CELEBI: Towards Reproducible and FAIR Physics Analysis

In the world of high-energy physics, scientists do not simply look through a telescope or a microscope to find answers. Instead, they build massive, complex machines that smash particles together, creating a storm of data that must be sorted, measured, and understood. This process is not a single event but a long, winding journey of discovery. A researcher takes raw data and runs it through a series of computer programs, making choices about which pieces of information to keep and which to discard. They adjust settings, test different theories, and refine their methods over months or even years. The final result is a scientific finding, but that finding is inextricably linked to the specific path the team took to get there. If the code changes, or if a parameter is tweaked, the result might change too. The great challenge in this field is not just getting the answer, but remembering exactly how it was found. Without a clear record of the steps taken, a result from last year might become impossible to reproduce today, leaving the scientific community unable to verify or build upon the work.

This is the problem Mingrui Zhao, a researcher at Peking University, set out to solve with a new system called Celebi. In the paper, Zhao describes how modern physics analyses evolve continuously. As datasets grow and software updates arrive, the conditions under which an earlier result was produced can become difficult to determine. Imagine a team working on a complex calculation where the code is constantly being rewritten. A result presented during an internal review might no longer match the current version of the analysis because the underlying rules have shifted. In large collaborations, where many people are working on different variations of the same project, this confusion can be overwhelming. The researcher found that existing tools could manage the flow of data or store the final files, but they struggled to keep a persistent, unbreakable link between a specific result and the exact state of the analysis that created it.

To address this, Zhao designed Celebi as a framework that treats the analysis not as a single, fluid document, but as a collection of structured, reusable parts. The system separates the lightweight instructions—such as the code, the configuration settings, and the dependencies—from the heavy data products that are generated. It organizes these instructions into a clear workflow where every piece knows what it depends on. The core innovation lies in how the system handles change. As the working analysis evolves, Celebi preserves specific moments in time called "impressions." An impression is a frozen snapshot of the analysis at a particular moment, capturing the code, the parameters, and the exact state of every dependency used to produce a result. These impressions are immutable, meaning they cannot be altered once saved. When a result is generated, it is permanently associated with the specific impression that created it. This means that even if the main analysis is updated a hundred times, the original result remains tied to the exact conditions under which it was first produced.

The practical effect of this organization is a system that allows scientists to modify their work without losing their history. Because the components are modular, a researcher can change one part of the analysis, such as the method used to filter data, without having to rewrite the entire project. The system automatically tracks which downstream tasks are affected by that change. If a scientist wants to run a systematic variation, perhaps testing how a different assumption affects the outcome, they can do so by reusing the unchanged parts of the analysis while only modifying the specific section that differs. This reduces duplication and makes the differences between various configurations explicit. The system also supports the controlled rerunning of these preserved states. If a team needs to verify an old result, they can execute the preserved impression again, provided the necessary inputs and software environments are still available, ensuring that the new run is a true reproduction of the old one.

This approach fundamentally changes how collaboration and review happen within a physics team. In a typical project, multiple collaborators might be working on different configurations, and results might be scattered across various files and personal directories. Celebi brings order to this by organizing all materials, documentation, and dependencies into a structured repository. A collaborator can inspect a specific component along with its local documentation and the exact state of its dependencies without needing to reconstruct the entire analysis from scratch. This makes it easier for new team members to understand the work and for reviewers to check the logic behind a result. When a reviewer asks for a change, the team can update the analysis and create a new impression, keeping the old one alongside the new. This allows for direct comparison between revisions, showing exactly what changed and how it affected the outcome. The system essentially builds a transparent timeline of the research, where every result can be traced back to its source.

The ultimate goal of this work is to make physics analysis more FAIR: Findable, Accessible, Interoperable, and Reusable. Currently, reusing a scientific result often requires not just the final data, but a deep understanding of the code and settings used to create it, which are often lost or buried. By associating results with identifiable, preserved states of the analysis, Celebi provides a foundation for finding and reusing these materials. The structured nature of the system allows computers to interpret the analysis logic, which could help different research infrastructures work together. While the paper does not claim that Celebi solves every problem in scientific preservation, it offers a practical foundation for extending these principles beyond just the raw data to the computational processes that generate scientific knowledge. The system does not rely on manual bookkeeping or retrospective reconstruction of history; instead, it captures the provenance of the work as it happens, day by day. In doing so, it ensures that the path to a discovery remains as clear and accessible as the discovery itself.