{"id":"b0bd2308-47dd-48b7-8c68-a0fa6158a32b","arxiv_id":"2501.06201","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A Deng-entropy-weighted pignistic average with Yager-style conflict redistribution is presented as a new evidence fusion rule, with several examples claiming sharper verdicts.","lead":"This paper proposes a new way to combine conflicting evidence by mixing pignistic probabilities, Z-number-style reliability weights, and an OWA-style aggregation inside Dempster-Shafer theory. The reported examples and applications show sharper decisions than classic rules, but the equations do not reproduce the numbers in the tables.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The method's reported superiority rests on an unjustified self-combination of the synthesized BPA; Section 3.7's n-1 Dempster self-fusion has no independent-source justification, and hand-recomputing from Eqs. 18-25 reproduces Table 2 only, not Tables 4, 6, or 8.","rationale":"The reader's REJECT verdict is well-founded, and my independent reading identifies the same load-bearing spot: the final combination stage. The paper's own contribution list claims a 'concise aggregation' and 'much better performance', but the only mechanism that produces the extreme sharpening in the tables is the self-combination in Section 3.7. Because Dempster's rule is only a valid pooling operator for distinct sources, combining a BPA with itself n-1 times manufactures confidence. This is not a minor technicality: all the favorable examples (Tables 2, 4, 6, 8) show the proposed method assigning higher mass to the top proposition than competitors. The magnitude of that advantage is controlled by the arbitrary number of self-combinations. The paper does not justify why n-1 is the right number, nor why self-combination should preserve the 'rational and correct' probabilities it claims.\n\nThe problem is compounded by the malformed OWA step in Eqs. 23-24: ωPi is introduced as a set of Δ values and multiplied into a sum of scalar probabilities, with no ordering or OWA weight vector defined. This means even the pre-combination is not fully specified. My hand recomputation from the natural reading of Eqs. 18-25 (reliability-weighted pignistic average plus conflict reallocation, then one self-combination) matches Example 1 exactly, but fails for Examples 2 and 3. The discrepancies are large (e.g., 0.1429 vs. 0.2574 for U in Example 2). This demonstrates the published tables are not determined by the stated method. In the absence of code or formal verification, the empirical support for the headline claim evaporates. No ad hominem is intended: the issue is that the paper's equations do not entail its reported numbers.\n\nThe concrete check I propose is a reproduction from the paper text only; if the check is run and the numbers match after some reasonable clarification of the OWA step, my concern would be weakened. But as written, the central claim is unsupported.","tokens_in":18778,"tokens_out":13783,"duration_ms":110150,"concrete_test":"Reimplement the method strictly from Sections 3.1-3.7: (i) compute ξ(Pi)^C and ξ(X) from Eqs. 18-19; (ii) compute Δ from Eq. 21; (iii) compute ε(Pi) using Eq. 24 as a reliability-weighted sum, interpreting ωPi as the Δ weights since no OWA sorting is specified; (iv) form the synthesized BPA from Eq. 25; (v) apply Dempster's rule to this BPA with itself n-1 times. Compare outputs to Tables 2, 4, 6, 8, and 10. If any entry deviates beyond rounding—Tables 4, 6, and 8 already do—then the published results are not consequences of the stated method, and the claim of superior accuracy is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the pipeline yields 'rational and correct judgments' with high accuracy requires Section 3.7's final step to be a legitimate evidence-combination operation. It is not. Eq. 25 produces a single synthesized BPA; Section 3.7 then says to combine the modified values using Dempster's rule n-1 times. Since there is only one BPA after synthesis, this is self-combination: the same evidence is fed into Dempster's rule as if it were n independent sources. Dempster's rule has no justification for identical inputs; self-combination monotonically inflates belief in dominant propositions without adding information. Example 1 makes the effect visible: without self-combination the synthesized BPA is (0.4950, 0.0101, 0.4950); one self-combination yields (0.4999, 0.0002, 0.4999), the published Table 2. The paper offers no argument that this sharpening preserves true probabilities. Moreover, the step is under-specified: no equation is given for the self-combination, and whether n is the number of original sources or the number of combination passes is ambiguous. A strict recomputation from Eqs. 18-25 plus one self-combination reproduces Example 1 but not Examples 2 or 3: for Example 2 the computed output is (0.5438, 0.1978, 0.2574, 0.0011) versus the published (0.6229, 0.2340, 0.1429, 0.00); for Example 3 it is (0.8896, 0.0685, 0.0418) versus (0.9057, 0.0659, 0.0283). Repeating self-combination overshoots. Since the reported outputs are not consequences of the stated equations, the claimed superiority over Dempster, Yager, Murphy, and others is unsupported. This is an internal under-specification, not a disagreement with consensus.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Dempster-Shafer evidence-theoretic fusion method that combines pignistic transformation, Deng entropy, Z-number-like reliability pairs, OWA aggregation, and an unnormalized conflict mass. The pipeline is: (i) obtain a conflict mass via Yager-style unnormalized combination (Eqs. 18-19); (ii) compute Deng entropy and softmax-based reliabilities (Eqs. 20-21); (iii) form Z-number pairs per proposition (Eq. 22); (iv) aggregate via OWA (Eqs. 23-24); (v) form a modified BPA (Eq. 25); (vi) combine the modified BPA by Dempster's rule n-1 times (Section 3.7). The paper claims high accuracy and superiority over Dempster, Yager, Murphy, He, and other methods, supported by four numerical examples and two applications.","tokens_in":19186,"tokens_out":8720,"duration_ms":87987,"significance":"If the method were correctly specified and the reported improvements were reproducible, it would be a useful contribution to conflict management in evidence fusion: it is parameter-free, deterministic, and combines several existing ideas in a motivated way. The paper's strength is its clear attempt to address Zadeh's classic counter-intuitive conflict example without discarding the conflicting mass. However, the central aggregation equation is not well-defined, the n-1 self-combination step has no evidential justification, and the numerical tables do not follow from the stated formulas. The claimed significance therefore cannot currently be assessed, and the validation does not provide an external accuracy criterion.","major_comments":[{"comment":"The quantity omega_Pi is defined in Eq. (23) as the set {Delta_Evi1, ..., Delta_Evin}, so the product eta(Pi)^Evik times omega_Pi in Eq. (24) is not a defined mathematical expression. Even if omega_Pi is read as the vector of all Delta values, Eq. (24) never orders the inputs and never specifies the OWA weight vector required by Eqs. (10)-(11). The core aggregation step is therefore not a well-specified OWA operation, and the subsequent numerical results cannot be derived from it.","section":"3.5, Eq. (24)"},{"comment":"After Eq. (25) there is only one synthesized BPA. Instructing the reader to combine the modified values by Dempster's rule n-1 times means feeding that single BPA into Dempster's rule as if it were n independent sources. Dempster's rule is justified for independent bodies of evidence; self-combining identical evidence has no such justification and merely sharpens dominant propositions without adding information. The paper offers no argument that iterated self-combination preserves the true probabilities, and it is ambiguous whether n denotes the original number of sources or the number of combination passes. This step is load-bearing because it creates the decisive margins reported in Tables 2, 4, 6, and 8.","section":"3.7"},{"comment":"The numerical results are not consequences of the stated algorithm. For Example 1, direct computation from Eqs. (18)-(25) gives approximately (0.4950, 0.0101, 0.4950), whereas Table 2 reports (0.4999, 0.0002, 0.4999); reproducing the table requires an undocumented self-combination step. For Examples 2 and 3, Eqs. (18)-(25) leave positive mass on the compound focal element {T,Y}, but Tables 4, 6, and 8 report zero mass for {T,Y} with no stated reallocation. The claimed 'high accuracy' is therefore not reproducible from the published equations.","section":"Tables 2, 4, 6, 8 vs Eqs. (18)-(25)"},{"comment":"The evaluation does not provide a falsifiable test of the central claim. Section 4.1 asserts that propositions T and U 'are supposed to be allocated a mass close to 0.5', but that is an assumption built into the example rather than a measured outcome. The IRIS application in Section 5.1 reports no classification accuracy against known labels, and the procedure that converts the data into the BPAs of Table 9 is not described. Without an external accuracy criterion, the abstract's claim that the method 'keeps high accuracy in producing rational and correct judgments' is unsupported.","section":"Sections 4 and 5"}],"minor_comments":[{"comment":"The sentence after Eq. (9) says 'xi(empty) is not equal to 0', but Eq. (4) defines xi(empty)=0; the intended condition should be clarified.","section":"2.3, Eq. (9)"},{"comment":"The OWA operator is typeset incorrectly: the displayed vector in Eq. (10) appears as all theta_1 entries rather than a vector of dimension n, and Eq. (12) is an incomplete statement of the weight-sum property.","section":"2.4, Eqs. (10)-(12)"},{"comment":"The notation ZEvi1 = {ZA1 = (xi(A1), Delta_Evi1), ...} mixes ordered-pair tuples with set notation; using a consistent tuple or vector notation would avoid ambiguity.","section":"3.3"},{"comment":"Table 9 reports BPAs for the IRIS data, but the procedure that converts attribute measurements into these BPAs is not described, so the application cannot be independently reproduced.","section":"5.1"},{"comment":"The word 'carnality' appears where 'cardinality' is meant in Sections 2.2 and 2.3, and several figure captions are duplicated as 'The detailed process of proposed method'; these should be corrected.","section":"Throughout"}],"recommendation":"reject","confidential_remarks":"The mathematical problems are substantive rather than cosmetic: the central OWA equation is malformed, the self-combination step is unjustified, and the reported tables do not follow from the stated formulas. In my view this is not a case where a standard revision cycle can fix the paper within its current scope; the algorithm would need to be re-specified and all numerical claims re-derived and re-validated. I would not invite a resubmission unless the authors provide a complete, executable specification and reproducible code."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this paper is a marginal variant of the author's own He's method and Pan-Deng's OWA-based probability transform, rebranded with Z-number language. It is not a 'completely new method' as claimed, but it does put the pieces together in a way I haven't seen verbatim. The examples are the main selling point: on the classic Zadeh conflict example it produces (0.4999, 0.0002, 0.4999), which is more decisive than Dempster or Yager. That result is reproducible from the stated steps if you read Section 3.7 literally, i.e., synthesize one BPA and then combine it with itself n-1 times via Dempster's rule.\n\nThe problems are real but somewhat narrower than the stress-test suggests. Eq 24 is malformed: ωPi is defined as a set, so the product η×ωPi is not well-defined as written. In practice the paper seems to mean a simple weighted average using the reliabilities Δ, not the OWA operator it claims to use. The OWA framing is misapplied—there's no sorting or OWA weight vector; it's just softmax-weighted averaging. That is cosmetic but needs fixing.\n\nThe more substantive issue is Section 3.7: combining the synthesized BPA with itself n-1 times has no justification as a model of independent evidence. It inflates the dominant proposition's mass, exactly as Dempster self-combination always does. The paper does not address this. That said, this move is very common in the conflict-management literature (Murphy's average combination does the same), so it is not unique to this paper; the authors should simply cite a justification or acknowledge it as a heuristic.\n\nThe numerical tables are mostly consistent with a literal implementation of the text, contrary to the stress-test's claim, but the notation is so loose that a reader cannot be sure. My own recomputation of Examples 2 and 3 gave values within a few thousandths of the published tables, so I don't think the paper is hiding an extra step. The IRIS application is under-specified: no explanation of how BPAs are constructed from the raw features, and no ground truth or accuracy measure. That's a weakness.\n\nOverall: this is a plausible incremental heuristic in a crowded field, presented with too much hype ('completely new', 'high accuracy') and not enough mathematical care. A serious referee could sort out the notation and ask the authors to compare against their own prior work and justify the self-combination. I'd send it to review, but with the expectation of major revision.\n\nRecommendation: not desk-reject, but reject after review unless the authors tighten the math and validation.","headline":"A marginal variant of the author's own He's method and Pan-Deng's OWA transform, presented with sloppy notation and an unjustified self-combination step; the numerical examples are roughly reproducible but the method is not as 'completely new' or as well-validated as claimed.","tokens_in":19759,"tokens_out":9500,"would_cite":false,"duration_ms":75861,"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":"The paper claims a new combination rule that splits off conflict, weights propositions by entropy-derived reliability, and uses pignistic-OWA aggregation, so that high-conflict evidence yields the intuitively correct target rather than…","keywords":["Dempster-Shafer evidence theory","Z-number","pignistic transformation","OWA operator","Deng entropy","conflicting evidence","information fusion","multi-sensor fusion"],"falsifier":"Compute the outputs of Equations (18)–(25) on the paper's own examples, then apply Section 3.7's self-combination exactly as written; if the resulting masses do not match the published tables (for example, ξ(T)=0.4999 in Example 1), the described algorithm is incomplete or the self-combination step is doing the work attributed to the Z-number weighting.","tokens_in":18497,"feed_emoji":"🎯","tokens_out":4791,"duration_ms":398270,"temperature":0.7,"pith_summary":"This paper proposes a new rule for combining pieces of evidence within Dempster-Shafer theory. It separates the conflicting part of the evidence, measures each source's uncertainty with Deng entropy, builds a Z-number-like reliability weight for each proposition, converts multi-element masses to single-proposition probabilities via the pignistic transformation, and aggregates the weighted probabilities with an OWA operator before recombining the conflict-free part. The author argues that this pipeline keeps the ranking of Dempster's rule in low-conflict cases while avoiding its counter-intuitive outcomes in high-conflict cases, and that it produces the most decisive separation between the leading hypothesis and the rest of the alternatives in every tested example.","feed_headline":"Evidence fusion that survives high conflict: Z-number pignistic rule","feed_subtitle":"The method keeps Dempster's ranking in calm cases and picks the right target when evidence clashes.","key_machinery":"The load-bearing object is a hybrid evidence transformation: for each atomic proposition, a Z-number-like pair consisting of a pignistic probability and an entropy-derived reliability weight, where reliability is the normalized exponential of negative Deng entropy. An OWA operator aggregates these pairs across sources to produce a refined mass for each proposition; the unnormalized conflict mass from a Yager-style combination is redistributed proportionally to these refined masses; and finally Dempster's rule combines the modified evidence with itself n−1 times. The pignistic transform's role is to resolve multi-element masses onto singletons, and the OWA's role is to suppress outlier sources while preserving the overall ranking of propositions.","core_discovery":"The central claim is that evidence fusion can be made both robust to conflict and more decisive by treating reliability at the proposition level rather than only at the source level. Each mass is paired with an entropy-derived reliability weight to form a Z-number-like object; the pignistic transform flattens multi-singleton masses onto atomic propositions; an OWA aggregation over the weighted pignistic probabilities produces a refined mass assignment; and the conflict mass from an unnormalized Yager-style combination is then reallocated proportionally to these refined masses. The paper asserts that this procedure yields the correct decision in the classic high-conflict example where Dempster's rule gives a counter-intuitive answer, and that it consistently sharpens the leading proposition's mass in mild cases and in applications to iris classification and fault diagnosis.","pith_inferences":["The proposition-level reliability weighting suggests a general recipe: any conflict-management scheme that redistributes conflict mass could use the same pignistic-plus-OWA refinement, and the choice of Deng entropy as the reliability measure could be replaced by other uncertainty measures and tested for sensitivity.","The reported gain in decisiveness may depend on the self-combination step in Section 3.7 rather than on the Z-number weighting itself; this is testable by comparing the published results against a version that combines the modified evidence from all sources only once.","A natural extension would be to apply the same construction to belief functions on larger frames of discernment, where the pignistic flattening and OWA aggregation could be evaluated against probabilistic outputs of other conflict-resolution rules."],"forward_implications":["In high-conflict situations, the method identifies the hypothesis supported by the majority of the evidence rather than the counter-intuitive winner produced by Dempster's rule.","In low-conflict situations, the relative ordering of propositions matches Dempster's rule while the separation between the top proposition and the rest increases, making the decision more straightforward.","The method handles set-valued focal elements through the pignistic transform, so it does not require discarding multi-element evidence before fusion.","Applied to real datasets represented as basic probability assignments, the method yields a higher belief in the correct target than the compared combination rules."],"supporting_citations":[{"why":"Supplies the Dempster combination rule and the baseline result that the proposed method aims to improve.","marker":"[65]"},{"why":"Supplies the pignistic transformation used to resolve multi-element masses onto singleton propositions.","marker":"[67]"},{"why":"Supplies the OWA operator used to aggregate the Z-number-like reliability-weighted probabilities.","marker":"[68]"},{"why":"Supplies the Yager-style unnormalized combination rule used to isolate the conflict mass.","marker":"[69]"},{"why":"Supplies the Z-number concept that motivates pairing each mass with a reliability component.","marker":"[70]"},{"why":"Supplies Deng entropy, the uncertainty measure from which the reliability weights are computed.","marker":"[47]"},{"why":"Supplies the classic high-conflict example that the proposed method must handle correctly.","marker":"[71]"}],"fun_headline_variants":["Z-number pignistic rule: robust fusion under conflict","New fusion rule handles high-conflict evidence with Z-numbers","Pignistic Z-fusion: decisive and conflict-proof evidence merging","Z-number based pignistic transform fixes Dempster's conflict flaw","Reliability-weighted pignistic fusion wins where Dempster fails"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The procedure assumes that fusing a single synthesized evidence with itself, repeated n−1 times through Dempster's rule, acts as a legitimate sharpening of the true probabilities even though the copies are not independent sources.","fun_headline_variants_meta":{"raw":{"variants":["Z-number pignistic rule: robust fusion under conflict","New fusion rule handles high-conflict evidence with Z-numbers","Pignistic Z-fusion: decisive and conflict-proof evidence merging","Z-number based pignistic transform fixes Dempster's conflict flaw","Reliability-weighted pignistic fusion wins where Dempster fails"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000879,"raw_usage":{"total_tokens":3736,"prompt_tokens":819,"completion_tokens":2917,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":435,"completion_tokens_details":{"reasoning_tokens":2829}},"tokens_in":435,"tokens_out":2917,"duration_ms":20307,"temperature":1.0,"reasoning_tokens":2829,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T23:53:02.271586+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the outputs of Equations (18)–(25) on the paper's own examples, then apply Section 3.7's self-combination exactly as written; if the resulting masses do not match the published tables (for example, ξ(T)=0.4999 in Example 1), the described algorithm is incomplete or the self-combination step is doing the work attributed to the Z-number weighting.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Yager-style unnormalized combination rule used to isolate the conflict mass."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the classic high-conflict example that the proposed method must handle correctly."},{"cited_title":"Smets, Decision making in the TBM: the necessity of the pignistic transformation, Int","cited_arxiv_id":null,"evidence_quote":"Supplies the pignistic transformation used to resolve multi-element masses onto singleton propositions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the OWA operator used to aggregate the Z-number-like reliability-weighted probabilities."}],"review_version":1}