Analysis Preservation and Reproducibility in Collider Physics: An ADL/CutLang Implementation of ATLAS Open Data Analyses

arXiv:2607.19034 · hep-ex · Submitted 2026-07-21 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.

Kai: We're looking at the paper: "Analysis Preservation and Reproducibility in Collider Physics".

Title and authors: Kai: We're looking at this paper, "Analysis Preservation and Reproducibility in Collider Physics: An ADL/CutLang Implementation of ATLAS Open Data Analyses." The core idea is using Analysis Description Language, or ADL, and the CutLang runtime interpreter to describe and execute analyses.

Mira: Basically, they are giving physicists a common language where all the physics objects and event selection criteria are written down in a way that is human-readable. This language lets you define what you want to see in the data for any experiment.

Lev: The authors argue that this decouples the actual physics logic from the specific software infrastructure used by experiments, which is crucial for reproducibility across different analyses and setups.

Kai: They introduce CutLang as a runtime interpreter that executes these ADL descriptions directly on the data without needing to compile them into a general-purpose program first. This makes it flexible and portable across different input data formats.

Mira: It’s also presented as a pedagogical tool, which is interesting because they mention this approach has been used in workshops and schools before. They want to show how students can get a clear understanding of the physics by writing these standardized descriptions.

Lev: That direct execution at runtime means we have to consider how that overhead lands when we try to run it on real hardware, especially with multi-core architectures where they showed it can execute analyses in parallel.

Kai: The study re-implemented nine benchmark analyses from the ATLAS Open Data portal and ran them using this ADL/CutLang setup on thirteen TeV data with ten femtobarns of integrated luminosity <ref:2607.19034#pg2>.

Mira: And the results they got matched the original publications quite well across a wide range of Standard Model processes and even some Beyond the Standard Model searches.

Lev: That consistency is important, but I wonder about the limitations they mention regarding how these descriptions are interpreted at runtime compared to a fully compiled C++ framework.

Kai: They did address that by showing performance comparisons, and for instance, the Data over Monte Carlo ratio showed fully consistent results between the two frameworks for most analyses.

Mira: That consistency across different final state topologies is what really makes this approach compelling for validating existing results in high energy physics.

Lev: It’s about seeing if we can get reliable results using a system that doesn't require us to write new keywords or functions for every single analysis we want to do.

The paper's summary: Kai: So, summarizing this paper, the authors are presenting ADL and CutLang as an innovative and sustainable solution for describing, executing, and preserving physics content from collider data analyses.

Mira: They show that by using ADL to express everything in a human-readable form for physics objects and event selections, you get event selections that can be fed right into statistical analysis tools.

Lev: The key benefit they emphasize is decoupling the physics logic itself from the experiment-specific software infrastructures, which is a big step toward better reproducibility and transparency.

Kai: They demonstrate this by taking nine benchmark analyses—covering SM processes and BSM searches—and re-implementing them in ADL, then executing those descriptions with CutLang.

Mira: They use data from the ATLAS Open Data at thirteen TeV with an integrated luminosity of ten femtobarns to test this workflow thoroughly <ref:2607.19034#pg2,an integrated luminosity of 10>.

Lev: The benchmark analyses they chose covered a broad range of final state topologies, which means they tested the system on many different kinds of events, not just one specific type.

Kai: They also show that CutLang can handle complex kinematic reconstructions through an optimization mechanism, for example in the Higgs boson production analysis where it identified more Z boson candidates than a more traditional Open Data implementation.

Mira: The overall summary is that ADL/CutLang provides a common language for describing analyses, which facilitates reproducibility and transparency, and it’s also useful as an educational tool.

Lev: But we have to remember the context of what they are comparing it against—they are showing consistency with existing frameworks like the thirteen TeV ATLAS Open Data analysis framework written in C++ integrated with ROOT <ref:2607.19034#pg2,the 13 TeV ATLAS Open Data analysis framework written in C>.

Kai: It’s a practical demonstration that you don't need to build a whole new software stack just to do an analysis; you just write the description in ADL.

Mira: So, they are showing that this approach is not just theoretical but has been rigorously tested against established results in SM and BSM searches.

The paper's improvements: Lev: One of the main improvements they point out is the decoupling of the analysis algorithm from its software implementation, which directly addresses the challenge of analysis preservation.

Kai: They also highlight how CutLang allows for direct runtime execution without prior compilation, which makes the workflow very flexible and portable across different input data formats.

Mira: They’ve improved handling complex physics selections by allowing complex angular combinations between multiple objects, ensuring consistency in event selection criteria even when dealing with things like specific angular cuts on small R jets and large R jets.

Lev: On the performance side, they show that CutLang can execute analyses in parallel using a multi-core architecture, which is important because it means we can scale up to handle larger amounts of collected data and simulation events efficiently.

Kai: For the H to ZZ star analysis specifically, they showed that their optimization mechanism could select more Z boson candidates than the Open Data framework’s implementation when using multiple independent if conditions.

Mira: They also pointed out that the overall Data over Monte Carlo ratio from most analyses showed fully consistent results between the ADL/CutLang and the Open Data frameworks.

Lev: The authors also mentioned that they re-implemented benchmark analyses from the ATLAS Open Data C++ Framework, which is a specific framework, not just any generic tool.

Kai: And they show that for many of these analyses, like Z boson production or W boson production, the data over Monte Carlo ratio was very close to one across both frameworks.

Conclusion: Mira: So wrapping up this paper on "Analysis Preservation and Reproducibility in Collider Physics: An ADL/CutLang Implementation of ATLAS Open Data Analyses," the main implication is that ADL and CutLang offer a practical way to describe, execute, and preserve physics content.

Kai: It means we can achieve better reproducibility by using a common language for describing analyses, which is crucial for result validation in high energy physics.

Lev: From an engineering side, it’s about having this system that allows us to run analysis descriptions directly at runtime without the overhead of a full compilation step, which helps with flexibility and portability across different data formats.

Mira: The consistency they showed across various Standard Model and Beyond the Standard Model analyses confirms that this same event selection algorithm can be successfully reproduced using ADL/CutLang.

Kai: It’s a practical system for performing open data analyses, especially in an educational context, because it lets students develop a clear understanding of the physics by writing these standardized descriptions.

Lev: For me, the real value is seeing how this works on real hardware; if it can scale and run in parallel effectively, that’s where we see its impact on actual computational workflows.

Mira: I think this work provides a reliable alternative to traditional analysis frameworks for SM and BSM searches because the results match existing literature so well.

Kai: So, we have ADL and CutLang as a tool to keep our physics logic separate from the experiment-specific software, making it easier to maintain and reuse.

Istanbul University · Boğaziżi University

hep-ex

Submitted: 2026-07-21

Updated: 2026-07-21

Code: https://github.com/unelg/cutlang

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 88/100

The gist: Analysis Description Language (ADL) and CutLang runtime interpreter offer an innovative and sustainable solution for describing, executing…

Key concepts

ADL (Analysis Description Language)
A domain-specific language designed to describe physics analyses in collider experiments, such as ATLAS. It allows physicists to express objects and event selection criteria in a structured, human-readable format. This language acts as the blueprint for an analysis.
CutLang
The runtime interpreter that executes ADL descriptions directly without needing prior compilation into a general programming language. CutLang processes the ADL file at runtime to produce event selections, making the workflow flexible and portable across various data inputs.
Analysis Preservation
The goal of decoupling physics logic from experiment-specific software infrastructure. By using ADL/CutLang, the core analysis algorithm is independent of the specific software used for data handling. This makes it easier to reproduce results later, even if the underlying software changes.
Data/MC Ratio Consistency
A performance metric comparing event yields obtained from real data versus simulated Monte Carlo events. The study showed that ADL/CutLang produced Data/MC ratios fully consistent with established Open Data frameworks for various analyses, validating its reliability.

Terminology

Summary

Analysis Description Language (ADL) and CutLang runtime interpreter offer an innovative and sustainable solution for describing, executing, and preserving the physics content of collider data analyses In this solution, all physics objects and event selection criteria are expressed in a human-readable form using ADL, a domain-specific language designed for collider physics ADL descriptions are interpreted directly at runtime by CutLang without translation into a general-purpose programming language, producing event selections that can be readily fed into statistical analysis tools This approach addresses the long-standing challenge of analysis preservation and analysis reproducibility in high energy physics by decoupling the physics logic from experiment-specific software infrastructures

ADL/CutLang Workflow and Implementation

The ADL/CutLang workflow facilitates reproducibility and transparency by providing a common language for describing analyses CutLang, the runtime interpreter for ADL, executes analysis descriptions directly without requiring prior compilation The system allows a single ADL file to be executed over different input data formats without any additional coding, making the workflow both flexible and portable Key features of ADL include structured blocks for object definitions, event selections, information, and parameter definitions For example, object definitions are expressed using syntax like object goodEles take ELE select isTightID(ELE) == 1

Validation through Benchmark Analyses

The study re-implemented several benchmark analyses from the ATLAS Open Data C++ Framework in ADL and executed them with CutLang using data from the ATLAS Open Data at a center-of-mass energy of 13 TeV and an integrated luminosity of 10 fb−1 The results yielded in good agreement with the original publications, validating the ADL/CutLang workflow as a reliable alternative to traditional analysis frameworks Nine analyses were selected to cover a broad range of standard model (SM) processes as well as Beyond the standard model (BSM) searches

Performance and Consistency Comparison

The performance of CutLang was evaluated using multi-core architecture, showing that it can execute analyses in parallel The comparison between ADL/CutLang and the Open Data framework demonstrated consistent results across various final state topologies For instance, the Data/MC ratio from most analyses showed fully consistent results between the two frameworks

Handling Complex Physics Selections

The ADL/CutLang framework demonstrates flexibility by allowing complex angular combinations between multiple objects In the search for direct production of pairs of sleptons, specific angular criteria on both small-R jets and large-R jets are explicitly defined within the object blocks to ensure consistent event selection The optimization mechanism in CutLang is shown to be able to select more Z boson candidates than the Open Data framework’s implementation with multiple independent if conditions, which is demonstrated in the H → ZZ∗ analysis

Conclusion on ADL/CutLang Utility

The study concludes that ADL/CutLang is a practical system for performing open data analyses, particularly in an educational context The consistent results across different SM and BSM analyses confirm that the same event selection algorithm can be successfully reproduced using ADL/CutLang The paper highlights the advantages of decoupling the analysis algorithm from its software implementation, ensuring reliable results The flexibility and expressiveness of ADL and CutLang are demonstrated by successfully implementing various analyses without needing new keywords or functions The overall Data/MC ratio from most analyses shows fully consistent results between the two frameworks

Acknowledgement

This work is being supported by the CHIST-ERA OpenMAPP project under grant 223N165 from The Scientific and Technological Research Council of T¨urkiye (TUBITAK) The authors acknowledge the efforts of the ATLAS Collaboration to record or simulate, reconstruct, and distribute the Open Data used in this paper, and to develop and support the software with which it was analyzed Finally, the authors would like to thank Sezen Sekmen, V. Erkcan Ozcan and G¨okhan ¨ Unel for their valuable contributions The references listed provide the necessary context for the methods and tools used in this research The list of ATLAS Open Data samples is provided in Table 27. p. 2 The list of predefined functions in CutLang used in this study is provided in Table 28. p. 2

References

[1] G. Unel and S. Sekmen, “Cutlang: A particle physics analysis description language and runtime interpreter,” Computer Physics Communications, vol. p. 0 233, p. 215–236, Dec. 2018

[2] G. Unel, S. Sekmen, A. M. Toon, B. Gokturk, B. Orgen, A. Paul, N. Ravel, and J. Setpal, “Cutlang v2: Advances in a runtime-interpreted analysis description language for hep data,” Frontiers in Big Data, vol. p. 0 4, June 2021

[3] A. Adiguzel, O. Cakir, U. Kaya, V. E. Ozcan, S. Ozturk, S. Sekmen, I Turk Cakir, and G Unel, “Cutlang as an analysis description language for introducing students to analyses in particle physics,” European Journal of Physics, vol. p. 0 42, p. 035802

[4] “ATLAS Outreach C++ Framework: 13 TeV.” https://github. p. 1 p. 1com/atlas-outreach-data-tools/atlas-outreach-cpp-framework-13tev

[5] ATLAS Collaboration, “Review of the 13 TeV ATLAS Open Data release,” tech. rep., CERN, Geneva, 2020

[6] “CMS Open Data Guide.” https://cms-opendata-guide. p. 6web.cern.ch

[7] H. B. Prosper, S. Sekmen, and G Unel, “Analysis Description Language (ADL) Web Portal. p. 2”

[8] R. Brun and F Rademakers, “Root — an object oriented data analysis framework,” Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment, vol. p. 1 389, no. 1, pp. 81–86, 1997

[9] “CutLang GitHub repository.” https://github.com/unelg/cutlang

[10] K. Karaca and K Sahan and A Sansar, “ADL Implementations for 13 TeV ATLAS Open Data,” https://github. p. 0com/ADL4HEP/ADL4LHCOpenData/tree/main/ATLAS-OTRC-2019-01

[11] “ATLAS Open Data and Tools for Education.” https://opendata. p. 6atlas.cern

[12] M Aaboud et al., “Probing the W tb vertex structure in t-channel single-top-quark production and decay in pp collisions at √s = 8 TeV with the ATLAS detector,” JHEP, vol. p. 2 04, p. 124, 2017

[13] M Aaboud et al., “Measurement of the inclusive cross-sections of single top-quark and top-antiquark t-channel production in pp collisions at √s = 13 TeV with the ATLAS detector,” JHEP, vol. p. 2 04, p. 086, 2017

[14] ATLAS Collaboration, “Optimisation of the ATLAS b-tagging performance for the 2016 LHC Run,” tech. rep., CERN, Geneva, 2016

Table 27: A list of Standard Model (SM) simulation samples in the ATLAS 13 TeV Open Data used in this study

Sample Dataset ID

W + jets 361100–361105

Z + jets 361106–361108

Diboson 363356, 363489–3649

Table 28: Predefined functions in CutLang and their ADL syntax

Syntax Meaning

Object attributes m Mass of q Charge of Phi Phi of Eta Eta of AbsEta Absolute value of Eta of Rap Rapidity of Pt pT of Pz pZ(E) Energy of P(P) Momentum

Table 26: Comparison of the Data/MC ratio obtained using the ADL/CutLang and Open Data frameworks

Data/Monte Carlo Analysis ADL/CutLang Open Data Framework (4.1) Zboson 0.977 0.977 (4.2) Wboson 1.013 1.013 (4.3) SingleTop 1. p. 6014 1.014 (4.4) TTbar 1.047 1.047 (4.5) WZ 1.053 1, The results were compared with those obtained using the ATLAS Open Data C++ Framework

Table 29: Event yields after applying the loose signal region selections

Sample ADL/CutLang Opendata Framework

Data 60.00 ± 7.75 58.00 Z + jets 5. p. 031 ± 1.15 6.00 Diboson 18. p. 221 ± 1.56 18, The results were compared with those obtained using the ATLAS Open Data C++ Framework

Table 30: Event yields after all selections for each sample in the single-lepton channel

Sample ADL/CutLang Opendata Framework

Data 13873.00 ± 117.78 Diboson 46, The results were compared with those obtained using the ATLAS Open Data C++ Framework

Table 31: Cutflow of event counts for each sample in the single-top channel

Selection Data tq ZVV tt, W t, tb ¯ W + jets All 168533115.00 ± 27442.09

Table 32: Event yields after tt¯→ W+W−b¯b → lνqq¯'b¯b channel selections

Sample ADL/CutLang Opendata Framework Data 150502.00 ± 387.95 ttbar 136844.62 ± 120.08

Table 33: Event yields after all the selections for each sample in the W ±Z → lνll′ channel

Sample ADL/CutLang Opendata Framework Data 918.00 ± 30.30 W Z 799.43 ± 3.45

Table 34: Event yields after all the selections for each sample in the W ±Z → lνll′ channel

Sample ADL/CutLang Opendata Framework Data 918.00 ± 30.30 W Z 799.43 ± 3.45

Table 18: Event yields after all the selections for each sample in the ZZ → l+l−l′+l′− channel

Sample ADL/CutLang Opendata Framework Data 93.00 ± 9.64 ZZ 85.22 ± 0.52

Table 17: Cutflow of event counts for each sample in the W ±Z → lνll′ channel

Selection Data WZ ZZ,WW V + jets Other All 73490.00 ± 289.27

Table 15: Cutflow of event counts for each sample in the H → ZZ∗ → l+l−l′+l′− channel

Selection Data Higgs ZZ Other All 832.0 ± 28.8444

Table 16: Event yields after all the selections for each sample in the H → ZZ∗ → l+l−l′+l′− channel

Sample ADL/CutLang Opendata Framework Data 8.50 ± 0.02 Higgs 8. p. 250

Table 19: Cutflow of event counts for each sample in the Z′ → tt¯→ W+b, W−¯b → qq¯'b, l−ν¯b channel

Selection Data Single Top V + jets Diboson ttbar All 466034.00 ± 1238452.00

Table 21: Event yields after applying the tight signal region selections

Sample ADL/CutLang Opendata Framework Data 6.00 ± 2.45 Z + jets 0. p. 004 ± 0.03

Table 25: Event yields after bTag(goodDRJets[0]) == 1 selection for each sample in the Z′ → tt¯→ W+b, W−¯b → qq¯'b, l−ν¯b channel

Sample ADL/CutLang Opendata Framework Data 8433.00 ± 91.83 Single Top 311. p. 271 ± 4.99

Data/Monte Carlo Analysis ADL/CutLang Open Data Framework (4.1) Zboson 0.977 0.977 (4.2) Wboson 1.013 1, The results were compared with those obtained using the ATLAS Open Data C++ Framework

Table 23: Event yields after Size(goodDRjets) ≥ 1 selection for each sample in the Z′ → tt¯→ W+b, W−¯b → qq¯'b, l−ν¯b channel

Sample ADL/CutLang Opendata Framework Data 76742.00 ± 277.02 Single Top 4919. p. 283 ± 20.45

Table 24: Event yields after Size(topLRjets) == 1 selection for each sample in the Z′ → tt¯→ W+b, W−¯b → qq¯'b, l−ν¯b channel

Sample ADL/CutLang Opendata Framework Data 13988.00 ± 118.27 Single Top 610. p. 286 ± 6.88

Table 22: Cutflow of event counts for each sample in the Z′ → tt¯→ W+b, W−¯b → qq¯'b, l−ν¯b channel

Selection Data Single Top V + jets Diboson ttbar All 466034.00 ± 1242954.00

Table 13: Cutflow of event counts for each sample in the W ±Z → lνll′ channel

Selection Data WZ ZZ,WW V + jets Other All 73490.00 ± 289.27

Table 14: Event yields after all selections for each sample in the W ±Z → lνll′ channel

Sample ADL/CutLang Opendata Framework Data 918.00 ± 30.30 W Z 799.43 ± 3.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved analysis preservation through decoupling physics logic from experiment-specific software infrastructures by using an Analysis Description Language (ADL) and CutLang runtime interpreter to produce event selections that can be readily fed into statistical analysis tools.

  2. Enhanced reproducibility and transparency by providing a common language for describing analyses where all physics objects and event selection criteria are expressed in a human-readable form using ADL.

  3. Increased educational accessibility by offering a pedagogical tool where the ADL/CutLang workflow allows students to develop a clear understanding of the physics being studied through standardized, declarative descriptions.

  4. Increased computational efficiency by allowing CutLang to execute analysis descriptions directly at runtime without requiring prior compilation, facilitating flexible and portable workflows across different input data formats.

  5. Improved performance scaling by leveraging the multi-core architecture of modern computers to execute analyses in parallel, with a comparison showing that CutLang can utilize this capability for large amounts of collected data and simulation events.

  6. Better handling of complex kinematic reconstructions by implementing an optimization mechanism within CutLang, such as the one used in Higgs boson production, which can identify more candidates close to the nominal Z boson mass compared to the Open Data framework’s implementation with multiple independent if conditions.

  7. Robust validation across different analysis types by demonstrating that ADL/CutLang results yield good agreement with those obtained from the original publications, providing a reliable alternative to traditional analysis frameworks for SM and BSM searches.

Related papers