{"id":"a49bc480-8a61-49ab-abe5-f3d16235019a","arxiv_id":"2604.15898","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":3.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Non-symbolic XAI methods such as SHAP lack rigor and can mislead; symbolic methods offer a rigorous alternative for relative feature importance.","lead":"This paper surveys rigorous symbolic methods for assigning feature importance in machine learning explanations, as alternatives to popular non-symbolic tools like SHAP. It matters because non-rigorous explanations can mislead decisions in high-stakes settings.","discovery_kind":"review","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the abstract-only limitation already reflected in the UNVERDICTED verdict.","rationale":"The Reader already set UNVERDICTED / LOW confidence precisely because only the abstract is available and therefore neither proofs, code, nor experiments can be examined. The strongest claim is a high-level contrast between non-rigorous non-symbolic methods and symbolic alternatives; the weakest assumption is that those alternatives actually deliver at useful scale. That is the correct load-bearing concern for an abstract-only survey, and no additional internal inconsistency or technical over-claim can be isolated without the body of the paper. Consequently the stress-test finds no reason to move the verdict, novelty score, or correctness-risk assessment. The recommended concrete check is simply to acquire the full text and verify whether any surveyed method is shown to scale while avoiding known SHAP pathologies; that single step settles whether the framing is supported or remains programmatic.","tokens_in":1865,"tokens_out":464,"duration_ms":5050,"concrete_test":"Obtain the full PDF and check whether any section supplies a concrete symbolic procedure (with complexity bounds or empirical scale) that demonstrably avoids a documented SHAP failure mode on a non-trivial model; if none does, the ‘viable alternative’ framing remains aspirational and the UNVERDICTED status is confirmed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper is a survey/position piece whose central claim (non-symbolic XAI, especially SHAP-style Shapley methods, lacks rigor and can mislead; rigorous symbolic methods are a viable alternative for relative feature importance) is stated at the level of framing and overview. With only the abstract available there is no internal derivation, theorem, experiment, or coverage claim that can be stress-tested for inconsistency, hidden assumptions, or over-reach. The reader’s weakest_assumption correctly flags the practical-delivery gap of the surveyed symbolic efforts, but that gap is definitional for an abstract-only review rather than a load-bearing flaw inside an argument we can inspect. No further concrete technical soft spot (equation, lemma, scaling claim, or evaluation protocol) is present to attack.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript is a survey/position paper arguing that non-symbolic XAI methods, particularly Shapley-value approaches such as SHAP, lack formal rigor and can mislead decision-makers, especially in high-stakes settings. It frames ongoing work on rigorous symbolic methods of XAI as a viable alternative for assigning relative feature importance. Only the abstract is available for review; no theorems, proofs, comparative evaluations, or detailed coverage of the surveyed methods appear in the provided text.","tokens_in":2013,"tokens_out":527,"duration_ms":5368,"significance":"If the full paper delivers a precise technical survey that maps concrete failure modes of SHAP-style methods to specific symbolic alternatives with formal guarantees, it would be a useful orientation piece for the XAI community. The abstract correctly identifies a known line of critique (lack of rigor in popular attribution methods) and points toward symbolic approaches as a remedy. However, significance cannot be fully assessed from the abstract alone, because the practical scope, formal guarantees, and comparative evidence of the surveyed symbolic methods are not shown.","major_comments":[{"comment":"Only the abstract is available. The central claim that non-symbolic methods (exemplified by SHAP) lack rigor and that symbolic methods provide a rigorous alternative for relative feature importance is stated as framing, but no proofs, counterexamples, formal comparisons, or coverage of the surveyed efforts appear. Without the full text it is impossible to verify whether the claim is supported or whether the practical-delivery gap of symbolic methods is addressed. A complete manuscript is required before a substantive technical assessment can be made.","section":null},{"comment":"The abstract asserts a 'provable lack of rigor' for Shapley-value XAI. In a full submission this claim must be backed by at least one concrete, citable formal result or counterexample (e.g., a named theorem or published impossibility result) rather than remaining at the level of overview. Absent that grounding, the load-bearing contrast with symbolic methods cannot be evaluated.","section":null}],"minor_comments":[],"recommendation":"uncertain","confidential_remarks":"This is an abstract-only review of arXiv:2604.15898. The manuscript cannot be properly refereed until the full text is supplied. The abstract is coherent as a position/survey framing but contains no inspectable technical content (equations, theorems, tables, or evaluation protocols). I recommend requesting the full paper and re-assigning for a normal review cycle; the present report should not be treated as a final accept/reject decision."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"Punchline: this is an overview, not a new theorem or algorithm. From the abstract alone it frames a decade of non-symbolic XAI (SHAP/Shapley as the prime example) as lacking rigor and potentially misleading in high-stakes settings, then points to ongoing symbolic methods as the route to rigorous relative feature importance.\n\nWhat is actually new is modest by design: an organizing survey/position of work already underway in formal/symbolic XAI. Credit where it is earned—the problem statement is clean and useful. The claim that Shapley-value tools can mislead is a real, documented line of critique in this literature, not invented here. Stating that relative feature importance should be assigned with rigor is a fair organizing claim for people who still treat SHAP as the default.\n\nSoft spots are proportional to what we have. We only have the abstract, so coverage, scaling, practical delivery, and independent benchmarks cannot be checked. The reader’s weakest assumption is right: that the surveyed symbolic efforts actually deliver usable rigorous alternatives at useful scale. That is definitional for an abstract-only overview, not a load-bearing flaw inside an argument we can inspect. Novelty is survey-level; circularity risk looks low if the body cites formal results properly rather than only the authors’ prior papers. The stress-test note is correct—there is no equation or protocol here to attack.\n\nWho this is for: researchers already in formal/symbolic XAI, or practitioners who need a map of alternatives to SHAP. A serious editor should send the full paper to referees rather than desk-reject a clear position survey on this topic. I would not bring the abstract alone to reading group, and I would not cite until I see the body. If the full text is a careful map with honest limits, it is useful; if it is mostly self-citation without external comparison, less so. Engage the full paper when it is available.","headline":"Abstract-only survey/position piece: SHAP-style attribution lacks rigor and symbolic methods are the alternative—known critique, value turns on how well the full text maps the alternatives.","tokens_in":2635,"tokens_out":494,"would_cite":false,"duration_ms":14278,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"Non-symbolic XAI such as SHAP lacks rigor and can mislead; rigorous symbolic methods offer a sound alternative for relative feature importance.","keywords":["explainable AI","XAI","Shapley values","SHAP","feature attribution","symbolic methods","rigorous explainability","machine learning"],"falsifier":"A high-stakes ML model for which a symbolic XAI method produces a relative feature-importance ranking that is later shown, by formal verification or controlled counter-example, either to be incomplete or to mis-rank a feature that a decision-maker relies on.","tokens_in":2731,"feed_emoji":"🔍","tokens_out":485,"duration_ms":4628,"temperature":0.7,"pith_summary":"This paper argues that for roughly a decade the dominant approach to explaining complex machine learning models has been non-symbolic methods that lack formal rigor and can therefore mislead human decision-makers, especially in high-stakes settings. The prime concrete example is the use of Shapley values in explainable AI, most visibly through the SHAP tool. The authors survey and frame ongoing work on rigorous symbolic methods of XAI as a viable alternative, specifically for the task of assigning relative importance to input features. A sympathetic reader cares because explanations that are not guaranteed to be correct can produce false confidence precisely where reliability is most needed; the paper positions symbolic techniques as the route to explanations that can be trusted.","feed_headline":"SHAP-style XAI lacks rigor; symbolic methods can fix feature importance","feed_subtitle":"A survey argues non-symbolic explanations can mislead and positions formal alternatives for high-stakes use","key_machinery":"Rigorous symbolic methods of XAI: formal, logic- or constraint-based techniques that assign relative feature importance with guarantees of correctness, offered as the alternative to non-rigorous Shapley-value approaches.","core_discovery":"Non-symbolic XAI methods, exemplified by Shapley-value tools such as SHAP, are not rigorous and can mislead decision-makers; rigorous symbolic methods of XAI constitute a sound alternative for assigning relative feature importance.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["SHAP can mislead on feature importance; symbolic XAI offers rigor","Non-symbolic XAI like SHAP lacks rigor for feature attributions","Symbolic methods provide rigorous feature importance unlike SHAP","Switch from SHAP to symbolic XAI for sound relative feature ranks","Rigorous symbolic XAI fixes misleading feature attributions of SHAP"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"That the ongoing symbolic XAI efforts the paper surveys actually deliver practical, rigorous relative feature-importance assignments that avoid the failure modes of SHAP-style methods at useful scale.","fun_headline_variants_meta":{"raw":{"variants":["SHAP can mislead on feature importance; symbolic XAI offers rigor","Non-symbolic XAI like SHAP lacks rigor for feature attributions","Symbolic methods provide rigorous feature importance unlike SHAP","Switch from SHAP to symbolic XAI for sound relative feature ranks","Rigorous symbolic XAI fixes misleading feature attributions of SHAP"]},"model":"grok-4.5","effort":"low","cost_usd":0.008368,"raw_usage":{"total_tokens":1872,"prompt_tokens":620,"num_sources_used":0,"completion_tokens":70,"cost_in_usd_ticks":83680000,"prompt_tokens_details":{"text_tokens":620,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1182,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":620,"tokens_out":70,"duration_ms":8387,"temperature":1.0,"reasoning_tokens":1182,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-12T19:31:35.110346+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"A high-stakes ML model for which a symbolic XAI method produces a relative feature-importance ranking that is later shown, by formal verification or controlled counter-example, either to be incomplete or to mis-rank a feature that a decision-maker relies on.","supporting_citations":[],"review_version":3}