pith. sign in

arxiv: 2606.29878 · v1 · pith:DDMXBTW6new · submitted 2026-06-29 · 💻 cs.LG · math.OC· stat.ML

Decision-Value Attribution in Predict-then-Optimize Systems

Konstantinos Ziliaskopoulos , Alexander Vinel , Alice E. Smith This is my paper

Pith reviewed 2026-06-30 06:55 UTC · model grok-4.3

classification 💻 cs.LG math.OCstat.ML
keywords decision value attributionshapley valuespredict-then-optimizeexplanation methodsoperational decision makinginformation attributiondesign parameters
0
0 comments X

The pith

Decision Value Attribution attributes operational payoffs to predictions and design choices via Shapley games on realized decision value.

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

The paper develops Decision Value Attribution to measure how much each input or setting contributes to the final payoff of a predict-then-optimize system. Standard methods explain forecasts, yet forecast changes often leave the chosen action unchanged while small shifts can alter the realized outcome. DVA instead runs cooperative games whose payoff equals the value of the decision the optimizer actually selects. Three variants assign value to information sources, to operational configurations, or to their joint interactions. Post-DVA uses realized outcomes and pre-DVA uses the model's own predictions, turning attribution into a check on whether the model's beliefs match performance.

Core claim

Decision Value Attribution is a Shapley-based framework that attributes the value of a fixed prediction-optimization pipeline by defining cooperative games whose payoff is the downstream decision value. Players may be information sources, optimization or design parameters, or both. The method includes InfoDVA for features, DesignDVA for configurations, and Decision-Value Interactions for their joint effect. It distinguishes post-DVA, which scores decisions on realized outcomes, from pre-DVA, which scores them under the model's full prediction, and expresses all attributions in the units of the operational objective while decomposing gain or loss relative to a baseline.

What carries the argument

Decision Value Attribution (DVA), the cooperative-game construction whose payoff equals the realized value of the optimizer's selected action.

If this is right

  • Predictive explanations can be poor proxies for operational value.
  • DVA can guide targeted information-control interventions.
  • Optimization configurations determine when predictive information is decision-relevant.
  • Attributions are expressed directly in the units of the operational objective and decompose gain or loss relative to a baseline.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Systems could prioritize data acquisition by expected impact on decisions rather than by forecast accuracy alone.
  • The pre-DVA versus post-DVA split offers a practical test for when a deployed model should be retrained or overridden by real-time observations.

Load-bearing premise

Defining the cooperative-game payoff strictly as downstream decision value rather than a surrogate such as prediction error produces attributions that stay stable across baselines.

What would settle it

Compute DVA attributions using decision value as payoff, then recompute the same attributions using prediction error as payoff; if the resulting feature or parameter rankings differ materially or shift with baseline choice, the stability claim fails.

Figures

Figures reproduced from arXiv: 2606.29878 by Alexander Vinel, Alice E. Smith, Konstantinos Ziliaskopoulos.

Figure 1
Figure 1. Figure 1: Overview of the Predict-then-Optimize (PtO) pipeline. We devise Decision-Value Attribution (DVA), a Shapley￾based framework for attributing the value of a PtO pipeline to the components that generate, or shape, the downstream decision. The main idea, following the Shapley method, is to define a cooperative game whose payoff is downstream deci￾sion value. The players may be information sources, design param… view at source ↗
Figure 2
Figure 2. Figure 2: Decision regions induced by the downstream optimization problem. The axes ˆy [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: EMS decision regime heatmaps. Panel A reports realized covered demand across nine budget-radius regimes. Panel [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Pre-InfoDVA / Post-InfoDVA value comparison across features in the validation set. The highlighted quadrant is the set of points where the feature provides positive decision value according to the model predictions (pre-DVA) but negative realized decision value (post-DVA). For each qualifying feature, we compute the mean actual daily regret improvement on the validation set when that feature is masked from… view at source ↗
Figure 5
Figure 5. Figure 5: A beeswarm plot for the post-InfoDVA values on the test dataset. Each point corresponds to a single day’s SHAP [PITH_FULL_IMAGE:figures/full_fig_p024_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spatial distribution of decision value di [PITH_FULL_IMAGE:figures/full_fig_p025_6.png] view at source ↗
read the original abstract

Predictive models are increasingly embedded in operational decision-making, yet standard explanation methods typically explain forecasts rather than the decisions those forecasts induce. This distinction is important in predict-then-optimize systems: large forecast changes may leave the optimizer's action unchanged, while small changes can alter the selected decision and its realized value. We propose Decision Value Attribution (DVA), a Shapley-based framework for attributing the value of a fixed prediction--optimization pipeline. The framework defines cooperative games whose payoff is the downstream decision value, allowing the players to be information sources, optimization or design parameters, or both. We present three variants: InfoDVA attributes value to features, DesignDVA attributes value to operational configurations, and Decision-Value Interactions (DVI) quantifies how information and design jointly create value. We further distinguish post-DVA, which evaluates decisions using realized outcomes, from pre-DVA, which evaluates decisions under the model's full prediction. This separation turns attribution into a decision-level diagnostic of whether the model's operational beliefs align with realized performance. The resulting attributions are expressed in the units of the operational objective and decompose the gain or loss relative to a baseline. Case studies in electricity storage arbitrage and emergency medical service coverage show that predictive explanations can be poor proxies for operational value, that DVA can guide targeted information-control interventions, and that optimization configurations determine when predictive information is decision-relevant.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 2 minor

Summary. The paper proposes Decision Value Attribution (DVA), a Shapley-based framework for attributing the value of a fixed prediction-optimization pipeline by defining cooperative games whose payoff is the downstream decision value. Players can be information sources, optimization or design parameters, or both. It introduces three variants (InfoDVA for features, DesignDVA for operational configurations, and Decision-Value Interactions (DVI) for joint effects) and distinguishes post-DVA (using realized outcomes) from pre-DVA (using the model's predictions). Case studies in electricity storage arbitrage and emergency medical service coverage illustrate that predictive explanations can be poor proxies for operational value and that DVA can guide interventions.

Significance. If the framework and its stability claims hold, the work provides a decision-level attribution method expressed directly in operational objective units, which addresses a gap between standard prediction explanations and realized decision value in predict-then-optimize systems. The pre/post-DVA distinction offers a diagnostic for model-reality alignment, and the case studies demonstrate concrete scenarios where the approach identifies when predictive information is decision-relevant.

minor comments (2)
  1. [Abstract / §3] The abstract states that attributions 'remain stable and meaningful across different baseline choices and value-function specifications,' but the manuscript should include an explicit statement or small example (e.g., in §3 or §4) showing how the value function is defined to avoid dependence on arbitrary baselines.
  2. [Case studies] The case-study sections would benefit from a table reporting the numerical DVA attributions (in objective units) alongside the corresponding prediction-error attributions for the same instances, to make the claimed superiority of DVA over standard explanations directly comparable.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary of the manuscript, recognition of the pre/post-DVA distinction as a diagnostic tool, and recommendation for minor revision. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper proposes a new framework (DVA) by directly defining cooperative games whose payoff equals downstream decision value, with variants InfoDVA, DesignDVA, and DVI. This is a constructive definition of attribution rather than a derivation chain. No equations or steps reduce a claimed result to fitted inputs, self-citations, or prior ansatzes by construction. The pre-DVA vs. post-DVA distinction and case studies are presented as applications of the definition, not as outputs forced by the inputs. The method is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the standard axioms of cooperative game theory for Shapley values; no free parameters, invented entities, or domain-specific axioms are stated in the abstract.

axioms (1)
  • standard math Shapley value axioms (efficiency, symmetry, dummy player, additivity) apply to the defined cooperative game with decision value as payoff
    Invoked by the choice of Shapley-based attribution.

pith-pipeline@v0.9.1-grok · 5782 in / 1284 out tokens · 29977 ms · 2026-06-30T06:55:23.433355+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

66 extracted references · 1 canonical work pages

  1. [1]

    Management Science , volume=

    From predictive to prescriptive analytics , author=. Management Science , volume=. 2020 , publisher=

    2020

  2. [2]

    European Journal of Operational Research , volume=

    A survey of contextual optimization methods for decision-making under uncertainty , author=. European Journal of Operational Research , volume=. 2025 , publisher=

    2025

  3. [3]

    Advances in Neural Information Processing Systems , volume=

    Task-based end-to-end model learning in stochastic optimization , author=. Advances in Neural Information Processing Systems , volume=

  4. [4]

    Journal of Artificial Intelligence Research , volume=

    Decision-focused learning: Foundations, state of the art, benchmark and future opportunities , author=. Journal of Artificial Intelligence Research , volume=

  5. [5]

    International Conference on Machine Learning , pages=

    Decision-focused learning: Through the lens of learning to rank , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  6. [6]

    Ribeiro, Marco Tulio and Singh, Sameer and Guestrin, Carlos , booktitle=. "

  7. [7]

    Advances in Neural Information Processing Systems , volume=

    A unified approach to interpreting model predictions , author=. Advances in Neural Information Processing Systems , volume=

  8. [8]

    1953 , publisher=

    A value for n-person games , author=. 1953 , publisher=

    1953

  9. [9]

    Explaining individual predictions when features are dependent: More accurate approximations to

    Aas, Kjersti and Jullum, Martin and L. Explaining individual predictions when features are dependent: More accurate approximations to. Artificial Intelligence , volume=. 2021 , publisher=

    2021

  10. [10]

    Explainable

    De Bock, Koen W and Coussement, Kristof and De Caigny, Arno and S. Explainable. European Journal of Operational Research , volume=. 2024 , publisher=

    2024

  11. [11]

    International Conference on Machine Learning , pages=

    Explainable data-driven optimization: From context to decision and back again , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  12. [12]

    Machine Learning , volume=

    The voice of optimization , author=. Machine Learning , volume=. 2021 , publisher=

    2021

  13. [13]

    European Journal of Operational Research , volume=

    A framework for inherently interpretable optimization models , author=. European Journal of Operational Research , volume=. 2023 , publisher=

    2023

  14. [14]

    Annals of Operations Research , pages=

    Towards robust interpretable surrogates for optimization , author=. Annals of Operations Research , pages=. 2026 , publisher=

    2026

  15. [15]

    European Journal of Operational Research , year=

    Counterfactual explanations for linear optimization , author=. European Journal of Operational Research , year=

  16. [16]

    IEEE Transactions on Systems Science and Cybernetics , volume=

    Information value theory , author=. IEEE Transactions on Systems Science and Cybernetics , volume=. 1966 , publisher=

    1966

  17. [17]

    2019 , organization=

    Ghorbani, Amirata and Zou, James , booktitle=. 2019 , organization=

    2019

  18. [18]

    Forty-first International Conference on Machine Learning , pages =

    Benchmarking Deletion Metrics with the Principled Explanations , author=. Forty-first International Conference on Machine Learning , pages =

  19. [19]

    International Conference on Machine Learning , pages=

    A Consistent and Efficient Evaluation Strategy for Attribution Methods , author=. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  20. [20]

    On the (In)fidelity and Sensitivity of Explanations , pages =

    Yeh, Chih-Kuan and Hsieh, Cheng-Yu and Suggala, Arun and Inouye, David I and Ravikumar, Pradeep K , booktitle =. On the (In)fidelity and Sensitivity of Explanations , pages =

  21. [21]

    Computers & Operations Research , volume=

    Improving polynomial estimation of the Shapley value by stratified random sampling with optimum allocation , author=. Computers & Operations Research , volume=. 2017 , publisher=

    2017

  22. [22]

    Proceedings of the British Machine Vision Conference (BMVC) , year =

    Vitali Petsiuk and Abir Das and Kate Saenko , title =. Proceedings of the British Machine Vision Conference (BMVC) , year =

  23. [23]

    Biometrika , volume=

    A new measure of rank correlation , author=. Biometrika , volume=. 1938 , publisher=

    1938

  24. [24]

    SIAM Journal on Discrete Mathematics , volume=

    Comparing top k lists , author=. SIAM Journal on Discrete Mathematics , volume=. 2003 , publisher=

    2003

  25. [25]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Shapley value approximation based on k-additive games , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  26. [26]

    The Journal of Machine Learning Research , volume=

    An efficient explanation of individual classifications using game theory , author=. The Journal of Machine Learning Research , volume=. 2010 , publisher=

    2010

  27. [27]

    Sundararajan, Mukund and Dhamdhere, Kedar and Agarwal, Ashish , booktitle=. The. 2020 , organization=

    2020

  28. [28]

    Tsai, Che-Ping and Yeh, Chih-Kuan and Ravikumar, Pradeep , journal=. Faith-

  29. [29]

    2026 , howpublished =

    2026

  30. [30]

    INFORMS Journal on Computing , volume=

    Optimal hour-ahead bidding in the real-time electricity market with battery storage using approximate dynamic programming , author=. INFORMS Journal on Computing , volume=. 2015 , publisher=

    2015

  31. [31]

    Operations Research , volume=

    The value of coordination in multimarket bidding of grid energy storage , author=. Operations Research , volume=. 2023 , publisher=

    2023

  32. [32]

    Production and Operations Management , volume=

    Data-driven storage operations: Cross-commodity backtest and structured policies , author=. Production and Operations Management , volume=. 2022 , publisher=

    2022

  33. [33]

    2018 International Symposium on Power Electronics, Electrical Drives, Automation and Motion (SPEEDAM) , pages=

    Opportunities for energy storage in CAISO: Day-ahead and real-time market arbitrage , author=. 2018 International Symposium on Power Electronics, Electrical Drives, Automation and Motion (SPEEDAM) , pages=. 2018 , organization=

    2018

  34. [34]

    Energy Economics , volume=

    Large-scale battery storage, short-term market outcomes, and arbitrage , author=. Energy Economics , volume=. 2022 , publisher=

    2022

  35. [35]

    IEEE Transactions on Power Systems , volume=

    Energy storage arbitrage under day-ahead and real-time price uncertainty , author=. IEEE Transactions on Power Systems , volume=. 2017 , publisher=

    2017

  36. [36]

    Chen, Tianqi and Guestrin, Carlos , booktitle=

  37. [37]

    2024 , publisher =

    Zippenfenig, Patrick , title =. 2024 , publisher =. doi:10.5281/zenodo.14582479 , url =

    work page doi:10.5281/zenodo.14582479 2024

  38. [38]

    Manufacturing & Service Operations Management , volume=

    Real-time ambulance dispatching and relocation , author=. Manufacturing & Service Operations Management , volume=. 2018 , publisher=

    2018

  39. [39]

    Operations Research , volume=

    Ambulance emergency response optimization in developing countries , author=. Operations Research , volume=. 2020 , publisher=

    2020

  40. [40]

    Geospatial Health , volume=

    Where to place emergency ambulance vehicles: use of a capacitated maximum covering location model with real call data , author=. Geospatial Health , volume=

  41. [41]

    Frichi et al , author=

    Ambulance location and relocation under budget constraints: investigating coverage-maximization models and ambulance sharing to improve emergency medical services performance: Y. Frichi et al , author=. Health Care Management Science , volume=. 2025 , publisher=

    2025

  42. [42]

    Mathematical Programming Computation , volume=

    Parallelizing the dual revised simplex method , author=. Mathematical Programming Computation , volume=. 2018 , publisher=

    2018

  43. [43]

    and Wolsey, Laurence A

    Nemhauser, George L. and Wolsey, Laurence A. and Fisher, Marshall L. , title =. Mathematical Programming , year =

  44. [44]

    Information Processing Letters , year =

    Khuller, Samir and Moss, Anna and Naor, Joseph , title =. Information Processing Letters , year =

  45. [45]

    Biometrika , volume=

    The treatment of ties in ranking problems , author=. Biometrika , volume=. 1945 , publisher=

    1945

  46. [46]

    Machine Learning , volume=

    Random forests , author=. Machine Learning , volume=. 2001 , publisher=

    2001

  47. [47]

    Journal of Machine Learning Research , volume=

    All models are wrong, but many are useful: Learning a variable's importance by studying an entire class of prediction models simultaneously , author=. Journal of Machine Learning Research , volume=

  48. [48]

    Artificial Intelligence , volume=

    Wrappers for feature subset selection , author=. Artificial Intelligence , volume=. 1997 , publisher=

    1997

  49. [49]

    Journal of the American Statistical Association , volume=

    Distribution-free predictive inference for regression , author=. Journal of the American Statistical Association , volume=. 2018 , publisher=

    2018

  50. [50]

    Advances in Neural Information Processing Systems , volume=

    Why do tree-based models still outperform deep learning on typical tabular data? , author=. Advances in Neural Information Processing Systems , volume=

  51. [51]

    European Journal of Operational Research , volume=

    Explainable subgradient tree boosting for prescriptive analytics in operations management , author=. European Journal of Operational Research , volume=. 2024 , publisher=

    2024

  52. [52]

    The value of information in human-

    Guo, Ziyang and Wu, Yifan and Hartline, Jason and Hullman, Jessica , booktitle=. The value of information in human-

  53. [53]

    2022 4th international conference on process mining (ICPM) , pages=

    Explainable process prescriptive analytics , author=. 2022 4th international conference on process mining (ICPM) , pages=. 2022 , organization=

    2022

  54. [54]

    European Journal of Operational Research , volume=

    The many Shapley values for explainable artificial intelligence: A sensitivity analysis perspective , author=. European Journal of Operational Research , volume=. 2024 , publisher=

    2024

  55. [55]

    2020 , howpublished =

    Modified. 2020 , howpublished =

    2020

  56. [56]

    Sampling permutations for

    Mitchell, Rory and Cooper, Joshua and Frank, Eibe and Holmes, Geoffrey , journal=. Sampling permutations for

  57. [57]

    Journal of Machine Learning Research , volume=

    Explaining by removing: A unified framework for model explanation , author=. Journal of Machine Learning Research , volume=

  58. [58]

    Advances in Neural Information Processing Systems , volume=

    Understanding global feature contributions with additive importance measures , author=. Advances in Neural Information Processing Systems , volume=

  59. [59]

    2017 , publisher=

    Design and analysis of experiments , author=. 2017 , publisher=

    2017

  60. [60]

    International Journal of Game Theory , volume=

    An axiomatic approach to the concept of interaction among players in cooperative games , author=. International Journal of Game Theory , volume=. 1999 , publisher=

    1999

  61. [61]

    Nature Machine Intelligence , volume=

    From local explanations to global understanding with explainable AI for trees , author=. Nature Machine Intelligence , volume=. 2020 , publisher=

    2020

  62. [62]

    Management Science , volume=

    Multilinear extensions of games , author=. Management Science , volume=. 1972 , publisher=

    1972

  63. [63]

    Advances in Neural Information Processing Systems , volume=

    When do neural nets outperform boosted trees on tabular data? , author=. Advances in Neural Information Processing Systems , volume=

  64. [64]

    2025 , month =

    2024 Special Report on Battery Storage , author =. 2025 , month =

    2024

  65. [65]

    Journal of Energy Storage , volume=

    Impact of battery degradation on energy arbitrage revenue of grid-level energy storage , author=. Journal of Energy Storage , volume=. 2017 , publisher=

    2017

  66. [66]

    2025 , month =

    Cost Projections for Utility-Scale Battery Storage: 2025 Update , author =. 2025 , month =

    2025