acceptodds
Under review as a conference paper at ICLR 2027

When Pairwise Counterfactual Bounds Are Exact: Conflict Certificates and Temporal Compression

Abstract

Pairwise counterfactual bounds can combine incompatible response mechanisms. We characterize when a supplied query attains this relaxation and construct a compact attained representation of the complete sharp function for a structured temporal class. For rational mechanism polytopes, occupied Bellman residuals and a backward no-leakage recursion give an exactness criterion and recover a rational witness, with polynomial bit cost at fixed coordinate dimension. For a structured three-state, binary-successor, same-action model with independent untied time laws and nonnegative rational one-defect terminal costs, original-domain repair, control of nonmetric histories, and first-entry counting bound the entire representation by , where counts exceptional layers. The guarantee transfers to standard exact full-cone pruning. Same-support complete-atom transport gives certified enclosures, with error scaled by terminal range and entrance mass. Coverage and chronological attainment are checked separately. Fixed three-state instances require attained values. A public token-circulation protocol supplies native dynamics for a shared-response completion query. Synthetic comparisons measure a block/generic cold-cost ratio of and changes in optimal route designs; supplied-query batches favor diagnosis or direct search. A growing-state construction establishes NP-hard evaluation and an exponential attained-family boundary.

Primary area:
causal reasoning
Keywords:
counterfactual inference, causal inference, partial identification, structural causal models, dynamic counterfactuals, sharp bounds, dynamic programming, exact certification

Reviews

Among:
Shape:
1234this paper, Bottom 10%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 1.67 over 3 reviews, above 10% of the others.
Reviewer_WghVOpenReview ↗
Rating
1
Confidence
2
Informativeness
1
Presentation
1
Soundness
3
Read review
Paper summary

The paper studies pairwise counterfactual bounds in temporal models, but its presentation is extremely poor and its practical contribution is limited.

Critical strengths

It presents a new model for sequential decision-making.

Critical weaknesses
  1. The paper reads as largely AI-generated and is extremely hard to follow. Many terms are neither formally defined nor explained. The sentences are also poorly connected, and the text contains a lot of defensive writing.
  2. The setting is too restricted to be of interest to the ICLR community.
Questions for authors

I have no questions whose answers are likely to change my assessment.

Reviewer_QhrrOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
1
Soundness
2
Read review
Paper summary

This paper studies the exactness of pairwise counterfactual relaxation bounds. The main contributions include: (1) a general exactness criterion based on support-level conditions (Theorem 2), and (2) a temporal compression theorem showing that for a structured three-state model, the sharp bound representation has polynomial size in the horizon when the number of exceptional layers is fixed (Theorem 4). Complexity hardness results and synthetic experiments are also provided.

Critical strengths
  1. The exactness criterion and the compression theorem are technically substantial, with thorough proofs provided in the appendix.

  2. The NP-hardness result for variable state counts complements the positive compression result and helps delineate tractability.

Critical weaknesses
  1. The paper is very difficult to read for the broader community. The writing is extremely dense, with heavy notation and many non-standard terms introduced before being motivated. The abstract contains specific constants (392, 490) that offer no conceptual insight. A clear running example and more intuitive exposition are needed. In its current form, the paper is not accessible to the general ICLR audience.

  2. No experiments on real or realistic data. All evaluations are on synthetic inputs or a small protocol model. The paper does not demonstrate that the framework is relevant to practical counterfactual inference problems. Without such evidence, the practical impact of the theoretical results remains unclear.

  3. Narrow structural assumptions. The compression theorem applies only to a very specific setting (three states, same action, binary successors). The paper does not convincingly argue why this subclass is broadly important or how the results might generalize.

Questions for authors
  1. Can you identify a concrete real-world application where the assumptions of Theorem 4 naturally hold?

  2. Could the paper be restructured with a simple motivating example early on to improve accessibility?

  3. Is there any path toward relaxing the three-state or binary-successor assumptions?

Reviewer_WWfbOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
1
Soundness
2
Read review
Paper summary

The paper studies when a pairwise relaxation of a counterfactual bound can be attained by one shared response mechanism. For a structured three-state example, it constructs a family of vectors representing the sharp value for each entrance distribution, which is polynomial in the horizon and exponential in the number of exceptional layers. It also provides hardness results and empirical comparisons of diagnosis, direct search, and compilation.

Critical strengths
  • The paper considers an interesting question concerning the compatibility of pairwise counterfactual bounds.
  • The paper provides theoretical analysis and empirical evaluations on synthetic examples.
Critical weaknesses
  1. The presentation is fragmented and difficult to follow. The abstract and introduction introduce many undefined terms without explanation. For example, in the abstract, nn is undefined in the bound (392n+81)490r(392n + 81)490^r and terms such as "pairwise relaxation", "occupied residual", and the structured three-state examples (lines 47-50) appear without any intuitions. I do not believe the paper in its current form is accessible to general readers.
  2. The paper lacks intuition and specific examples to illustrate its ideas, and some definitions are unclear. For example, Definition 3 seems to rely on the previous three-state example, but this dependency is never clarified. Also, the notions of “omission” and “fair” are not clearly defined.
  3. The main result (Theorem 4) only considers a three-state model with specific restrictions, which greatly limits its scope.
  4. More motivations should be provided. For example, the discussion beginning near line 264 follows Theorem 4 without clearly stating whether it illustrates the theorem, motivates an assumption, or begins its proof.
Questions for authors
  1. How practical is the bound (392n+81)490r(392n + 81)490^r? It seems that it only works when rr is small.
  2. How difficult would it be to generalize the compression result beyond the structured three-state setting? Which parts rely on the three-state assumption?

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.