{"id":"55f6d23b-64e7-4159-b655-cff92709bc17","arxiv_id":"2607.29077","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Counterfactual explanation by distance minimization is re-derived as MAP inference in a generalized Bayes posterior, and additional posterior-based decision rules are proposed and evaluated.","lead":"The paper shows that the standard counterfactual-explanation objective—a weighted sum of prediction loss and change cost—is the maximum a posteriori estimate of a Gibbs posterior whose prior is chosen to be the negative exponential of the change cost. It then proposes Bayes and CVaR decision rules and a model-mixture extension, and tests them on simulated and Google Trends data.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 1 is correct but the claimed 'theoretical foundation' is a definitional identity: the prior exp(-d) is chosen to force the MAP to equal the CE objective, so the equivalence does not independently justify distance-minimization CE.","rationale":"The reader's weakest_assumption pinpoints that the equivalence rests entirely on the prior choice, making the 'theoretical foundation' claim definitional rather than substantive. I agree with this assessment. The proposition itself is correct, so there is no mathematical error; the issue is overclaimed novelty. The empirical unreliability of the 10D MAP approximation is a secondary concern that reinforces the need for caution but does not change the central analytical point. Since the paper's extensions (Bayes decision, CVaR-CE, ModelUnc) are presented and evaluated independently of the depth of the equivalence, a CONDITIONAL verdict remains appropriate: the paper is honest and the methods are reasonable, but the core equivalence is trivial. The proposed concrete test would determine whether the posterior machinery yields any practical advantage over direct optimization, thereby settling whether the equivalence is load-bearing for the paper's contributions.","tokens_in":16825,"tokens_out":14235,"duration_ms":152904,"concrete_test":"Implement the CVaR-CE decision rule directly on the original CE objective: minimize CVaRτ(ℓ(f(z+δ),y*)) over z in the input space, without restricting z to posterior samples and without forming a Gibbs posterior. Compare its performance (Lpt, Rb, Dpt) to the paper's posterior-restricted CVaR-CE on the same simulated and real datasets. If the direct version matches or outperforms, then the posterior framework adds no decision-theoretic value beyond the original objective, confirming that the equivalence is purely re-descriptive and that the 'theoretical foundation' claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 1 is mathematically correct: with prior p(˜x|xb) ∝ exp(−d(xb,˜x)), the MAP of the Gibbs posterior is argmin[ηℓ + d]. However, the log-prior is exactly −d, so the equivalence is an identity by construction. For any regularized objective of the form min[ηℓ + r(xb,˜x)], one can define a prior exp(−r) and obtain the same MAP. The paper concedes that p(˜x|xb) is user-specified, and the variational derivation in the appendix (Eq. 20) is a generic property of Gibbs posteriors, not a justification of the distance prior specifically. Thus the central claim to provide a 'theoretical basis' or 'probabilistic foundation' for distance-minimization CE overstates the novelty: the equivalence is a re-description, not a derivation. The genuine contributions—Bayes decision, CVaR-CE, and ModelUnc—are independent of whether this equivalence is deep, but the paper's framing makes the definitional identity load-bearing for its claimed significance.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a generalized-Bayes interpretation of counterfactual explanations. It defines a Gibbs posterior over counterfactuals as p(˜x|xb,y*) ∝ exp(−ηℓ(f(˜x),y*)) p(˜x|xb), chooses a distance-based prior p(˜x|xb) ∝ exp(−d(xb,˜x)), and shows in Proposition 1 that the MAP estimate equals the standard CE objective argmin[ηℓ + d]. Building on this posterior, the paper introduces two additional decision rules—a Bayes decision that minimizes posterior expected decision loss and CVaR-CE, a risk-averse rule—together with a ModelUnc construction that mixes model-specific posteriors using Bayesian weights to address model multiplicity. It defines pointwise and distributional evaluation metrics and reports experiments on 2D and 10D simulated data and on Google Trends data. The central claim is that this equivalence provides a theoretical/probabilistic foundation for conventional distance-minimization CE.","tokens_in":17179,"tokens_out":4506,"duration_ms":49044,"significance":"Proposition 1 is mathematically correct, and the paper is honest about several limitations, including the difficulty of MAP estimation from finite samples in high dimensions and the arbitrariness of the temperature η. The genuine contributions are the distributional decision rules (especially CVaR-CE), the model-uncertainty mixture in Eq. (8), and the distribution-level evaluation metrics. The model-agnostic nature of the framework is a practical strength. However, the paper overstates the novelty of Proposition 1: because the prior is defined as exp(−d), the equivalence is a definitional identity rather than an independent justification of distance-minimization CE. This overreach is load-bearing for the paper's stated significance and needs to be reframed. The empirical support is also limited by single-seed runs and by the acknowledged unreliability of the 10D MAP approximation.","major_comments":[{"comment":"The claimed 'theoretical basis' for distance-minimization CE is a definitional identity. The prior p(˜x|xb) ∝ exp(−d(xb,˜x)) is chosen so that the log-prior equals −d, making the MAP of the Gibbs posterior equal to argmin[ηℓ+d] by construction. Any regularized objective of the form min[ηℓ+r(xb,˜x)] can be rewritten in the same way by setting the prior to exp(−r). The paper acknowledges that the prior is user-specified, but the abstract and Section 5 nevertheless claim that this 'provides a broader theoretical foundation.' This overstates what is shown. I recommend reframing Proposition 1 as a representational equivalence—distance-minimization CE is the MAP of a Gibbs posterior under an exponential distance prior—rather than as an independent justification of the distance-minimization formulation.","section":"Section 2.3, Eq. (5), Proposition 1"},{"comment":"The empirical comparison of MAP is not informative in 10D. The reported Gibbs MAP has Lpt=1.816 and Rb=0.000, which the paper attributes to the difficulty of mode estimation from finite samples using KDE/kNN. This is a candid limitation, but it means that Table 3 cannot support the claim that the approximate Gibbs MAP behaves as Proposition 1 predicts, nor can it support substantive comparisons between MAP and other decision rules in 10D. The paper should either replace the sample-based mode with a more reliable MAP estimate (e.g., direct constrained optimization of the objective or a better mode-seeking procedure) or explicitly state that the 10D MAP rows are not a validation of Proposition 1 and should not be interpreted as such.","section":"Section 3.1.2, Table 3"},{"comment":"All experiments use a single random seed (seed 42), and the thresholds and hyperparameters are hand-set (εsp/εrb = 0.25/0.35/15.0, τ = 0.9, σδ = 0.2/0.15/5.0). The paper reports means and standard deviations over base points but not over repeated seeds or over the stochastic sampling/optimization procedures. Since the paper makes quantitative claims about trade-offs (e.g., ModelUnc increases Tail and Stability, CVaR-CE improves robustness at greater distance), the absence of any repeated-seed or bootstrap uncertainty weakens the empirical conclusions. I am not demanding a full benchmark study, but a small multi-seed or bootstrap analysis would materially improve the reliability of the reported differences.","section":"Sections 3.1–3.2"}],"minor_comments":[{"comment":"The heading reads 'Ten-Dimensional Data (Table 2)' but should refer to Table 3.","section":"Section 3.1.2 heading"},{"comment":"The sentence 'The 2D analysis uses only x′1, summer vacation_lag1, and x′2, Jump_lag1' is awkwardly phrased; it should say the 2D analysis uses two variables, summer vacation_lag1 and Jump_lag1.","section":"Section 3.2"},{"comment":"The italicized comment on line 1 ('Match the proposal distribution to the distance prior ∝ exp(−d(xb, ˜x))') is a useful clarification, but it would be better placed in the main text or as a footnote to avoid confusion about whether it is part of the algorithm's pseudocode.","section":"Algorithm 1"},{"comment":"For P lu, the notation ∥˜x−xi∥ appears without a norm subscript; specifying the norm (Euclidean) at first use would improve clarity.","section":"Section 2.5.2"},{"comment":"The variational derivation in Eq. (20) is a standard Donsker–Varadhan result. It would be helpful to state explicitly that this derivation applies to any loss and prior and thus does not specifically justify the distance prior—this connects to Major Comment 1.","section":"Appendix"}],"recommendation":"major_revision","confidential_remarks":"The paper's framing overstates the depth of the theoretical contribution, but the underlying framework and decision rules are reasonable and potentially useful. The main fix is reframing Proposition 1 as a representation theorem and tightening the empirical claims around the 10D MAP. I do not see grounds for rejection, but the revision should be carefully checked for consistency between the abstract's 'theoretical basis' claim and the actual content."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is honest and technically correct, but the central \"theoretical basis\" claim is a re-description, not a derivation. Choosing the prior p(x~|xb)∝exp(−d) makes the MAP of the Gibbs posterior exactly argmin[ηℓ+d] by construction. Any regularized objective can be written this way, and the paper itself concedes the prior is user-specified. So Proposition 1 is true, but calling it a probabilistic foundation overstates what's new.\n\nWhat is genuinely useful: the posterior view does open the door to decision rules beyond MAP. The CVaR-CE rule, which minimizes the tail of the perturbed loss, is a sensible risk-averse choice. The ModelUnc mixture, averaging posteriors across models with Bayesian weights, is a practical way to address model multiplicity. The distribution-level metrics (SP, Tail, Stability, VarImp) are reasonable and clearly defined. The writing is clear, and the authors are admirably candid about limitations—they explicitly flag the 10D MAP approximation as unreliable and report the failure in Table 3.\n\nThe soft spots are real but proportionate. The empirical section is single-run (seed 42), with hand-set thresholds, and no code is released. That makes the quantitative comparisons illustrative rather than robust. The 10D Gibbs MAP result (Lpt=1.816, Rb=0.000) is an artifact of density estimation from finite samples, so the decision-rule comparisons in Table 3 are partly comparing estimation failures. ModelUnc is only tested with equal weights, not the performance-based weights. The CVaR-CE implementation restricts candidates to posterior samples, which is pragmatic but could miss better points outside the sample.\n\nIs it worth taking seriously? Yes, but as a modest contribution. The framing is coherent, the extensions are reasonable, and the related work is handled fairly. The authors' claims of a \"broader theoretical foundation\" should be tempered. With code and multi-seed results, this could be a useful paper for practitioners who want to move from point CEs to distribution-aware decisions.\n\nMy recommendation: send it to a serious referee. The work is not groundbreaking, but it is clearly presented and the CVaR-CE and model-mixture ideas deserve engagement. The referee should press for code release, repeated seeds, and a more careful separation of the definitional identity from the genuine extensions.","headline":"The MAP–Gibbs equivalence is a definitional identity, not a deep theoretical result; the paper's real value is the CVaR-CE and model-mixture extensions, though the empirical support is thin.","tokens_in":17612,"tokens_out":1814,"would_cite":false,"duration_ms":22042,"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 proves that the standard distance-minimization counterfactual explanation is exactly the maximum a posteriori estimate of a Gibbs posterior built from a distance-based prior, and uses that equivalence to derive new decision rules","keywords":["counterfactual explanations","generalized Bayes","Gibbs posterior","MAP estimation","model multiplicity","CVaR","algorithmic recourse","Bayesian decision theory"],"falsifier":"Run the paper's importance-sampling MAP approximation side by side with direct multistart optimization of the same objective on many base points. If the sample-based MAP is consistently far from the direct optimum (as the paper's own 10D results already suggest, e.g., Gibbs MAP loss 1.816 versus direct optimization loss 0.017), then the practical claim that the method recovers the posterior mode is not supported; the theoretical equivalence itself would remain intact, but its algorithmic realization would be falsified.","tokens_in":16739,"feed_emoji":"🎯","tokens_out":4989,"duration_ms":49750,"temperature":0.7,"pith_summary":"The paper tries to establish that the common practice of finding a counterfactual by minimizing a weighted sum of goal-attainment loss and change distance is not just a heuristic: it is exactly MAP estimation of a Gibbs posterior when the prior is chosen as the exponential of the distance. If true, every existing distance-minimizing counterfactual method inherits a probabilistic interpretation, and the full posterior over counterfactual candidates becomes available for decision-making. The paper builds on this posterior to define three decision rules—MAP, posterior-mean Bayes, and risk-averse CVaR—along with distribution-level metrics such as success probability, tail loss, stability, and variable importance. A reader should care because this recasts a widely used explainability tool as a Bayesian inference problem, enabling model averaging across multiple competing models and principled handling of recourse uncertainty.","feed_headline":"Proof: distance-minimizing counterfactuals are Bayesian MAP estimates","feed_subtitle":"The equivalence turns standard recourse into a posterior problem, enabling decision rules like CVaR and model-averaged explanations.","key_machinery":"The load-bearing object is the Gibbs posterior over counterfactual candidates, p(˜x|xb, y*) ∝ exp(−ηℓ(f(˜x), y*)) exp(−d(xb, ˜x)), with the exponential of the change distance serving as the prior. Proposition 1 shows that maximizing this posterior is identical to minimizing the conventional counterfactual objective. In experiments, the posterior is approximated by importance sampling from a Gaussian proposal centered at the base point, after which decision rules (MAP via density estimation, posterior mean, CVaR-CE) and distribution-level metrics are applied. The model-uncertainty extension, ModelUnc, replaces the single loss with a weighted mixture of exponentials from multiple models, refle","core_discovery":"The central discovery is Proposition 1: the minimizer of ηℓ(f(˜x), y*) + d(xb, ˜x) is exactly the maximizer of the Gibbs posterior p(˜x|xb, y*) ∝ exp(−ηℓ(f(˜x), y*)) exp(−d(xb, ˜x)), formed by exponentiating the goal-attainment loss and using the distance as a prior. The paper calls this the Distance-Prior Generalized Bayes CE (DP-GBCE). This is an exact algebraic equivalence, not a numerical approximation, and it holds for any loss and any distance that make the posterior integrable. On this basis the paper argues that distance-minimization counterfactual explanations are a special case of probabilistic inference, and that the entire posterior distribution—not just its mode—is a legitimate","pith_inferences":["The equivalence is reversible: any objective of the form loss plus regularizer can be read as a MAP problem by taking the prior proportional to the exponential of the regularizer. This means the Bayesian framing supplies a vocabulary and decision-theoretic tools, while the substantive novelty lies in the decision rules and metrics built on the posterior.","The posterior-mean decision is prone to landing in low-density, high-loss regions when the success region is multimodal or curved; the paper's own experiments show this failure. A natural extension is a constrained posterior mean or a decision rule that explicitly forbids low-density regions.","CVaR-CE depends on an assumed execution-noise distribution; matching that distribution to observed deployment noise rather than using an isotropic normal is a direct, testable refinement.","The ModelUnc mixture is an OR-type aggregation—it assigns mass to regions where at least one model succeeds—so it does not provide worst-case guarantees. Replacing the mixture with a loss-side aggregation such as exp(−η max_k ℓ_k) would yield a more conservative posterior, a step the paper leaves for future work."],"forward_implications":["Every existing distance-minimizing counterfactual method is, in effect, computing a MAP estimate of a Gibbs posterior; the equivalence provides a probabilistic justification for the cost-minimization formulation.","Replacing the point estimate with the full posterior enables two additional decision rules—Bayes decision (posterior mean or geometric median) and CVaR-CE—which trade off goal attainment, change distance, and execution robustness differently.","Using a mixture of model-specific posteriors with Bayesian weights gives a distributional counterfactual that explicitly accounts for model multiplicity, and the paper shows that this mixture increases tail risk and posterior dispersion compared with single-model posteriors.","Distribution-level metrics (success probability, tail loss, stability, and distributional variable importance) make it possible to evaluate the whole counterfactual posterior, not just one explanation, and to identify which variables move most in constructing recourse.","The framework is model-agnostic: it requires only that the loss can be evaluated, not differentiability or a generative model, so it applies to black-box predictors such as tree ensembles and external APIs."],"fun_headline_variants":["Distance-minimizing CEs are Bayesian MAP estimates","Counterfactual explanations are Gibbs posterior modes","Recourse as posterior inference: beyond MAP decisions","Bayes view explains distance-minimized counterfactuals","Generalized Bayes: counterfactuals as posterior decision"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire equivalence rests on choosing the prior over counterfactuals to be exactly the exponential of the change distance; if that prior is not a faithful representation of what counts as a feasible or natural change, the posterior and the decision rules inherit that arbitrary choice.","fun_headline_variants_meta":{"raw":{"variants":["Distance-minimizing CEs are Bayesian MAP estimates","Counterfactual explanations are Gibbs posterior modes","Recourse as posterior inference: beyond MAP decisions","Bayes view explains distance-minimized counterfactuals","Generalized Bayes: counterfactuals as posterior decision"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000503,"raw_usage":{"total_tokens":2305,"prompt_tokens":766,"completion_tokens":1539,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":510,"completion_tokens_details":{"reasoning_tokens":1465}},"tokens_in":510,"tokens_out":1539,"duration_ms":11946,"temperature":1.0,"reasoning_tokens":1465,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T14:02:19.988809+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the paper's importance-sampling MAP approximation side by side with direct multistart optimization of the same objective on many base points. If the sample-based MAP is consistently far from the direct optimum (as the paper's own 10D results already suggest, e.g., Gibbs MAP loss 1.816 versus direct optimization loss 0.017), then the practical claim that the method recovers the posterior mode is not supported; the theoretical equivalence itself would remain intact, but its algorithmic realization would be falsified.","supporting_citations":[],"review_version":1}