Pith. sign in

REVIEW 3 major objections 4 minor 36 references

Causal identification with $Y_0$

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Y0 is a Python package that claims to implement the most complete suite of causal identification algorithms of any causal inference package, spanning interventional, counterfactual, and transportability queries.

desk verdict A real, useful open-source package for causal identification; the missing piece is validation of the implementations, and that is fixable in revision. read the letter →

arxiv 2508.03167 v1 pith:CDOCBVQI submitted 2025-08-05 cs.AI

Charles Tapley Hoyt , Craig Bakker , Richard J. Callahan , Joseph Cottam , August George , Benjamin M. Gyori , Haley M. Hummel , Nathaniel Merrill
show 6 more authors
Sara Mohammad Taheri Pruthvi Prakash Navada Marc-Antoine Parent Adam Rupe Olga Vitek Jeremy Zucker
This is my paper
classification cs.AI
keywords causalidentificationinferencecounterfactualqueriestransportabilityacyclicdirectedmixedgraphssymbolicestimandssoftwarepackagehierarchy
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Y0 is an open-source Python package for the qualitative part of causal inference: before estimating any effect size, it takes a causal graphical model, a query, and information about which variables were observed or randomized, and decides whether the query is identifiable at all. The central claim of the paper is that Y0 implements the most complete suite of causal identification algorithms of any causal inference package, covering interventional queries (ID, IDC), counterfactual queries (ID*, IDC*), surrogate outcomes, tian-ID, transport, and counterfactual transport. If the claim is true, a researcher can express a causal question in one domain-specific language, run the appropriate algorithm, and receive a symbolic estimand that can be estimated non-parametrically from the available data. The paper illustrates the workflow on the smoking-tar-cancer example, including a confounded version that the basic ID algorithm cannot handle, and on COVID-19 transportability.

What carries the argument

The object that carries the argument is the acyclic directed mixed graph (ADMG), whose directed edges represent direct causal influence and whose bidirected edges represent unobserved confounding, together with a domain-specific language for symbolic probabilistic expressions such as $P_{\mathrm{do}(X=x)}(Y=y)$ and counterfactual conditional contrasts. The ADMG is the common input format that all identification algorithms operate on; the domain-specific language is what lets a user's query become an expression that the algorithms rewrite, through graph surgeries and summations over latent variables, into an estimand over observed variables. Around those two objects sit graph-conversion utilities and falsification workflows that check whether a proposed graph is consistent with the data.

What would settle it

Take an ADMG and a query for which the published literature states no estimand exists; if Y0 returns a symbolic expression instead of declaring non-identifiability, the implementation is wrong. A second falsifier is to run a large corpus of random graphs through Y0 and through independent reference implementations and compare the returned estimands symbolically.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes Y0 as a unified implementation of eight identification algorithms: ID, IDC, ID*, IDC*, TRSO, tian-ID, transport, and counterfactual transport. These span Pearl's causal hierarchy from association through intervention to counterfactuals, and they accept data from randomized controlled trials, observational studies, or mixtures of both. The package represents causal models with unobserved confounders as acyclic directed mixed graphs, encodes queries and estimands as symbolic probabilistic expressions, and, when identification succeeds, returns a non-parametric estimand in terms of observed quantities. The stated claim is that this collection is the most complete suite of identification algorithms available in any causal inference package.

Load-bearing premise

The load-bearing premise is that every implemented identification algorithm is coded correctly; the paper lists what is covered but does not report unit tests or side-by-side validation, so one faulty algorithm would undermine the package's practical value even if its breadth claim is accurate.

Editorial extensions

If this is right

  • A single Python package can replace the current patchwork of R and Python tools for identification queries, including the interventional, conditional, and counterfactual cases.
  • Researchers can determine before any estimation step whether an effect is identifiable from observational data, a randomized trial, or a combined data source.
  • Queries that transport findings from one population to another become directly computable, yielding a symbolic recipe for combining evidence from multiple studies.
  • Graph falsification and identification live in the same library, so the same graph object can be checked against data before a causal claim is derived.
  • Because the expression and graph layers are modular, new identification algorithms can be added without redesigning the query language.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A consequence the authors leave implicit is that failed identification is itself a useful experimental-design result: the contrast between a query that ID rejects and one that TRSO or transport accepts points to exactly the data source or population that should be collected next.
  • The symbolic outputs make Y0 a natural front end for automated estimator generation, which the paper identifies as an open problem for anything more sophisticated than ID.
  • The same graph layer could host algorithms that are not yet implemented, such as cyclic or missing-data identification, without forcing users to change how they express queries.
  • Interoperability with existing graph formats could let teams keep their current modeling front end and use Y0 purely for the identification step, lowering the cost of adoption.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper presents Y0, an open-source Python package for symbolic causal identification. Y0 implements a domain-specific language for probabilistic expressions, graph structures for causal models with unobserved confounders (ADMGs and Evans-type latent graphs), falsification workflows, and a suite of identification algorithms: ID, IDC, ID*, IDC*, surrogate outcomes (TRSO), tian-ID, transport, and counterfactual transport. The authors claim this is the most complete identification suite of any causal inference package, and they illustrate the package with a smoking-tar-cancer case study showing estimands from ID and TRSO, plus a second, notebook-referenced transportability case study on COVID-19. The paper introduces no new mathematical theory; its central claim is that Y0 is a useful, complete, and correct implementation of published identification algorithms.

Significance. If the implementations are correct, Y0 would be a valuable community resource: it is open source, provides a symbolic DSL with SymPy and LaTeX rendering, includes graph conversion utilities, and covers nested counterfactual and transportability algorithms that are rarely available in a single package. The public repository, documentation, and case-study notebooks are concrete strengths. However, the paper offers no machine-checked proofs, no reported test suite, and no side-by-side validation against existing implementations such as causaleffect or cfid. Because the algorithms are complex and exact symbolic outputs are checkable, the correctness gap is fixable within the paper's scope, but as written the central practical claim rests on unverified assertions about the implementations.

major comments (3)
  1. [Identification section and Case Study section] The paper claims implementations of ID, IDC, ID*, IDC*, TRSO, tian-ID, transport, and counterfactual transport, but it reports no unit tests, randomized trials, or comparisons against reference implementations. Since these algorithms are nontrivial and their outputs are exact symbolic estimands, a single implementation bug in ID*, IDC*, or counterfactual transport could silently produce invalid results. The paper should add a validation section with at least (i) a battery of test graphs covering multiple outcomes, conditional queries, selection bias, and complex latent structures, and (ii) side-by-side comparison with causaleffect, cfid, or Ananke on shared examples. Without this, the central claim that Y0 returns correct estimands for accepted inputs is not supported.
  2. [Identification section] The assertion that 'Y0 has the most complete suite of identification algorithms of any causal inference package' is unsupported as written. The Statement of Need itself says that causaleffect implements ID, IDC, TRSO, and transport, and that cfid implements ID* and IDC*, so a reader cannot evaluate whether the claim is per-package, per-language, or identification-only. A feature matrix listing each algorithm and each existing package's coverage, together with an explicit scope statement, is needed to make the claim meaningful and testable.
  3. [Case Study section] The worked examples exercise only ID and TRSO on a simple three-variable graph. No output is shown for IDC, ID*, IDC*, tian-ID, or counterfactual transport, and the second case study is referenced only via a notebook rather than summarized in the paper. At minimum, a worked symbolic output for each algorithm family would demonstrate that the implemented code executes on accepted inputs and yields expected estimands; the present paper does not provide such evidence.
minor comments (4)
  1. [Abstract and Availability and Usage section] The abstract states that the source code is under the MIT License, while the Availability and Usage section says it is under a BSD 3-clause license; please make the license statement consistent.
  2. [Statement of Need section] The sentence about CausalFusion, 'neither open source, available for registration of new users, nor provides documentation', is grammatically incomplete; it should read 'neither open source nor available for registration of new users, nor does it provide documentation'.
  3. [Case Study section] The estimand for Figure 1B uses the notation P^{\pi_1}_{Smoking}(Tar); the superscript and subscript conventions for population and intervention variables are not defined in the text. Please define them near the equation or point explicitly to the notebook for notation.
  4. [Case Study section] The COVID-19 transport and counterfactual transport case study is only referenced via a Jupyter notebook; for a self-contained paper, include the graph and the resulting estimand in the text, or at least state the key structural assumptions used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper implements and cites published identification algorithms and does not derive new results from fitted or self-referential inputs.

full rationale

The paper is a software description, not a derivation of new causal identification results. Its central content is the claim that Y0 implements published algorithms (ID, IDC, ID*, IDC*, TRSO, tian-ID, transport, counterfactual transport), each attributed to external references such as Shpitser & Pearl, Tikka & Karvanen, S. Lee et al., and Correa et al. The case studies are illustrative applications of existing algorithms (ID and TRSO) to simple graphs; the resulting estimands are straightforward symbolic outputs of those published methods, not quantities fitted from data or defined in terms of the package's own outputs. The 'used in several scientific studies' bullet list cites prior applications of Y0, but these applications are reported as downstream uses and do not feed back into any argument establishing the package's correctness or completeness. The claim that Y0 has 'the most complete suite of identification algorithms of any causal inference package' is an unsupported comparative assertion, and the absence of unit tests or side-by-side comparisons is a correctness and evidence concern, but neither constitutes circular reasoning under the stated criteria. No self-definitional reduction, fitted-input-as-prediction, load-bearing self-citation, or imported uniqueness theorem appears in the paper.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper contributes software, not a new mathematical theory. The central claim rests on standard causal inference formalism and on the correctness of the cited algorithms, neither of which introduces free parameters or invented entities.

assumptions (3)
  • domain assumption Causal models with unobserved confounders can be represented as acyclic directed mixed graphs (ADMGs).
    Y0's graph data structure and all identification algorithms rely on this representation, which is standard in the cited literature (Data Structure section).
  • standard math The identification algorithms cited from the literature are correct as originally published.
    The central claim depends on the correctness of ID, IDC, ID*, IDC*, TRSO, tian-ID, transport, and counterfactual transport; Y0 is an implementation of these published results, not a re-derivation (Identification section).
  • standard math Pearl's causal hierarchy and its three levels (association, intervention, counterfactual) are the appropriate formalization of causal questions.
    The package's DSL and query semantics are built on this framework (Bareinboim et al., 2022; Probabilistic Expressions section).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Causal identification with $Y_0$." pith.science (2026). https://pith.science/paper/CDOCBVQI

@misc{pith2026250803167,
  author       = {Pith},
  title        = {Pith review of: Causal identification with $Y_0$},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CDOCBVQI}},
  note         = {Machine review of arXiv:2508.03167}
}
abstract

We present the $Y_0$ Python package, which implements causal identification algorithms that apply interventional, counterfactual, and transportability queries to data from (randomized) controlled trials, observational studies, or mixtures thereof. $Y_0$ focuses on the qualitative investigation of causation, helping researchers determine whether a causal relationship can be estimated from available data before attempting to estimate how strong that relationship is. Furthermore, $Y_0$ provides guidance on how to transform the causal query into a symbolic estimand that can be non-parametrically estimated from the available data. $Y_0$ provides a domain-specific language for representing causal queries and estimands as symbolic probabilistic expressions, tools for representing causal graphical models with unobserved confounders, such as acyclic directed mixed graphs (ADMGs), and implementations of numerous identification algorithms from the recent causal inference literature. The $Y_0$ source code can be found under the MIT License at https://github.com/y0-causal-inference/y0 and it can be installed with pip install y0.

Figures

Figures reproduced from arXiv: 2508.03167 by the authors.

Figure 1
Figure 1. A) A simplified acyclic directed graph model representing prior knowledge on smoking and cancer and B) a more complex acyclic directed mixed graph that explicitly represents confounding variables. The identification algorithm (ID) (Shpitser & Pearl, 2006) estimates the effect of smoking on the risk of cancer in Figure 1A as P T ar P(Cancer|Smoking, T ar)P(T ar|Smoking). However, the model in Figure 1A is inaccurate … view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

36 extracted references · 26 canonical work pages

  1. [1]

    Agrawal, R., Witty, S., Zane, A., & Bingham, E. (2024). Automated efficient estimation using monte carlo efficient influence functions. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, & C. Zhang (Eds.), Advances in neural information processing systems(Vol. 37, pp. 16102– 16132). Curran Associates, Inc. https://proceedings.neurips....

    work page 2024

  2. [2]

    (2015).pgmpy: Probabilistic Graphical Models using Python

    Ankan, A., & Panda, A. (2015).pgmpy: Probabilistic Graphical Models using Python. 6–11. https://doi.org/10.25080/majora-7b98e3ed-001

    doi:10.25080/majora-7b98e3ed-001 2015

  3. [3]

    D., Ibeling, D., & Icard, T

    Bareinboim, E., Correa, J. D., Ibeling, D., & Icard, T. (2022). On pearl’s hierarchy and the foundations of causal inference. InProbabilistic and causal inference: The works of judea pearl(1st ed., pp. 507–556). Association for Computing Machinery. https://doi.org/10.1145/3501714.3501743

    arXiv 2022

  4. [4]

    P., Jankowiak, M., Obermeyer, F., Pradhan, N., Karalet- sos, T., Singh, R., Szerlip, P., Horsfall, P., & Goodman, N

    Bingham, E., Chen, J. P., Jankowiak, M., Obermeyer, F., Pradhan, N., Karalet- sos, T., Singh, R., Szerlip, P., Horsfall, P., & Goodman, N. D. (2018).Pyro: Deep universal probabilistic programming. https://arxiv.org/abs/1810.09538

    arXiv 2018

  5. [5]

    Bingham, E., & Witty, S. (2025). Causal reasoning with ChiRho. InGitHub repository. https://github.com/BasisResearch/chirho; GitHub

    work page 2025

  6. [6]

    D., Lee, S., & Bareinboim, E

    Correa, J. D., Lee, S., & Bareinboim, E. (2022). Counterfactual transportability: A formal approach. In K. Chaudhuri, S. Jegelka, L. Song, C. Szepesvari, G. Niu, & S. Sabato (Eds.),Proceedings of the 39th international conference on machine learning(Vol. 162, pp. 4370–4390). PMLR. https://proceedings.ml r.press/v162/correa22a.html

    work page 2022

  7. [7]

    D., Lee, S., & Bareinboim, E

    Correa, J. D., Lee, S., & Bareinboim, E. (2021). Nested counterfactual identifica- tion from arbitrary surrogate experiments.Advances in Neural Information Processing Systems,34. https://arxiv.org/abs/2107.03190

    arXiv 2021

  8. [8]

    Causal Inference Using Tractable Circuits

    Darwiche, A. (2022).Causal inference using tractable circuits. https://arxiv.or g/abs/2202.02891

    work page Pith review arXiv 2022

Show all 36 references
  1. [9]

    Drton, M., & Richardson, T. S. (2004). Iterative conditional fitting for gaussian ancestral graph models.Proceedings of the 20th Conference on Uncertainty in Artificial Intelligence, 130–137. ISBN: 0974903906

    2004

  2. [10]

    A., Blöbaum, P., Hardt, M., & Janzing, D

    Eulig, E., Mastakouri, A. A., Blöbaum, P., Hardt, M., & Janzing, D. (2023). 6 Toward falsifying causal graphs using a permutation-based test. https://arxiv. org/abs/2305.09565

    2023 arXiv

  3. [11]

    Evans, R. J. (2016). Graphs for margins of bayesian networks.Scandinavian Journal of Statistics,43(3), 625–648. https://doi.org/10.1111/sjos.12194 Forré, P., & Mooij, J. M. (2018).Constraint-based causal discovery for non- linear structural causal models with cycles and latent...

    2016 arXiv

  4. [12]

    A., Schult, D

    Hagberg, A. A., Schult, D. A., & Swart, P. J. (2008). Exploring network structure, dynamics, and function using NetworkX. In G. Varoquaux, T. Vaught, & J. Millman (Eds.),Proceedings of the 7th python in science conference(pp. 11–15). https://aric.hagberg.org/papers/hagberg-200...

    2008

  5. [13]

    Lee, J. J. R., Bhattacharya, R., Nabi, R., & Shpitser, I. (2023).Ananke: A python package for causal inference using graphical models. https://arxiv.or g/abs/2301.11477

    2023 arXiv

  6. [14]

    D., & Bareinboim, E

    Lee, S., Correa, J. D., & Bareinboim, E. (2019). General identifiability with arbitrary surrogate experiments.Proceedings of the 35th Conference on Uncertainty in Artificial Intelligence. https://proceedings.mlr.press/v115/lee 20b.html

    2019

  7. [15]

    D., & Bareinboim, E

    Lee, S., Correa, J. D., & Bareinboim, E. (2020). General transportability – synthesizing observations and experiments from heterogeneous domains. Proceedings of the AAAI Conference on Artificial Intelligence,34(06), 10210– 10217. https://doi.org/10.1609/aaai.v34i06.6582

    2020 doi

  8. [16]

    E., Muller, R

    Granger, B. E., Muller, R. P., Bonazzi, F., Gupta, H., Vats, S., Johansson, F., Pedregosa, F., ... Scopatz, A. (2017). SymPy: Symbolic computing in python.PeerJ Computer Science,3, e103. https://doi.org/10.7717/peerj- cs.103

    2017 doi

  9. [17]

    Gyori, B., & Vitek, O. (2024). Eliater: A python package for estimating outcomes of perturbations in biomolecular networks.Bioinformatics. https: //doi.org/10.1093/bioinformatics/btae527

    2024 doi

  10. [18]

    T., Zucker, J., & Vitek, O

    Mohammad-Taheri, S., Tewari, V., Kapre, R., Rahiminasab, E., Sachs, K., Hoyt, C. T., Zucker, J., & Vitek, O. (2022).Experimental design for causal query estimation in partially observed biomolecular networks. https://arxiv.org/ab s/2210.13423

    2022 arXiv

  11. [19]

    T., Zucker, J., & Vitek, O

    Mohammad-Taheri, S., Tewari, V., Kapre, R., Rahiminasab, E., Sachs, K., Hoyt, C. T., Zucker, J., & Vitek, O. (2023). Optimal adjustment sets for causal query estimation in partially observed biomolecular networks.Bioinformatics, 39(Supplement_1), i494–i503. https://doi.org/10....

    2023 doi

  12. [20]

    Mohan, K., & Pearl, J. (2021). Graphical models for processing missing data. Journal of the American Statistical Association,116(534), 1023–1037. https: //doi.org/10.1080/01621459.2021.1874961 7

    2021

  13. [21]

    (2024).Causal AI

    Ness, R. (2024).Causal AI. O’Reilly Media. ISBN: 978-1-63343-991-7

    2024

  14. [22]

    (2009).Causality(2nd ed.)

    Pearl, J. (2009).Causality(2nd ed.). Cambridge University Press. ISBN: 978- 0521895606

    2009

  15. [23]

    Pearl, J., Geiger, D., & Verma, T. S. (1989). Conditional independence and its representations.Kybernetika,25(Suppl), 33–44. http://eudml.org/doc/28568

    1989

  16. [24]

    (2021).Algorithmic causal effect identification with causaleffect

    Pedemonte, M., Vitrià, J., & Parafita, Á. (2021).Algorithmic causal effect identification with causaleffect. https://arxiv.org/abs/2107.04632

    2021 arXiv

  17. [25]

    (2020).DoWhy: An end-to-end library for causal inference

    Sharma, A., & Kiciman, E. (2020).DoWhy: An end-to-end library for causal inference. https://arxiv.org/abs/2011.04216

    2020 arXiv

  18. [26]

    Shpitser, I., & Pearl, J. (2006). Identification of joint interventional distributions in recursive semi-markovian causal models.Proceedings of the 21st National Conference on Artificial Intelligence - Volume 2, 1219–1226. https://doi.org/ 10.5555/1597348.1597382

    2006

  19. [27]

    Shpitser, I., & Pearl, J. (2008). Complete identification methods for the causal hierarchy.Journal of Machine Learning Research,9(64), 1941–1979. http: //jmlr.org/papers/v9/shpitser08a.html

    2008

  20. [28]

    (2012).What counterfactuals can be tested

    Shpitser, I., & Pearl, J. (2012).What counterfactuals can be tested. https: //arxiv.org/abs/1206.5294

    2012 arXiv

  21. [29]

    (2012).On the testable implications of causal models with hidden variables

    Tian, J., & Pearl, J. (2012).On the testable implications of causal models with hidden variables. https://arxiv.org/abs/1301.0608

    2012 arXiv

  22. [30]

    Tian, J., & Shpitser, I. (2010). On identifying causal effects.Heuristics, Proba- bility and Causality: A Tribute to Judea Pearl (R. Dechter, H. Geffner and J. Halpern, Eds.). College Publications, UK, 415–444. https://faculty.sites. iastate.edu/jtian/files/inline-files/tian-s...

    2010

  23. [31]

    Tikka, S. (2023). Identifying counterfactual queries with the r package cfid.The R Journal,15, 330–343. https://doi.org/10.32614/RJ-2023-053

    2023 doi

  24. [32]

    Tikka, S., Hyttinen, A., & Karvanen, J. (2021). Causal effect identification from multiple incomplete data sources: A general search-based approach.Journal of Statistical Software,99(5), 1–40. https://doi.org/10.18637/jss.v099.i05

    2021 doi

  25. [33]

    Tikka, S., & Karvanen, J. (2017a). Identifying causal effects with the r package causaleffect.Journal of Statistical Software,76(12), 1–30. https://doi.org/ 10.18637/jss.v076.i12

    2017 doi

  26. [34]

    Tikka, S., & Karvanen, J. (2019). Surrogate outcomes and transportability. International Journal of Approximate Reasoning,108, 21–37. https://doi.or g/10.1016/j.ijar.2019.02.007

    2019 doi

  27. [35]

    (2023).Tractable probabilistic models for causal learning and reasoning [Doctoral Dissertation]

    Wang, B. (2023).Tractable probabilistic models for causal learning and reasoning [Doctoral Dissertation]. https://ora.ox.ac.uk/objects/uuid:2fafc463-3a9f- 40cb-a48c-e33272c691b8

    2023

  28. [36]

    N., & Blei, D

    Weinstein, E. N., & Blei, D. M. (2024).Hierarchical causal models. https: //arxiv.org/abs/2401.05330 8

    2024 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.