes all self-attacking arguments before evaluation; 'scc' is the SCC-recursive scheme that splits a framework into strongly connected components and evaluates them in topological order, deleting arguments attacked by accepted sets from later components; 'SCOOC-naive' is the base semantics applied inside each single SCC, picking subset-maximal conflict-free sets that are strongly complete outside odd cycles. This composition is what carries the argument: nsa yields INRA, scc yields Directionality, and SCOOC-naive yields SCOOC, and the paper proves no other combination of these switches achieves all three.","core_discovery":"SCF2 is defined as nsa(scc(SCOOC-naive)): delete all self-attacking arguments, then run the SCC-recursive scheme that CF2 uses, but replace naive semantics inside each single SCC with SCOOC-naive semantics, which selects subset-maximal conflict-free sets that are strongly complete outside odd cycles. The paper proves that among seven semantics formed by switching the nsa, scc, and SCOOC-naive ingredients on and off, SCF2 alone satisfies Directionality, INRA, and SCOOC (Theorems 6–8 and Table 1), and that every finite AF has at least one SCF2 extension (Theorem 5). The proofs rest on a lemma showing that nsa(scc(σ)) satisfies INRA whenever σ is SCC-semi-rich, a condition it establishes for SCOOC-naive semantics. The empirical section reports that 27 coherent non-grounded participants in a cognitive study judged most frameworks in line with SCF2, including the six-cycle framework where CF2 accepts two opposite arguments but SCF2 does not.","pith_inferences":["The nsa-then-scc recipe is a general design template: any SCC-semi-rich base semantics could be turned into an INRA-satisfying recursive semantics in the same way, so the technique may outlive SCF2 itself.","INRA gives SCF2 a clean dynamic behaviour — universally rejected arguments can be dropped without revisiting earlier conclusions — which the paper does not explore but which matters for incremental argumentation systems.","The empirical case is still open: the study's improvement of SCF2 over CF2 and grounded semantics was not statistically significant, so a larger experiment with frameworks that separate these semantics is the decisive test of the descriptive claim.","A brute-force check for INRA counterexamples would settle whether the omitted part of Lemma 3(b) is a harmless abbreviation or a real gap."],"forward_implications":["Every finite argumentation framework has at least one SCF2 extension, so SCF2 never leaves a framework without an evaluation.","Deleting a universally rejected argument leaves the SCF2 verdicts unchanged, so such arguments can be omitted from the evaluation without affecting the outcome.","Within every SCF2 extension, an argument whose attackers are all rejected must be accepted unless it lies on an odd cycle, so SCF2 eliminates unexplained undecidedness outside paradox zones.","Among the seven naive-based variants, SCF2 is the only one satisfying Directionality, INRA, and SCOOC simultaneously.","The six-cycle case in which CF2 accepts two opposite arguments is repaired in SCF2, since SCOOC forces one of them to be rejected."],"supporting_citations":[{"why":"It defines abstract argumentation frameworks and the admissibility-based semantics that are the baseline for comparison.","marker":"[14]"},{"why":"It introduces the principle-based approach and the Directionality principle that the paper adopts as one of the three required properties.","marker":"[3]"},{"why":"It introduces SCC-recursiveness and the CF2 semantics that SCF2 builds on by replacing naive semantics with SCOOC-naive inside each component.","marker":"[4]"},{"why":"It defines stage2 semantics and criticizes CF2's six-cycle behavior, which the paper uses to motivate SCOOC and to contrast stage2 with SCF2.","marker":"[15]"},{"why":"It is the first cognitive study showing that naive-based semantics such as CF2 predict human judgments better than admissibility-based semantics.","marker":"[12]"},{"why":"It is the second cognitive study whose results suggest SCF2 matches human judgments on frameworks that distinguish SCF2 from CF2 and stage2.","marker":"[13]"},{"why":"It defines persistence and monotony properties for argumentation semantics that are closely related to the new INRA principle.","marker":"[19]"}],"fun_headline_variants":["SCF2 alone satisfies three key argumentation principles","New semantics SCF2 matches human rational judgment","SCF2: only semantics passing both new principles","SCF2 deletes self-attacks, blocks odd cycles, matches humans","SCF2 proven to satisfy INRA and directionality unlike CF2"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof that SCF2 satisfies INRA depends on an induction step that is not fully written out in Lemma 3(b): after deleting an argument attacked by every extension, the graph's strongly connected parts may split, and the proof says 'we continue as in the proof of part (a)' without supplying that induction; if it fails for some graph, SCF2 would not actually satisfy INRA.","fun_headline_variants_meta":{"raw":{"variants":["SCF2 alone satisfies three key argumentation principles","New semantics SCF2 matches human rational judgment","SCF2: only semantics passing both new principles","SCF2 deletes self-attacks, blocks odd cycles, matches humans","SCF2 proven to satisfy INRA and directionality unlike CF2"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001231,"raw_usage":{"total_tokens":5048,"prompt_tokens":929,"completion_tokens":4119,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":4035}},"tokens_in":545,"tokens_out":4119,"duration_ms":29272,"temperature":1.0,"reasoning_tokens":4035,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:40:58.793739+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a brute-force search over all finite argumentation frameworks with up to, say, five arguments, looking for a framework in which some argument is attacked by every SCF2 extension yet deleting that argument changes the set of extensions. A single such framework refutes Theorem 8; if none exists, the open gap in Lemma 3(b) becomes a clear target for a full rewrite of the proof.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"It defines abstract argumentation frameworks and the admissibility-based semantics that are the baseline for comparison."},{"cited_title":"Baroni and M","cited_arxiv_id":null,"evidence_quote":"It introduces the principle-based approach and the Directionality principle that the paper adopts as one of the three required properties."},{"cited_title":"Baroni, M","cited_arxiv_id":null,"evidence_quote":"It introduces SCC-recursiveness and the CF2 semantics that SCF2 builds on by replacing naive semantics with SCOOC-naive inside each component."},{"cited_title":"Dvoˇ r´ ak and S","cited_arxiv_id":null,"evidence_quote":"It defines stage2 semantics and criticizes CF2's six-cycle behavior, which the paper uses to motivate SCOOC and to contrast stage2 with SCF2."},{"cited_title":"Cramer and M","cited_arxiv_id":null,"evidence_quote":"It is the first cognitive study showing that naive-based semantics such as CF2 predict human judgments better than admissibility-based semantics."},{"cited_title":"Cramer and M","cited_arxiv_id":null,"evidence_quote":"It is the second cognitive study whose results suggest SCF2 matches human judgments on frameworks that distinguish SCF2 from CF2 and stage2."},{"cited_title":"Rienstra, C","cited_arxiv_id":null,"evidence_quote":"It defines persistence and monotony properties for argumentation semantics that are closely related to the new INRA principle."}],"review_version":1}