Pith. sign in

REVIEW 3 major objections 5 minor 45 references

For conditional exponential-family models, a falsifier that samples parameters and checks candidate formulas almost surely verifies them: identifying formulas pass, non-identifying formulas fail with probability one.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Verification of candidate causal formulas is formalized as a decision problem, and a Gaussian/exponential-family falsifier almost surely certifies whether a formula matches the interventional target within the chosen parametric family.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection Solid new framing of verification as a standalone problem, with a sound but parametric-only guarantee that the abstract and gateway-test claims overstate. the 3 major comments →

arxiv 2607.13883 v1 pith:PPKP32QV submitted 2026-07-15 stat.ME cs.AIcs.LGstat.ML

Verifying formulas for interventional distributions

Francesco Freni , Leonard Henckel , Sebastian Weichwald This is my paper
classification stat.ME cs.AIcs.LGstat.ML MSC 62D20
keywords causal inferenceverificationfalsificationidentificationinterventional distributionscausal graphical modelsexponential familiesfront-door formula
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The reading

This paper introduces verification as a decision problem in causal graphical models: given a graph, a treatment/outcome pair, and a candidate observational formula, decide whether the formula identifies the target interventional distribution. It shows that existing identification machinery does not answer this question, and that direct do-calculus proof search can certify but not refute formulas. As a practical route, it proposes a falsifier that samples parameters from a conditional exponential-family submodel and checks whether the candidate formula agrees with the interventional target; under regularity assumptions, this falsifier is an almost-surely correct verifier relative to that submodel (Theorem 4.9). This yields the gateway test, which enumerates all front-door candidate sets and returns exactly those whose formulas are identifying, showing that verification can also characterize entire formula classes.

Core claim

The central claim is Theorem 4.9: for conditional exponential-family parametrizations satisfying Assumptions 4.4–4.7, the falsifier of Definition 4.1 induces an almost-surely correct verifier relative to the parametric family PΘ(XV). Concretely, an identifying formula is never rejected, while a non-identifying formula is rejected with probability one over the sampled parameters, because the set of parameters on which a non-identifying formula accidentally matches the target has Lebesgue measure zero (Proposition 4.8). The paper also shows verification is not a by-product of identification: sound and complete identification procedures need not decide whether an arbitrary given formula is iden

What carries the argument

The falsifier (Definition 4.1) samples independent parameter values θ1,…,θK from an absolutely continuous prior over a parametric submodel PΘ(XV) of the causal graph and checks, for each sample, whether the candidate formula's output equals the interventional target density for all intervention values. The proof rests on analyticity: under a conditional exponential-family parametrization with open connected parameter space and analytic natural parameters, both the candidate formula's output and the interventional density are real-analytic functions of θ (Assumptions 4.4–4.7); a nonzero real-analytic function on an open connected domain has measure-zero zero set, so accidental agreement is ne

Load-bearing premise

The load-bearing premise is Assumption 4.6: for every intervention value, the post-intervention distribution family must form a regular exponential family with open natural parameter space, and correctness is only claimed relative to the user-supplied parametric family PΘ(XV); if that family is too narrow to represent the true model, a formula that passes the falsifier may not be identifying in the full non-parametric causal model.

What would settle it

A concrete check would be to take a regular exponential-family model satisfying Assumptions 4.4–4.7, choose a formula known to be non-identifying, and symbolically compute the set of parameters where it agrees with the target interventional density; Proposition 4.8 predicts this set has Lebesgue measure zero. In the linear Gaussian implementation, the same test is exact: sample parameters from a fine integer grid and verify that the non-identifying formula is rejected unless the polynomial discrepancy is identically zero.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Verification is a separate decision problem: even a sound and complete identification algorithm does not decide whether an arbitrary candidate formula is identifying, so new tools are needed.
  • Do-calculus proof search is only semi-decidable: it can eventually certify derivable identifying formulas, but may run forever on non-identifying formulas.
  • Under Assumptions 4.4–4.7, the falsifier is an almost-surely correct verifier relative to the parametric family: no false rejections, and false acceptances have probability zero.
  • The gateway test enumerates and verifies all front-door candidate sets; it returns exactly the sets whose front-door formulas are identifying, so the front-door criterion itself is not exhaustively complete relative to that formula class.
  • In the linear Gaussian implementation, verification reduces to polynomial identity testing, and sampling parameters from finite integer sets with exact arithmetic gives a concrete bound on the probability of falsely accepting a non-identifying formula.

Where Pith is reading between the lines

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

  • Because the guarantee is relative to the user-chosen parametric family, a certified formula may fail outside that family; the paper leaves open when parametric correctness transfers to the full non-parametric model. A practical extension is to run the falsifier over a sequence of growing parametric families and check whether certified formulas stabilize.
  • The analyticity argument is not tied to Gaussianity: any conditional exponential family satisfying the regularity assumption should admit the same almost-sure falsifier, so Poisson-type or discrete exponential-family implementations are a natural next step.
  • The enumeration idea behind the gateway test extends to other finite formula classes, such as adjustment formulas on acyclic directed mixed graphs, where no sound and exhaustively complete graphical criterion is currently known; verification could fill that gap.
  • One could also test the falsifier's one-sided behavior empirically: for a known non-identifying formula, the rejection probability should approach one as sampling becomes dense; deviations would indicate a violation of the regularity assumptions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper introduces and formalizes the verification problem for causal graphical models: given a graph, a target interventional distribution, and a candidate observational formula, decide whether that formula identifies the target. It first argues that sound and complete identification procedures do not solve verification, and that do-calculus proof search is only semi-decidable. It then proposes a falsifier that samples parameters from a user-chosen parametric family PΘ(XV) and checks agreement between the candidate formula and the post-intervention density. The central theoretical result, Theorem 4.9, states that, under Assumptions 4.4–4.7 on conditional exponential-family parametrizations, the falsifier is an almost-surely correct verifier relative to PΘ(XV). The paper applies this falsifier to the class of front-door formulas, introducing the gateway test that enumerates candidate mediator sets and reports all sets whose front-door formula is accepted. An extensive appendix supplies proofs, a full grammar for observational formulas, a Gaussian closure result, and a Schwartz–Zippel-based finite-sample probability bound for the implemented exact-arithmetic version.

Significance. The formalization of verification as a distinct decision problem is a useful contribution, and the paper's main theorem is internally sound under its stated assumptions. The work also provides concrete technical tools of independent value: a typed, kernel-preserving grammar for observational formulas; a Gaussian closure lemma; a detailed proof that non-identifying formulas fail on a measure-zero parameter set; and a careful treatment of the gap between idealized almost-sure guarantees and floating-point/tolerance-based implementations. The front-door examples in Section 5 and Appendix H.1 are interesting and go beyond the classical front-door criterion. However, the advertised claims in the abstract and Section 5 are stronger than what Theorem 4.9 proves: correctness is established only relative to the chosen parametric submodel PΘ(XV), not relative to the full non-parametric model PG(XV). Since Section 6 explicitly leaves transfer to PG(XV) open, the word 'verifier' as defined in Definition 3.1 is not currently justified by the theorem. The central ideas are valuable, but the manuscript needs to recalibrate its claims.

major comments (3)
  1. [§4.2 / Theorem 4.9 vs. Definition 3.1] Theorem 4.9 proves that the falsifier is an almost-surely correct verifier relative to PΘ(XV), but Definition 3.1 defines a verifier over the full non-parametric model PG(XV). These are different objects. If PΘ is a proper subfamily on which a non-identifying formula φ agrees with the target for every θ (for example, if the subfamily imposes constraints that force identification), the falsifier accepts with probability one, even though φ is not identifying in PG(XV). Proposition 4.8 only gives a measure-zero set when a witness θ* exists inside Θ; it says nothing when the disagreement occurs only outside Θ. Section 6 acknowledges this by stating that transfer from PΘ(XV) to PG(XV) remains open. The abstract and Section 5 must therefore not describe the procedure as a verifier in the sense of Definition 3.1, or must include the qualifier 'relative to PΘ(XV)' in every such claim.
  2. [Abstract and §5] The abstract states that the falsifier 'induces an almost-surely correct verifier for regular exponential-family models' and that the gateway test 'finds all sets admissible for use in a front-door formula'. These claims omit the essential qualifier 'relative to the chosen parametric family PΘ(XV)'. Section 5 similarly says that with an exact verifier the gateway test returns exactly the identifying front-door sets, but no exact full-model verifier is constructed: Theorem 4.9 is only almost-surely correct relative to PΘ under Assumptions 4.4–4.7. The false-acceptance scenario above is therefore not addressed by the theorem. The manuscript should either prove transfer to PG(XV) or consistently state that the gateway test is sound and complete only within the parametric submodel used by the falsifier.
  3. [§5 / Appendix G] Section 5 says that in practice, when the falsifier is used, the gateway test is 'almost-surely sound and exhaustively complete' relative to the chosen exponential family. The actual implementation in Appendix G samples parameters from finite integer sets and uses exact arithmetic; Theorem G.3 only bounds the probability of false acceptance by (2n−1)(1+∏γ_l)/min|A_i|. A single run therefore does not achieve the almost-sure guarantee of Theorem 4.9, which assumes absolutely continuous parameter sampling and exact evaluation. The manuscript is transparent about this in Appendix G, but Section 5 should reference the finite-sample bound and should not present the implemented gateway test as literally almost-surely correct without qualification.
minor comments (5)
  1. [Abstract] The phrase 'regular exponential-family models' should be 'conditional exponential-family parametrizations satisfying Assumptions 4.4–4.7', since Assumption 4.6 also requires the post-intervention family to be a regular exponential family for every intervention value t.
  2. [§4.2] The title 'Almost-sure correct verifier' should be qualified as 'relative to PΘ(XV)' to avoid confusion with Definition 3.1.
  3. [Appendix E.2] Minor typo in the proof of Proposition 4.8: 'pΔ .0' appears to mean 'pΔ is not the zero polynomial'; please correct the formatting.
  4. [§4.1 / Definition 4.1] Definition 4.1 says 'Let G be any latent-variable causal directed acyclic graph...' and then fixes a parametric family depending on G. Since the almost-sure guarantee is relative to that chosen family, the definition should explicitly state that the falsifier is parameterized by the chosen graph and parametric family, rather than suggesting that the choice of G is immaterial.
  5. [§5] The sentence 'with an exact verifier, the gateway test returns exactly those candidate sets whose front-door formulas are identifying' should be flagged as conditional on the existence of an exact verifier, which the paper does not provide; Theorem 4.9 only gives an almost-sure verifier relative to PΘ.

Circularity Check

0 steps flagged

No significant circularity: Theorem 4.9 adds a genuine measure-zero genericity argument rather than restating Definition 4.1, and the parametric-relative scope is explicitly disclosed.

full rationale

The paper's central derivation is not circular. Definition 4.1 specifies the falsifier's output as agreement between the candidate formula and the interventional target at finitely many sampled parameter values. Theorem 4.9 claims that, under Assumptions 4.4–4.7, this falsifier is an almost-surely correct verifier relative to PΘ(XV). That claim is not true by definition: the nontrivial content is Proposition 4.8, which shows that for a non-identifying formula the set S of parameters where the formula happens to agree with the target has Lebesgue measure zero, using analyticity of conditional exponential families and the identity theorem. The sampled-parameter agreement in Definition 4.1 and the relative-identification property in Definition 2.2 are different objects, connected by a real measure-zero proof rather than by construction. No fitted parameter is renamed as a prediction: parameters are sampled from an absolutely continuous prior, not fitted to data. The load-bearing external inputs are the ID algorithm (Shpitser and Pearl, 2006), Barndorff-Nielsen's exponential family theory, Boeken et al.'s analyticity results, and the Schwartz–Zippel lemma; none of these are self-citations of the present authors, and no argument reduces to an unverified claim by the authors. The main caveat—that correctness is relative to the chosen parametric family and need not transfer to the full non-parametric model PG(XV)—is stated explicitly in Section 6: 'it need not be valid in the full non-parametric graphical model PG(XV). Characterizing when correctness relative to PΘ(XV) transfers to correctness in PG(XV) remains open.' This is an honest scope limitation, not a hidden circular step. Likewise, the gateway test's guarantees are explicitly qualified as relative to the conditional exponential family used by the falsifier. The paper may overstate the abstract's 'verifier' phrasing relative to the non-parametric Definition 3.1, but that is a correctness or presentation risk, not circularity. Accordingly, no circularity steps are identified and the score is 0.

Axiom & Free-Parameter Ledger

0 free parameters · 6 axioms · 0 invented entities

No empirical data or fitted constants appear in the central result. The almost-sure correctness is conditional on the user-chosen exponential-family submodel and its regularity assumptions; the gateway test is exhaustive only relative to that submodel unless an exact verifier is used.

axioms (6)
  • domain assumption Causal DAG semantics: densities factorize according to G and interventions are truncated factorizations (g-formula).
    Section 2 and Appendix A.2 define the object being verified; all causal claims are relative to this semantics.
  • domain assumption Identifiability depends only on the latent projection Gp of G onto observed variables.
    Invoked after Definition 2.2 via Richardson et al. (2023, Corollary 49); the verifier's input is a latent projection.
  • ad hoc to paper The parametric family admits a conditional exponential-family parametrization satisfying Assumptions 4.4-4.7.
    These regularity conditions are imposed by this paper; if they fail, Proposition 4.8 and Theorem 4.9 do not follow.
  • domain assumption Gaussian closure: under a Gaussian observational law, every admissible formula returns a linear Gaussian kernel.
    Appendix B.1 makes equality checking tractable by reducing density equality to mean and covariance comparison.
  • domain assumption The ID algorithm is sound and complete for identification.
    Used for the phi=none case in the falsifier (Shpitser and Pearl 2006).
  • standard math Identity theorem for real-analytic functions and Schwartz-Zippel lemma.
    Used to prove measure-zero agreement sets (Proposition 4.8) and to bound false acceptance in exact arithmetic (Theorem G.3).

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Verifying formulas for interventional distributions." pith.science (2026). https://pith.science/paper/PPKP32QV

@misc{pith2026260713883,
  author       = {Pith},
  title        = {Pith review of: Verifying formulas for interventional distributions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PPKP32QV}},
  note         = {Machine review of arXiv:2607.13883}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We formalize verification in causal graphical models: deciding whether a given observational formula identifies a target interventional distribution. This opens a problem complementary to identification, asking not whether any identifying formula exists, but whether the given formula is identifying. We show that even sound and complete solutions to identification do not solve verification. We propose a falsifier as a first practical route forward, prove that it induces an almost-surely correct verifier for regular exponential-family models, and use the resulting verifier to develop the gateway test, which finds all sets admissible for use in a front-door formula.

Figures

Figures reproduced from arXiv: 2607.13883 by Francesco Freni, Leonard Henckel, Sebastian Weichwald.

Figure 1
Figure 1. Figure 1: The ID algorithm returns one identifying formula for [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Acyclic directed mixed graph with unobserved confounding between [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗

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

45 extracted references · 2 linked inside Pith

  1. [1]

    Bareinboim and J

    E. Bareinboim and J. Pearl. Causal inference by surrogate experiments: z-identifiability. In Proceedings of the Twenty-Eighth Conference on Uncertainty in Artificial Intelligence, page 113–120. AUAI Press, 2012

    2012

  2. [2]

    Barndorff-Nielsen

    O. Barndorff-Nielsen. Information and Exponential Families: In Statistical Theory . John Wiley & Sons, 2014

    2014

  3. [3]

    Boeken, P

    P. Boeken, P. Forré, and J. M. Mooij. Are Bayesian networks typically faithful? arXiv preprint arXiv : 2410.16004 , 2026

    arXiv 2026

  4. [4]

    Chen and J

    L. Chen and J. M. Mooij. Complete Causal Identification from Ancestral Graphs under Selection Bias . arXiv preprint arXiv : 2603.26301 , 2026

    arXiv 2026

  5. [5]

    J. B. Conway. Functions of One Complex Variable I . Springer New York, 2nd edition, 1978

    1978

  6. [6]

    Correa and E

    J. Correa and E. Bareinboim. A Calculus for Stochastic Interventions: Causal Effect Identification and Surrogate Experiments . Proceedings of the AAAI Conference on Artificial Intelligence, 34 0 (06): 0 10093--10100, 2020

    2020

  7. [7]

    de Moura and S

    L. de Moura and S. Ullrich. The Lean 4 Theorem Prover and Programming Language . In Automated Deduction -- CADE 28, volume 12699, pages 625--635. Springer, 2021

    2021

  8. [8]

    R. A. DeMillo and R. J. Lipton. A probabilistic remark on algebraic program testing. Information Processing Letters, 7 0 (4): 0 193--195, 1978

    1978

  9. [9]

    Drton, R

    M. Drton, R. Foygel, and S. Sullivant. Global identifiability of linear structural equation models . The Annals of Statistics, 39 0 (2): 0 865 -- 886, 2011

    2011

  10. [10]

    R. J. Evans. Margins of discrete Bayesian networks . The Annals of Statistics, 46 0 (6A): 0 2623 -- 2656, 2018

    2018

  11. [11]

    F. R. Guo, E. Perković, and A. Rotnitzky. Variable elimination, graph reduction and the efficient g-formula. Biometrika, 110 0 (3): 0 739--761, 2023

    2023

  12. [12]

    Henckel, E

    L. Henckel, E. Perković, and M. H. Maathuis. Graphical Criteria for Efficient Total Effect Estimation Via Adjustment in Causal Linear Models . Journal of the Royal Statistical Society Series B: Statistical Methodology, 84 0 (2): 0 579--599, 2022

    2022

  13. [13]

    Henckel, M

    L. Henckel, M. Buttenschoen, and M. H. Maathuis. Graphical tools for selecting conditional instrumental sets. Biometrika, 111 0 (3): 0 771--788, 2023

    2023

  14. [14]

    Henckel, T

    L. Henckel, T. W\" u rtzen, and S. Weichwald. Adjustment Identification Distance: A gadjid for Causal Structure Learning . In Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence, 2024

    2024

  15. [15]

    Huang and M

    Y. Huang and M. Valtorta. Identifiability in causal Bayesian networks: a sound and complete algorithm . In Proceedings of the 21st National Conference on Artificial Intelligence - Volume 2, page 1149–1154, 2006

    2006

  16. [16]

    Ibeling, T

    D. Ibeling, T. Icard, and M. Mossé. On probabilistic and causal reasoning with summation operators. Journal of Logic and Computation, 35 0 (8): 0 exae068, 2025

    2025

  17. [17]

    Jaber, A

    A. Jaber, A. Ribeiro, J. Zhang, and E. Bareinboim. Causal Identification under Markov equivalence: Calculus, Algorithm, and Completeness . In Advances in Neural Information Processing Systems, volume 35, pages 3679--3690, 2022

    2022

  18. [18]

    M. H. Maathuis and D. Colombo. A generalized back-door criterion . The Annals of Statistics, 43 0 (3): 0 1060--1088, 2015

    2015

  19. [19]

    B. S. Mityagin. The Zero Set of a Real Analytic Function . Mathematical Notes, 107 0 (3): 0 529--530, 2020

    2020

  20. [20]

    J. Pearl. Causal Diagrams for Empirical Research . Biometrika, 82 0 (4): 0 669--688, 1995 a

    1995

  21. [21]

    J. Pearl. On the testability of causal models with latent and instrumental variables. In Proceedings of the Eleventh Conference on Uncertainty in Artificial Intelligence, page 435–443. Morgan Kaufmann Publishers Inc., 1995 b

    1995

  22. [22]

    J. Pearl. Causality: Models, Reasoning, and Inference . Cambridge University Press, 2nd edition, 2009

    2009

  23. [23]

    Perkovi\'c, J

    E. Perkovi\'c, J. Textor, M. Kalisch, and M. H. Maathuis. Complete Graphical Characterization and Construction of Adjustment Sets in Markov Equivalence Classes of Ancestral Graphs . Journal of Machine Learning Research, 18 0 (220): 0 1--62, 2018

    2018

  24. [24]

    Peters, D

    J. Peters, D. Janzing, and B. Sch\"olkopf. Elements of Causal Inference: Foundations and Learning Algorithms . The MIT Press, 2017

    2017

  25. [25]

    Richardson

    D. Richardson. Some Undecidable Problems Involving Elementary Functions of a Real Variable . The Journal of Symbolic Logic, 33 0 (4): 0 514--520, 1968

    1968

  26. [26]

    Richardson

    T. Richardson. Markov Properties for Acyclic Directed Mixed Graphs . Scandinavian Journal of Statistics, 30 0 (1): 0 145--157, 2003

    2003

  27. [27]

    Richardson and P

    T. Richardson and P. Spirtes. Ancestral graph Markov models . The Annals of Statistics, 30 0 (4): 0 962 -- 1030, 2002

    2002

  28. [28]

    T. S. Richardson, R. J. Evans, J. M. Robins, and I. Shpitser. Nested Markov properties for acyclic directed mixed graphs . The Annals of Statistics, 51 0 (1): 0 334 -- 361, 2023

    2023

  29. [29]

    J. Robins. A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect . Mathematical Modelling, 7 0 (9): 0 1393--1512, 1986

    1986

  30. [30]

    Rotnitzky and E

    A. Rotnitzky and E. Smucler. Efficient Adjustment Sets for Population Average Causal Treatment Effect Estimation in Graphical Models . Journal of Machine Learning Research, 21 0 (188): 0 1--86, 2020

    2020

  31. [31]

    M. C. Sachs, E. E. Gabriel, R. J. Evans, and A. Sjölander. Deriving Complete Constraints in Hidden Variable Models . arXiv preprint arXiv : 2601.11242 , 2026

    Pith/arXiv arXiv 2026

  32. [32]

    J. T. Schwartz. Fast Probabilistic Algorithms for Verification of Polynomial Identities . Journal of the ACM, 27 0 (4): 0 701--717, 1980

    1980

  33. [33]

    Shpilka and A

    A. Shpilka and A. Yehudayoff. Arithmetic Circuits: A survey of recent results and open questions . Foundations and Trends in Theoretical Computer Science, 5 0 (3–4): 0 207–388, 2010

    2010

  34. [34]

    Shpitser and J

    I. Shpitser and J. Pearl. Identification of joint interventional distributions in recursive semi-markovian causal models. In Proceedings of the 21st National Conference on Artificial Intelligence - Volume 2, page 1219–1226. AAAI Press, 2006

    2006

  35. [35]

    Shpitser and J

    I. Shpitser and J. Pearl. Complete Identification Methods for the Causal Hierarchy . Journal of Machine Learning Research, 9 0 (64): 0 1941--1979, 2008

    1941

  36. [36]

    Shpitser, T

    I. Shpitser, T. VanderWeele, and J. M. Robins. On the validity of covariate adjustment for estimating causal effects. In Proceedings of the Twenty-Sixth Conference on Uncertainty in Artificial Intelligence, page 527–536, 2010

    2010

  37. [37]

    Shpitser, R

    I. Shpitser, R. J. Evans, T. S. Richardson, and J. M. Robins. Introduction to Nested Markov Models . Behaviormetrika, 41 0 (1): 0 3--39, 2014

    2014

  38. [38]

    Shpitser, R

    I. Shpitser, R. Evans, and T. Richardson. Acyclic linear SEMs obey the Nested Markov property . In Proceedings of the Thirty-Fourth Conference on Uncertainty in Artificial Intelligence, pages 735--745, 2018

    2018

  39. [39]

    Spirtes, C

    P. Spirtes, C. Glymour, and R. Scheines. Causation, Prediction, and Search . The MIT Press, 2000

    2000

  40. [40]

    Tian and J

    J. Tian and J. Pearl. A general identification condition for causal effects. In Eighteenth National Conference on Artificial Intelligence, page 567–573. American Association for Artificial Intelligence, 2002

    2002

  41. [41]

    Verma and J

    T. Verma and J. Pearl. Equivalence and synthesis of causal models. In Proceedings of the Sixth Annual Conference on Uncertainty in Artificial Intelligence, page 255–270, 1990

    1990

  42. [42]

    Wien\" o bst, B

    M. Wien\" o bst, B. van der Zander, and M. Li\' s kiewicz. Linear-time algorithms for front-door adjustment in causal graphs. In Proceedings of the Thirty-Eighth AAAI Conference on Artificial Intelligence and Thirty-Sixth Conference on Innovative Applications of Artificial Intelligence and Fourteenth Symposium on Educational Advances in Artificial Intelli...

    2024

  43. [43]

    Witte, L

    J. Witte, L. Henckel, M. H. Maathuis, and V. Didelez. On Efficient Adjustment in Causal Graphs . Journal of Machine Learning Research, 21 0 (246): 0 1--45, 2020

    2020

  44. [44]

    Yvernes, E

    C. Yvernes, E. Devijver, M. Clausel, and E. Gaussier. Unveiling the Structure of Do-Calculus Reasoning via Derivation Graphs . arXiv preprint arXiv : 2606.03719 , 2026

    Pith/arXiv arXiv 2026

  45. [45]

    R. Zippel. Probabilistic Algorithms for Sparse Polynomials . In Proceedings of the International Symposiumon on Symbolic and Algebraic Computation, pages 216--226. Springer, 1979

    1979

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.