Pith. sign in

REVIEW 3 major objections 5 minor 26 references

Choices and their Provenance: Explaining Stable Solutions of Abstract Argumentation Frameworks

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Stable solutions of abstract argumentation frameworks can be explained by a minimal set of critical attacks: delete those attacks and the chosen stable extension becomes the grounded extension of the repaired graph.

desk verdict A sensible incremental idea for explaining stable AF semantics, but the paper's own Clingo program doesn't actually implement Definition 16 as stated, and the undec-only restriction is unproven. read the letter →

arxiv 2506.01087 v1 pith:LBYNOFWG submitted 2025-06-01 cs.AI cs.SC

Bertram Ludäscher , Yilin Xia , Shawn Bowers This is my paper
classification cs.AIcs.SC
keywords abstractargumentationstablesemanticsgroundedwell-foundedprovenancecriticalattacksetsanswersetprogrammingexplanation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

An abstract argumentation framework is a directed graph whose nodes are arguments and whose edges are attacks; the skeptical grounded semantics assigns each argument in, out, or undecided in a unique bottom-up way, whereas stable semantics chooses a 2-valued model and then tests it. This paper claims that every stable solution can be explained as the grounded solution of a minimally repaired graph: a minimal set of critical attack edges exists whose deletion makes the chosen stable extension exactly the grounded extension of the modified framework. The critical edges mark the choices and assumptions a stable model makes among arguments left undecided by the skeptical semantics, turning an ambiguous region into a definite verdict. If the claim is right, stable reasoning becomes diagnosable in the same concrete way grounded reasoning already is—one can point to the exact attacks to suspend, attach lengths and attack types to every argument in the explanation, and offer users minimal repairs that resolve ambiguity. The paper establishes the construction for stable semantics and implements the search for minimal critical attack sets in Answer Set Programming (Clingo).

What carries the argument

The central object is the critical attack set (Definition 16): a minimal set $\Delta_{i,j}$ of attack edges such that a chosen stable extension $S_i$ is the grounded extension of the framework with those edges removed, $G'=(V,E\setminus \Delta_{i,j})$. This definition lets the authors transfer the well-founded game provenance of grounded semantics—optimal argument lengths, and edge types such as primary and secondary successful attacks, failed attacks, undecided attacks, and blunders—to stable solutions via a provenance overlay, in which undecided arguments are recolored by their stable verdict and lengths are taken from the repaired grounded labeling. Computationally, the search is a generate-and-test loop written in Clingo: the choice rule at line 11 guesses a subset of undec-to-undec attacks, lines 17–19 recompute the grounded labeling of the reduced graph, and the constraint at line 22 plus the minimize directive at line 26 select cardinality-minimal critical attack sets.

What would settle it

Construct a finite AF and a stable extension $S$ such that every set of attacks whose deletion makes $S$ the grounded extension contains at least one attack with an endpoint labeled in or out in the original grounded labeling; running the Clingo program of Fig. 6 would then return no critical attack set, contradicting the promise that such a minimal set exists for the stable solution.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the non-deterministic part of stable semantics can be localized and made well-founded. Concretely, for a framework $G=(V,E)$ with grounded extension $S_0$ and a stable extension $S_i$, there exists a minimal set of attacks $\Delta_{i,j} \subseteq E$ such that $S_i$ is the grounded extension of the repaired framework $G'=(V, E \setminus \Delta_{i,j})$. The paper calls these sets critical attack sets; they identify exactly which attacks a stable solution discounts in order to turn its chosen 2-valued model into the unique skeptical solution. Because the repaired graph has a well-founded grounded labeling, the full game-based provenance machinery—argument lengths, primary and secondary successful attacks, failed attacks, and blunders—can be reused to annotate the stable solution as a provenance overlay. In the examples, different stable solutions of the same framework correspond to different minimal critical attack sets, such as suspending $C \to D$ versus $D \to C$ for the two stable models of a two-cycle, so the choice between alternative rational positions is made explicit.

Load-bearing premise

The load-bearing premise is that suspending only attacks whose two endpoints are both undecided in the original grounded labeling is always enough to turn a stable extension into the grounded extension; no lemma proves that a required repair could never involve an attack touching an already decided argument.

Editorial extensions

If this is right

  • A stable solution together with a critical attack set can be re-expressed as the grounded well-founded solution of the graph with those attacks suspended, so the game-based provenance of grounded semantics—lengths, primary and secondary attacks, failed attacks—applies to stable semantics.
  • The minimal critical attack sets expose the choices a stable model makes: alternative stable solutions of the same framework correspond to different minimal edge deletions, so the source of ambiguity is localized to specific edges.
  • Provenance overlays give each argument in a stable solution a length and attack-type provenance drawn from the repaired grounded labeling, enabling the layered visualization and regular-path-query style inspection previously available only for grounded solutions.
  • The Clingo implementation enumerates all cardinality-minimal critical attack sets for a finite AF and a given stable extension, and a variant that first finds all candidate sets and then post-filters yields subset-minimal sets, making diagnosis and minimal repair computable.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves open whether the undec-attacks-only restriction is complete; a natural benchmark is to search random AFs for a stable extension whose only repairs touch decided endpoints.
  • The repair view yields a counterfactual reading—if the critical attacks were suspended, the stable model would be the unique grounded solution—which suggests connecting critical attack sets to counterfactual explanation frameworks for argumentation.
  • Overlay lengths offer a secondary criterion for choosing among multiple minimal critical attack sets of the same stable solution, for example preferring the repair that minimizes the maximal derivation length.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes to explain stable extensions of abstract argumentation frameworks by identifying minimal sets of critical attacks whose removal makes the chosen stable extension equal to the grounded extension of the repaired framework. It defines provenance overlays that annotate stable solutions with grounded derivation lengths and critical edges, illustrates the idea on several examples (Figs. 1, 2, 4, 5), and gives a naive algorithm plus an ASP (Clingo) implementation (Fig. 6). The central claim is that every stable solution can be diagnosed as a well-founded solution of a minimally repaired graph, combining grounded derivation steps with explicit choice steps.

Significance. The notion of critical attack sets is appealing and goes beyond the authors' prior grounded-provenance work by making the nondeterministic choice component of stable semantics explicit. The paper ships concrete examples and a concrete ASP program, and the central definition does not rely on fitted parameters or self-derived benchmarks. If the correctness and completeness of the computation are established, the approach would provide a useful diagnostic tool for explaining ambiguity in AFs, with natural applications in legal reasoning and visualization. However, the current manuscript's computational core has a correctness gap and an unproven search-space restriction, so the central claim is not yet fully supported.

major comments (3)
  1. [Fig. 6, lines 17–22] The Clingo program does not enforce that the grounded labeling of the modified AF equals the target stable extension. Lines 17–19 compute the grounded labeling of G'=(V,E\critical), and line 22 only rejects candidates with any undec argument; there is no constraint relating the computed in/out labeling to the desired stable solution S_i. As a result, any deletion set that makes the grounded extension total is accepted, including sets that correspond to other stable extensions. For the AF of Fig. 1, both {C→D} and {D→C} make the grounded extension total, so both are returned as cardinality-minimal, although they correspond to the two different stable solutions S1 and S2. If the intended target constraints are assumed to be supplied externally, the listing is incomplete and the caption's claim that the program computes critical attack sets for a specified stable solution is misleading.
  2. [Section 4, step (1) and Fig. 6, line 11] The restriction of candidate critical attacks to edges whose endpoints are both undecided in the original grounded labeling is unproven. No lemma shows that every minimal critical attack set of a stable extension can be chosen from the undec-induced subgraph, nor that no repair outside this set is ever required. Since the algorithm enumerates only undec-undec edges, completeness of the computation depends on this missing result. The paper should either prove the restriction or remove it and adjust the search space accordingly.
  3. [Definition 16 and Fig. 6, line 26] The notion of minimality is ambiguous. Definition 16 says “minimal” without qualification, while the algorithm minimizes cardinality and the text mentions subset-minimal sets as an alternative. If the intended notion is subset-minimality, the #minimize directive in line 26 is insufficient because it selects only cardinality-minimal sets and can discard larger subset-minimal sets. The definition and the implementation should be aligned, and the paper should state explicitly which notion of minimality is used.
minor comments (5)
  1. [Section 4, Definition 16] The existence of a critical attack set for an arbitrary stable extension is asserted implicitly but never proved; a short argument (e.g., deleting all attacks from arguments outside S_i to arguments inside S_i) would make the definition self-contained.
  2. [Section 4, paragraph after Definition 16] The definition of a provenance overlay is informal; the notation for dashed labels, primed lengths, and the provenance-relevant edge types could be formalized so that the examples can be reproduced by readers.
  3. [Figure 3] The labels b-1, b-2, b-3 and n/a in the type graph are not explained in the text; a short explanation in the caption or surrounding prose would improve readability.
  4. [Section 5, related work] The comparison with Baumann and Ulbricht's explanation approach is high-level; a more precise account of the differences and overlaps would clarify the novelty of critical attack sets.
  5. [Fig. 6, comments] The source code comments contain spacing artifacts (e.g., “E li mi na te c an di da te s ol ut io ns” and “c r i t i c a l _ c n t”) that should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the critical-attack construction is defined and computed directly from the AF, with only background self-citation.

full rationale

The paper's central notion, Definition 16, defines a critical attack set as a minimal set of attacks whose deletion makes a specified stable extension the grounded extension of the repaired graph. The computation in Section 4 and Fig. 6 directly searches for such sets and evaluates grounded semantics on the repaired AF; it does not fit a parameter from the target and then rename that parameter as the output. The provenance/overlay machinery borrows language and background from the authors' prior game-provenance work, but the core claim about critical attacks is stated and computed from the AF itself rather than imported from those papers. The substantive caveat is a correctness/completeness issue, not a circularity one: Fig. 6 lines 17-22 only reject candidates leaving undecided arguments, and the printed listing does not show the constraints relating the repaired grounded extension to the selected stable solution. That is a gap in the algorithm as presented, but it is not a case of an output being definitionally equal to an input. There are no fitted inputs called predictions, no uniqueness theorems imported from the authors' previous work, and no reduction of the central derivation to a self-citation chain.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim rests on standard Dung semantics plus two unproved structural assumptions: that stable extensions can be made grounded by deleting only undec attacks, and that the ASP program's 2-valuedness check correctly targets the specified stable extension. No numerical parameters are fit. No physical or empirical entities are introduced; critical attack sets are formal constructs defined from the AF input.

assumptions (4)
  • domain assumption The grounded extension is the well-founded model of P_AF1 and stable extensions are the stable models of P_AF1, following Dung (1995).
    Invoked in Section 2.1 as the logical foundation for translating AF semantics into the fixpoint and game provenance framework used throughout the paper.
  • ad hoc to paper For any stable extension S_i there exists a minimal set of attacks Delta whose deletion makes S_i the grounded extension of the modified AF.
    This existence and minimality fact is the content of Definition 16 but is asserted without proof in Section 4.
  • ad hoc to paper Only attacks between undecided arguments in the original grounded labeling need to be considered as candidate critical attacks.
    Restricts the search space in the computation description and in Fig. 6 line 11; no lemma establishes the sufficiency of this restriction.
  • domain assumption The provenances for well-founded games transfer to grounded AF labelings as established in the authors' prior work [9,10].
    Used to justify the game-based provenance labels for overlays; this is prior work rather than a result derived in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Choices and their Provenance: Explaining Stable Solutions of Abstract Argumentation Frameworks." pith.science (2026). https://pith.science/paper/LBYNOFWG

@misc{pith2026250601087,
  author       = {Pith},
  title        = {Pith review of: Choices and their Provenance: Explaining Stable Solutions of Abstract Argumentation Frameworks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LBYNOFWG}},
  note         = {Machine review of arXiv:2506.01087}
}
abstract

The rule $\mathrm{Defeated}(x) \leftarrow \mathrm{Attacks}(y,x),\, \neg \, \mathrm{Defeated}(y)$, evaluated under the well-founded semantics (WFS), yields a unique 3-valued (skeptical) solution of an abstract argumentation framework (AF). An argument $x$ is defeated ($\mathrm{OUT}$) if there exists an undefeated argument $y$ that attacks it. For 2-valued (stable) solutions, this is the case iff $y$ is accepted ($\mathrm{IN}$), i.e., if all of $y$'s attackers are defeated. Under WFS, arguments that are neither accepted nor defeated are undecided ($\mathrm{UNDEC}$). As shown in prior work, well-founded solutions (a.k.a. grounded labelings) "explain themselves": The provenance of arguments is given by subgraphs (definable via regular path queries) rooted at the node of interest. This provenance is closely related to winning strategies of a two-player argumentation game. We present a novel approach for extending this provenance to stable AF solutions. Unlike grounded solutions, which can be constructed via a bottom-up alternating fixpoint procedure, stable models often involve non-deterministic choice as part of the search for models. Thus, the provenance of stable solutions is of a different nature, and reflects a more expressive generate & test paradigm. Our approach identifies minimal sets of critical attacks, pinpointing choices and assumptions made by a stable model. These critical attack edges provide additional insights into the provenance of an argument's status, combining well-founded derivation steps with choice steps. Our approach can be understood as a form of diagnosis that finds minimal "repairs" to an AF graph such that the well-founded solution of the repaired graph coincides with the desired stable model of the original AF graph.

Figures

Figures reproduced from arXiv: 2506.01087 by the authors.

Figure 1
Figure 1. AF solutions: (a) Input AF, (b) 3-valued grounded solution 𝑆0; (c), (d): stable solutions 𝑆1 and 𝑆2 (as provenance overlays). The grounded solution is shown in Fig. 1b: Since A is a source node, it is not attacked by any other argument and thus can be accepted immediately and labeled in (blue color). Because A is in and attacks B, the latter is defeated and labeled out (orange). We also label decided nodes with a na… view at source ↗
Figure 2
Figure 2. AF solutions for Wild Animals cases [5]: (a) The ranked layout of 𝑆0 uses the length of arguments: e.g., F.4 requires at most four rounds of argumentation to prove that F is accepted (in). Yellow nodes (length = ∞) are undecided (undec). Distinct edge types are used to account for their semantic roles [9]. The overlays in (b) and (c) represent alternative resolutions 𝑆 ′ 1,1 and 𝑆 ′ 2,1 : The undec nodes E, J, M, N,… view at source ↗
Figure 3
Figure 3. Classification of attack types: Argument 𝑥 is out if there exists (∃) a primary or secondary attack from an argument 𝑦 that is in (𝑦’s attack is successful). Argument 𝑥 is in if all (∀) of its attackers are out (all attacks on 𝑥 fail). The length of 𝑥 indicates the number of attacks in an optimal dialogue (game): A defeat can be forced from an out argument in at most 2𝑛+1 attacks; acceptance can be delayed for at le… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: b, argument D is displayed as D.1’, denoting that its length value in the overlay is 1. Arguments with the same length values in the original grounded solution are denoted as such, e.g., H.1 is the same in both the original and the modified grounded labeling. Critical …
Figure 5
Figure 5. Figure 5: Layered visualizations of the 𝑆1–𝑆4 overlays from [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Example Clingo (ASP) program for computing the [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

26 extracted references · 25 canonical work pages

  1. [3]

    Baumann, S

    R. Baumann, S. Doutre, J.-G. Mailly, and J.P. Wallner. 2021. Enforcement in Formal Argumentation. InHandbook of Formal Argumentation. Vol. 2. 445–510

    work page 2021

  2. [4]

    Baumann and M

    R. Baumann and M. Ulbricht. 2021. Choices and their Consequences - Explaining Acceptable Sets in Abstract Argumentation Frameworks. InKR. 110–119

    work page 2021

  3. [1]

    Alfano, S

    G. Alfano, S. Greco, F. Parisi, and I. Trubitsyna. 2024. Counterfactual and semi- factual explanations in abstract argumentation: formal foundations, complexity and computation. InKR

    work page 2024

  4. [2]

    Baroni, D

    P. Baroni, D. Gabbay, M. Giacomin, and L. van der Torre. 2018.Handbook of Formal Argumentation. London: College Publications

    work page 2018

  5. [5]

    Bench-Capon

    T. Bench-Capon. 2002. Representation of Case Law as an Argumentation Frame- work. InJURIX. 103–112

    work page 2002

  6. [6]

    Bench-Capon

    T. Bench-Capon. 2020. Before and after Dung: Argumentation in AI and Law. Argument & Computation.11, 1-2 (2020), 221–238

    work page 2020

  7. [7]

    Bench-Capon and S

    T. Bench-Capon and S. Modgil. 2009. Case law in extended argumentation frameworks. InICAIL. 118–127

    work page 2009

  8. [8]

    Borg and F

    A. Borg and F. Bex. 2024. Minimality, necessity and sufficiency for argumentation and explanation.Int. J. Approx. Reasoning168, C (May 2024)

    work page 2024

Show all 26 references
  1. [9]

    Bowers, Y

    S. Bowers, Y. Xia, and B. Ludäscher. 2024. The Skeptic’s Argumentation Game or: Well-Founded Explanations for Mere Mortals. InWorkshop on Systems and Algorithms for Formal Argumentation (SAFA) (CEUR, Vol. 3757). 104–118

    2024

  2. [10]

    Bowers, Y

    S. Bowers, Y. Xia, and B. Ludäscher. 2024. On the Structure of Game Provenance and its Applications. InTaPP. 602–609

    2024

  3. [11]

    Cabrio and S

    E. Cabrio and S. Villata. 2012. Combining Textual Entailment and Argumentation Theory for Supporting Online Debates Interactions. InAnnual Meeting of the Association for Computational Linguistics. 208–212

    2012

  4. [12]

    Caminada

    M. Caminada. 2006. On the Issue of Reinstatement in Argumentation. InLogics in Artificial Intelligence. 111–123

    2006

  5. [13]

    Caminada

    M. Caminada. 2018. Argumentation Semantics as Formal Discussion. See [ 2], Chapter 10, 487–518

    2018

  6. [14]

    Caminada and P

    M. Caminada and P. Dunne. 2020. Strong Admissibility Revisited: Theory and Applications.Argument & Computation10, 3 (2020), 277–300

    2020

  7. [15]

    Caminada, S

    M. Caminada, S. Sá, J. Alcântara, and W. Dvořák. 2015. On the Equivalence between Logic Programming Semantics and Argumentation Semantics.Approx. Reasoning58 (2015), 87–111

    2015

  8. [16]

    P.M. Dung. 1995. On the Acceptability of Arguments and Its Fundamental Role in Nonmonotonic Reasoning, Logic Programming and n-Person Games.AI77, 2 (1995), 321–357

    1995

  9. [17]

    Gelfond and V

    M. Gelfond and V. Lifschitz. 1988. The Stable Model Semantics for Logic Pro- gramming. InILPS. 1070–1080

    1988

  10. [18]

    Köhler, B

    S. Köhler, B. Ludäscher, and D. Zinn. 2013. First-Order Provenance Games. InIn Search of Elegance in the Theory and Practice of Computation. 382–399

    2013

  11. [19]

    Longo and L

    L. Longo and L. Hederman. 2013. Argumentation Theory for Decision Support in Health-Care: A Comparison with Machine Learning. InBrain and Health Informatics. 168–180

    2013

  12. [20]

    Ludäscher, S

    B. Ludäscher, S. Bowers, and Y. Xia. 2023. Games, Queries, and Argumentation Frameworks: Towards a Family Reunion. InAdvances in Argumentation in AI, Vol. 3546. CEUR

    2023

  13. [21]

    Odekerken, A.M

    D. Odekerken, A.M. Borg, and M. Berthold. 2023. Demonstrating PyArg 2.0. In Advances in Argumentation in AI (AI 3), Vol. 3546. CEUR

    2023

  14. [22]

    Rahwan and K

    I. Rahwan and K. Larson. 2009.Argumentation and Game Theory. 321–339

    2009

  15. [23]

    Van Gelder

    A. Van Gelder. 1993. The Alternating Fixpoint of Logic Programs with Negation. J. Comput. System Sci.47, 1 (1993), 185–221

    1993

  16. [24]

    Van Gelder, K.A

    A. Van Gelder, K.A. Ross, and J.S. Schlipf. 1991. The Well-founded Semantics for General Logic Programs.J. ACM38, 3 (1991), 619–649

    1991

  17. [25]

    Y. Xia, D. Odekerken, S. Bowers, and B. Ludäscher. 2024. Layered Visualization of Argumentation Frameworks. InCOMMA, Vol. 388. 373–374

    2024

  18. [26]

    Y. Xia, H. Zheng, S. Bowers, and B. Ludäscher. 2025. AF-Xray: Visual Explanation and Resolution of Ambiguity in Legal Argumentation Frameworks. In20th Intl. Conf. on Artificial Intelligence and Law (ICAIL). https://github.com/idaks/xray

    2025

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.