Making contributions visible: An interactive authorship viewer for team science

Introduction¶

The author byline has served as the primary unit of academic credit for over three centuries. Despite radical transformation in how science is conducted, from the solitary natural philosopher to hundred-person collaborations spanning continents, the format of authorship attribution has remained essentially unchanged: a linear, ordered list of names.

This compression of multi-dimensional contributions into a one-dimensional sequence creates well-documented tensions. Early-career researchers depend on first-author publications for career advancement. Principal investigators expect last-author or corresponding-author designations to signal intellectual leadership. Middle authors, often building critical infrastructure, contributing specialized techniques, or performing essential validation, receive minimal credit differentiation.

The consequences are not merely symbolic. Authorship disputes are among the leading causes of research misconduct complaints Rennie et al. (1997). Junior researchers report delaying publications and avoiding collaborations to protect their authorship position Sauermann & Haeussler (2017). The incentive structures that flow from linear author lists actively hinder the collaborative, interdisciplinary team science that modern challenges demand Wuchty et al. (2007).

We propose a fundamentally different approach. Rather than compressing contributions into a single ordered list, we present a multi-dimensional authorship framework that allows readers, institutions, and metrics systems to explore contributions along multiple axes simultaneously. This framework builds on the CRediT (Contributor Roles Taxonomy) standard while extending it with section-level contributions, project timelines, and rich author profiles. Critically, we reject the premise that any single ordering of authors is definitive. Using our tool, the reader can dynamically re-sort the author list by alphabetical order, number of CRediT roles, specific role contributions, project join date, or institution.

This paper serves simultaneously as a description of the framework and as a live demonstration. Built with MyST (Markedly Structured Text), the interactive authorship display above illustrates how rethinking author display can bring transparency, reduce conflict, and credit the diverse contributions that make modern science possible.

The CRediT taxonomy¶

The Contributor Roles Taxonomy (CRediT), developed by the Consortia Advancing Standards in Research Administration Information (CASRAI) and first published in 2015 Brand et al. (2015), defines fourteen distinct contributor roles: Conceptualization, Methodology, Software, Validation, Formal Analysis, Investigation, Resources, Data Curation, Writing—Original Draft, Writing—Review & Editing, Visualization, Supervision, Project Administration, and Funding Acquisition. For each role, contributors can be assigned a degree of contribution: lead, equal, or supporting.

CRediT has gained nominal adoption. Many submission platforms now include CRediT fields McNutt et al. (2018), and publishers such as Elsevier encourage authors to provide a CRediT author statement. In practice, however, these statements typically appear as a brief, unstructured paragraph at the end of the paper. They list role names per author without indicating contribution levels, without machine-readable structure, and without any prominence in the article layout. Some publishers, including Springer Nature, do not use CRediT at all, relying instead on free-text author contribution statements. The gap between the taxonomy’s potential and its current implementation underscores the need for richer, reader-facing attribution tools Allen et al. (2014)Larivière et al. (2016).

Limitations of current approaches¶

Current authorship display suffers from several interconnected limitations. First, the linear ordering implies a ranking that may not exist. In team science, contributions are often orthogonal: the person who built the data infrastructure, the person who designed the experiments, the person who wrote the theoretical framework, and the person who led the experimental work each contribute indispensably but along entirely different dimensions. Compressing these into positions 1 through N forces a false hierarchy.

Second, the convention of “first author did most work” and “last author is the PI” is not universal. In high-energy physics, for example, collaborations involving hundreds or thousands of researchers have stretched the traditional authorship model to the point where author position conveys little about individual contributions Birnholtz (2006). In mathematics and economics, authors are traditionally listed alphabetically.

Third, the static nature of the author list conceals the temporal dynamics of contributions. Projects increasingly involve contributors who participate intensely during one phase but not others. A publication often begins with a key conceptual insight, followed by intensive experimental design and execution, and later by data curation and analysis: each phase potentially led by different people. The byline flattens this rich temporal structure into a single snapshot, hiding who contributed when and to what phase of the work.

Team science and rising author lists¶

The average number of authors per paper has grown steadily across virtually all scientific fields Wuchty et al. (2007). In biomedicine, papers with more than fifty authors are no longer unusual, and in fields like particle physics and climate science, author lists routinely exceed one thousand names. Even in traditionally small-team fields like psychology and economics, the median author count has roughly doubled in the past two decades.

#!/usr/bin/env python3
"""Generate the author-count-growth figure as PNG."""
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import matplotlib.ticker as ticker
import yaml

with open('author-count-growth.yaml') as f:
    raw = yaml.safe_load(f)

data = raw['data']
years = [d['year'] for d in data]
means = [d['mean'] for d in data]
medians = [d['median'] for d in data]

fig, ax = plt.subplots(figsize=(8, 4))
ax.plot(years, medians, 'o-', color='#4c6ef5', linewidth=2.5, markersize=6, label='Median', zorder=3)
ax.plot(years, means, 's--', color='#93c5fd', linewidth=1.8, markersize=5, label='Mean', zorder=2)
ax.fill_between(years, medians, means, alpha=0.08, color='#4c6ef5')
ax.set_xlabel('Year', fontsize=11)
ax.set_ylabel('Authors per paper', fontsize=11)
ax.set_title('Mean number of authors per paper in biomedical sciences (2003\u20132025)',
             fontsize=12, fontweight='bold', pad=12)
ax.legend(frameon=True, fancybox=True, shadow=False, fontsize=10)
ax.grid(True, alpha=0.3, linestyle='--')
ax.set_xlim(2002, 2026)
ax.xaxis.set_major_locator(ticker.MultipleLocator(4))
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
fig.text(0.99, 0.01, 'Source: OpenAlex API \u00b7 200 highly-cited articles/year \u00b7 Medicine & Biology',
         ha='right', fontsize=7, color='#9ca3af', style='italic')
plt.tight_layout()
plt.savefig('author-count-growth.png', dpi=150, bbox_inches='tight')
print('Saved author-count-growth.png')

# ============================================================
# Mean number of authors per paper over time
# ============================================================
# AUTO-GENERATED by figures/fetch_author_count_growth.py
# Generated: 2026-03-15
# Source: OpenAlex API (https://openalex.org). Filtered to journal articles in Medicine and Biology concepts. Sample of 200 highly-cited articles per year.
# ============================================================

title: Mean authors per paper in biomedical sciences
xLabel: Year
yLabel: Mean authors per paper
source: OpenAlex API (https://openalex.org). Filtered to journal articles in Medicine
  and Biology concepts. Sample of 200 highly-cited articles per year.
generated: '2026-03-15'
data:
- year: 2003
  mean: 7.0
  median: 4.0
- year: 2005
  mean: 8.36
  median: 5.0
- year: 2007
  mean: 10.29
  median: 6.0
- year: 2009
  mean: 10.96
  median: 7.0
- year: 2011
  mean: 12.64
  median: 8.0
- year: 2013
  mean: 15.68
  median: 9.0
- year: 2015
  mean: 16.14
  median: 9.0
- year: 2017
  mean: 20.52
  median: 9.0
- year: 2019
  mean: 19.98
  median: 10.0
- year: 2021
  mean: 24.36
  median: 15.0
- year: 2023
  mean: 30.79
  median: 19.5
- year: 2025
  mean: 22.12
  median: 12.0

This growth reflects a genuine change in how research happens. Modern scientific problems, from understanding complex systems to building research infrastructure, require deep specialization in design, analysis, engineering, data management, and domain knowledge that no individual or small group possesses. The question is not whether to do team science but how to credit it fairly.

Current systems fail team science in specific, quantifiable ways: middle authors, who often include infrastructure contributors such as engineers and data curators, contribute to fewer documented tasks and occupy positions associated with less credit than first or last authors Larivière et al. (2016). Trainees who make crucial but bounded contributions face pressure to negotiate for first-author status Sauermann & Haeussler (2017). The proliferation of workarounds (co-first authors, co-second authors, co-senior authors, equal-contribution footnotes) reveals how strained the linear model has become: teams are inventing ad hoc mechanisms to inject dimensionality into a format that was never designed for it. A richer attribution framework would reduce these tensions by making the nature and extent of each contribution transparent Rennie et al. (1997)Holcombe (2019).

Data model¶

Our framework extends the CRediT vocabulary Brand et al. (2015) with three additional layers of contribution information. The first layer is section-level contributions: for each section of the manuscript, we record which authors contributed to the work described in that section and in what capacity. This goes beyond documenting who wrote the text: it captures who performed the research, designed the methodology, built the tools, or generated the results that a given section reports. A section on data infrastructure, for instance, credits the engineers who built the pipeline, not merely the person who drafted the prose.

The second layer is figure-level contributions, documenting who generated data, created visualizations, or designed the presentation for each figure and table. Figures often represent the most labor-intensive elements of a paper and the elements most often reused and cited. Attributing them individually adds significant granularity to the contribution record.

The third layer is a project timeline: when each contributor joined the project, key milestones in their participation, and optionally when their active contribution ended. This temporal dimension makes visible the trajectory of a project and the different phases of contribution that characterize modern collaborative research.

Each author profile includes structured metadata: institutional affiliations (linked to ROR IDs where available) and persistent identifiers such as ORCID. The core data model focuses on permanent, publication-relevant data: the facts about contributions that form part of the scholarly record. Beyond this archival layer, the display supports optional presentational metadata such as a profile photo (avatar_url). When a photo is provided, it replaces the default colored-initials avatar across all views, giving readers a more immediate, human connection to each contributor.

Interactive display¶

A central design principle is that there is no single correct author ordering. The system supports dynamic re-sorting along multiple dimensions: alphabetical, number of CRediT roles, project join date, institution, and by any specific CRediT role. A reader interested in who led the conceptualization can sort by that role; a hiring committee can sort by Software; a student can sort by Methodology. By making re-sorting immediate, we shift the message from “these authors are ranked” to “these authors contributed along multiple dimensions that you can explore.”

The interactive panel also allows the reader to switch between multiple datasets: for this paper, a simulated small team, a simulated large team of fifty authors, and the real contributor data. This makes it possible to explore how the framework scales and adapts to different team sizes.

The framework presents authorship information through six complementary views, accessible via tabs in the interactive panel above. The Collaboration tab offers four switchable network modes: a collaboration layout (the default) where a force-directed simulation positions authors so that proximity reflects collaboration strength, and three circle-of-circles views grouping authors by institution, role, or manuscript section and figure. The CRediT tab encodes the contribution landscape as a colored grid, with lead, equal, and supporting contributions distinguished by intensity. The Sections tab maps which authors contributed to each manuscript section and figure. The Timeline reveals when each contributor joined and what phases they participated in. The Sorted List allows dynamic re-sorting by alphabetical order, number of CRediT roles, any individual role, project join date, or institution, making explicit that author ordering is a choice, not a fact. The Profiles view displays structured author metadata including affiliation, CRediT roles, ORCID identifiers, and profile photos when available.

Methods¶

The framework is implemented as a MyST (Markedly Structured Text) plugin, leveraging the MyST ecosystem for scientific publishing. MyST extends Markdown with structured roles and directives designed for technical and scholarly content, supporting cross-references, citations, mathematical notation, and, critically for this work, executable and interactive components embedded directly in the document.

Contribution data is stored in YAML files that follow a schema extending the CRediT taxonomy. Each author entry includes standard metadata (name, affiliation, ORCID), CRediT roles with contribution levels (lead, equal, supporting), section-level contributions with free-text descriptions, figure-level attributions, and a timeline with join date and milestones. This structured format is both human-readable and machine-parseable, making it straightforward for authors to maintain and for downstream tools to consume.

The interactive display is implemented as a custom MyST directive ({authorship-explorer}) backed by an anywidget-compatible ESM module. At build time, the plugin reads the YAML author files, validates the schema, walks the document AST to extract section heading labels, and injects the structured data into the document as a widget node. At render time, the widget constructs the full interactive interface entirely in the browser using vanilla JavaScript and SVG, with no external dependencies beyond the MyST theme’s anywidget renderer.

The collaboration layout uses a force-directed simulation with repulsive forces between all nodes and attractive spring forces proportional to collaboration weight (shared CRediT roles plus shared section contributions), converging over 300 iterations with velocity damping and collision resolution to position authors so that spatial proximity directly reflects collaboration strength. Connections between authors are drawn as per-role minimum spanning trees: for each CRediT role, a minimum spanning tree connects all authors who share that role, with edge weights derived from spatial distance. This reduces visual clutter from O(n²) all-pairs edges to O(n) edges per role while preserving the network’s topology. On hover, only the MST trees for the hovered author’s roles are highlighted, with individual role strands rendered as parallel colored lines.

The implementation renders inside a Shadow DOM to isolate widget styles from the host document, ensuring consistent appearance across different MyST themes. The widget also injects minimal styles into the host document to relocate itself into the article’s frontmatter region and hide the default linear author list, replacing it with the interactive display.

The plugin and widget code were developed with the assistance of Claude Opus 4.6 (Anthropic), used as a coding agent within VS Code. The LLM generated the initial implementation of the MyST plugin, the anywidget ESM module, and the SVG-based visualizations, with the author providing architectural direction, iterative feedback, and manual refinements. This paper was also drafted collaboratively with the same model.

Discussion¶

The framework presented here addresses a structural problem in scholarly communication that has been recognized for decades. As early as 1997, Rennie, Yank, and Emanuel argued in JAMA that traditional authorship was failing and proposed replacing it with explicit contributorship Rennie et al. (1997). Their core insight, that a single byline cannot meaningfully represent the diverse roles within a research team, remains as urgent today as it was then, perhaps more so given the continued growth of team science documented by Wuchty, Jones, and Uzzi Wuchty et al. (2007).

Institutional momentum has been building. The call by Allen et al. in Nature to move “beyond authorship” toward transparent attribution Allen et al. (2014), the cross-publisher consensus statement by McNutt et al. on CRediT adoption McNutt et al. (2018), and Holcombe’s argument that CRediT should replace traditional authorship entirely Holcombe (2019) have established a clear direction. Yet none of these efforts have changed what the reader actually sees.

On the technical side, several tools and standards have advanced structured contributorship, but each addresses only part of the problem. A feature-by-feature comparison clarifies what remains missing.

Publisher displays. PLOS was an early adopter of CRediT and now displays per-author roles on a dedicated Authors tab for each article. Each author entry lists CRediT role names (e.g. “Conceptualization, Data curation, Formal analysis”), an institutional affiliation, and an ORCID link. However, the display is static text, roles are binary (present or absent, with no lead/equal/supporting distinction), and there is no visualization, no interactivity, and no section-level attribution. The contribution information lives on a secondary tab rather than in the article’s main view. eLife similarly lists CRediT roles and ORCID links beneath each author name at the bottom of the article. Both publishers surface structured contribution data, but the display remains static text with binary roles, no visualization, and no way for the reader to re-sort or explore contributions dynamically.

Authorship documentation tools. Tenzing Holcombe et al. (2020) is a web application and R package that lets teams document CRediT contributions via a shared spreadsheet and generate outputs such as prose contribution statements, JATS XML metadata, and structured affiliation blocks. It is an input tool, not a display tool: its outputs are static text or metadata files, not interactive reader-facing components.

Ontologies and metadata standards. The Contributor Role Ontology (CRO) by Vasilevsky et al. Vasilevsky et al. (2021) extends CRediT into a formal OWL ontology with finer-grained roles, making contribution data machine-readable and linkable to ORCID identifiers. The FORCE11 community has advocated for structured scholarly metadata more broadly Wilkinson et al. (2016). These efforts provide essential infrastructure for computational attribution, but they define schemas and vocabularies, not reader-facing tools. No existing ontology or metadata standard produces a display that a reader can explore.

What distinguishes the present work is its focus on the reader-facing display and interaction layer. The existing landscape addresses pieces of the attribution problem: CRediT defines the vocabulary, CRO makes it machine-readable, Tenzing streamlines data collection, and PLOS and eLife surface structured roles on article pages. None of these efforts, however, provides an interactive, multi-dimensional view of contributions embedded directly in the paper. Our framework closes this gap. It records contribution levels (lead, equal, supporting) rather than binary presence, attributes contributions to individual manuscript sections and figures, captures temporal dynamics, and presents all of this through six complementary interactive views that the reader can explore, re-sort, and query. The explicit design choice that no single author ordering is privileged shifts the message from “these authors are ranked” to “these authors contributed along dimensions you can explore.”

Several implications deserve emphasis. First, this framework makes visible the contributions of individuals who are systematically under-credited in current systems. Data curators, engineers, research coordinators, and technical specialists are increasingly essential to modern science but remain largely invisible in traditional author lists Larivière et al. (2016). Students and postdocs fare little better: their contributions are routinely compressed into an ambiguous middle-author position that tells a hiring committee almost nothing. Structured contribution records change this. When a reader can sort by “Data Curation” or “Software” and see who led these contributions, the value of that work becomes self-evident. This gives early-career researchers and support staff a concrete artifact they can present to hiring committees, fellowship panels, and promotion boards.

Second, the framework reduces the stakes of author ordering by making position less consequential. When every contribution is transparently documented and can be viewed from multiple angles, the specific position in the byline matters less. This alone could significantly reduce authorship disputes, which remain among the most common sources of friction in collaborative research Rennie et al. (1997).

Third, rather than pressuring teams to restrict author lists, this framework allows them to grow. When the only way to credit someone is a position in a linear ranking, every addition dilutes the signal for everyone else, creating an incentive to exclude contributors whose roles are harder to categorize. With structured contribution data, adding an author costs nothing: their specific roles are documented, and no one else’s credit is diminished. This removes an artificial gatekeeping mechanism and encourages teams to include everyone who contributed.

Fourth, the machine-readable nature of the contribution data opens new possibilities for research evaluation Vasilevsky et al. (2021)Wilkinson et al. (2016). Funding agencies, hiring committees, and promotion panels currently rely on crude proxies such as author position and journal prestige. Structured contribution data enables genuinely nuanced evaluation, crediting researchers for the specific roles they performed.

Fifth, structured contribution data could evolve into a new de facto standard for scholarly attribution. Today the ordered author list is that standard, not because it is adequate but because nothing better has displaced it. The YAML format used in this framework offers a concrete alternative: a simple, human-readable, machine-parseable file that any researcher can create, any version-control system can track, and any tool can consume. If journals, preprint servers, and funding agencies converged on a common contribution schema, even a minimal one recording CRediT roles and contribution levels, an author’s ORCID profile could aggregate structured contribution records across all their publications, producing a contribution portfolio rather than a publication list. Hiring committees would no longer see a CV listing “Author 4 of 12” on a Nature paper; they would see that the candidate led Software and Data Curation across a body of work and contributed equally to Formal Analysis in two others.

The interactive display operates within the scope of a single paper and replaces the traditional byline with a richer, multi-dimensional view. It is fully compatible with existing citation infrastructure: DOI metadata, BibTeX records, and in-text citations such as “Smith et al. (2025)” continue to work unchanged. For the author ordering registered in citation databases and reference managers, we recommend alphabetical listing, a convention already standard in mathematics, economics, and high-energy physics, so that the recorded order carries no implied ranking. We view the interactive display as a necessary step toward shifting the culture of credit attribution: once contribution data is routinely recorded and visible, the community can build on it. Future work on citation conventions could further reduce name-based credit biases, for instance by citing large collaborations under a project name or by replacing “first-author et al.” with human-readable short titles linked to DOIs.

Limitations of this framework include the documentation burden on authors, the need for journal and publisher adoption, and the challenge of verifying self-reported contributions. We believe these limitations are addressable: modern submission systems already collect CRediT data McNutt et al. (2018), section-level contributions map naturally to how teams already divide work, with each contributor knowing which parts of the project they led or supported, and the transparency of public contribution records itself serves as a check on misrepresentation. The contribution data itself is stored in simple YAML files that any researcher can edit in a text editor, version-control with Git, and validate without specialized tools. This low barrier to entry means adoption does not depend on any particular publishing platform.

Moreover, the documentation effort is modest compared to the time researchers already spend negotiating authorship order. In our experience, these conversations can consume hours of discussion, generate lasting resentment, and occasionally derail collaborations entirely. Filling out a structured contribution record takes minutes. More importantly, the process itself can diffuse tensions: when contributors see their specific roles documented transparently, the pressure to fight for a particular byline position diminishes. The documentation is not an added cost but a replacement for a far more expensive and time-consuming process.

Adoption does not require journals to migrate to MyST or any particular publishing platform. The interactive display demonstrated here is implemented entirely in standard JavaScript and SVG, with no framework dependencies. Any journal website that can embed a script tag can host an interactive contribution viewer alongside the traditional article layout. Journals already routinely embed interactive figures, supplementary viewers, and altmetric widgets. An authorship explorer is no different in kind. Rather than treating authorship disputes as an inevitable cost of publication, journals could provide structured contribution tools as part of the submission process, helping authors document and resolve attribution questions before they become conflicts. The workflow is straightforward: each author fills in their own contributions in parallel, the tool aggregates the entries and flags discrepancies, and the team reviews and approves the final record together. This is a solvable engineering problem, not a conceptual one.

The future of scientific authorship must evolve to match the reality of scientific practice. Team science is here to stay and growing Wuchty et al. (2007). The question is whether our attribution systems will continue to compress multi-dimensional contributions into a format designed for solo scholars, or whether we will build tools that celebrate the collaborative nature of discovery and give every contributor the credit they deserve.