Hidden Heroes in Software Ecosystems

• Hidden heroes are underrecognized yet pivotal software packages and developer contributions that enable the backbone of complex research and software systems.
• They are identified using network analysis methods such as Katz centrality and dependency graph mapping, which reveal their critical roles despite low direct citation.
• Empirical findings show that these hidden figures enhance project stability and efficiency, underscoring the need for improved recognition and funding models.

In computational research and software engineering, "hidden heroes" denote the foundational, often underrecognized entities—be they critical software packages in scientific ecosystems or key contributors in development teams—that underpin the visible success of complex projects. Despite their pivotal role, both varieties of hidden heroes tend to remain in the background, accruing minimal direct credit even as they are indispensable to downstream performance and sustainability (Nesbitt et al., 2024, Agrawal et al., 2017).

1. Conceptualizing Hidden Heroes in Software Ecosystems

Hidden heroes in the context of biomedical open-source software refer to low-visibility, high-impact packages positioned deep in dependency hierarchies. These are foundation libraries enabling the usability and effectiveness of more prominent, user-facing tools. Users and even tool developers predominantly cite only the visible top-level utilities (e.g., ggplot2, pandas), neglecting the substantial contributions of the foundational dependencies that these tools themselves invoke. Consequently, the scientific credit system overlooks these foundational elements, leaving them without adequate recognition or institutional support despite their critical enabling function (Nesbitt et al., 2024).

In collaborative software development, particularly in open-source and enterprise contexts, the term also describes "hero" developers—individuals responsible for a disproportionate share of the substantive work in a codebase. Hero developers are formally defined as the smallest 20% of the team whose combined contributions (commits) account for at least 80% of all work, following the Pareto principle (Agrawal et al., 2017).

2. Identification and Network Analysis Methodologies

Within the biomedical software domain, the detection of hidden heroes leverages the Chan Zuckerberg Initiative Software Mentions Dataset, which utilizes an NLP model trained on the Softcite corpus to extract software mentions from full-text biomedical articles. Detected mentions are mapped to packages in three major open-source software ecosystems: PyPI (Python), CRAN (R), and Bioconductor (R). For each directly mentioned package, all declared dependencies (including development, testing, and documentation packages) are recursively resolved, constructing an exhaustive list of direct and transitive dependencies and their relationships to published works.

The resulting structure is encoded as a directed, weighted bipartite graph $G=(V,E,W)$ with four node types—DOI-identified papers, PyPI packages, CRAN packages, and Bioconductor packages. Edges from papers to software are weighted by the OpenAlex citation count of the citing paper, while software-to-software dependency edges are unweighted but directed according to the declared dependency graph.

In software development teams, the extraction of hidden hero contributors is algorithmic: full project histories are mined for commit data, developers are sorted by contributions, and the minimal prefix covering 80% of commits is checked against the team’s 20% threshold to identify hero projects (Agrawal et al., 2017).

3. Centrality Metrics and Formal Definitions

To evaluate the structural importance of software packages within dependency graphs, standard network centrality measures are employed:

• Degree Centrality: $C_D(v) = \text{deg}(v) / (N - 1)$, where $\text{deg}(v)$ is the number of incident edges and $N$ is the total number of nodes.
• Betweenness Centrality: $$C_B(v) = \sum_{s \ne v \ne t} \sigma_{st}(v) / \sigma_{st}$$, quantifying control over information flow via shortest paths.
• Closeness Centrality: $C_C(v) = (N-1) / \sum_{t} d(v,t)$, measuring average proximity to all other nodes.
• Eigenvector Centrality: $C_E(v) = (1/\lambda)\sum_{u \in N(v)} C_E(u)$, where $\lambda$ is the leading eigenvalue.

However, due to the directed, weighted structure and the one-way character of paper-to-software edges, Katz centrality is preferred:

$C_K(v) = \alpha \sum_{u} A_{uv} C_K(u) + \beta b_v$

where $A$ is the adjacency matrix, $b_v$ incorporates external boosts (e.g., citation weight), and $\alpha$ is a damping factor (Nesbitt et al., 2024).

For hero contributors, the formal criterion is:

$$H = \{\,d\in Dev\mid contributions(d)\ge 0.8\cdot\sum_{d'\in Dev}contributions(d')\}\,.$$

4. Empirical Findings: Exemplars and Prevalence

Analysis of 661 public and 171 enterprise GitHub projects demonstrates the ubiquity of hero contributors: 77% of public and 78% of enterprise projects qualify as hero projects. The prevalence increases with team size—rising to 95% for large public projects and 93% in large enterprise groups. Notably, the presence of hero contributors in large teams is nearly universal, contradicting the assumption that collaborative diffusion occurs at scale (Agrawal et al., 2017).

In the software package ecosystem, plotting mention counts against Katz centrality for PyPI, CRAN, and Bioconductor reveals the quadrant occupation of hidden heroes: they are characterized by low direct mention but high centrality. Principal exemplars include:

Ecosystem	Package	Mentions	Katz Centrality	Functional Role
PyPI	velvet	197	0.105	de Bruijn-graph-based genome assembler
CRAN	rlang	~82	0.093	core metaprogramming for tidyverse, critical dependency
Bioconductor	S4Vectors	~62	0.089	foundational S4 vector and list containers

These examples represent keystone packages supporting extensive downstream tools, in many cases without receiving explicit citation or recognition (Nesbitt et al., 2024).

5. Impact on Software Ecosystem Stability and Project Health

The indispensability of hidden hero software packages is illustrated by their support of fundamental algorithms or data structures relied upon by a majority of higher-level tools. For example, "velvet" provides de novo assembly algorithms foundational for sequence analysis, "rlang" enables tidy evaluation underpinning nearly every modern R-based analytics tool, and "S4Vectors" furnishes interoperable containers for genomic and single-cell data structures.

In development teams, analysis shows that hero-dominated projects do not suffer slower issue or bug resolution rates. Median resolution times for overall issues and bugs are statistically indistinguishable between hero and non-hero projects in both public and enterprise contexts. The critical divergence emerges in enhancement (feature) resolution: in public projects, non-hero teams outperform hero teams, while in enterprise settings, hero teams resolve enhancements at a significantly higher rate (median $E_r/E_t = 0.68$ for enterprise hero projects versus $0.52$ for non-hero) (Agrawal et al., 2017).

6. Implications for Recognition, Funding, and Management

The marginalization of hidden heroes engenders risks to the sustainability of both scientific infrastructures and software projects. Recommendations for supporting these critical elements include:

• Mandating formal citation of all direct dependencies in software metadata (e.g., via standardized citation files).
• Allocating maintenance funding to packages with high centrality even if their direct usage metrics are low.
• Incorporating transitive-credit metrics, derived from network-based centralities, in evaluations of developer contributions for hiring, tenure, and promotion.
• Publishing dependency graphs and centrality dashboards to surface the true software-infrastructure backbone to users, reviewers, and funders.
• Strategically tracking and retaining hero contributors in enterprise environments, and rotating enhancement responsibility among top contributors in OSS to prevent burnout (Nesbitt et al., 2024, Agrawal et al., 2017).

A plausible implication is that failure to recognize and incentivize hidden heroes may compromise the long-term reproducibility, maintainability, and progress in scientific and enterprise software ecosystems.

7. Statistical and Methodological Tables

Prevalence of Hero Projects by Team Size

Team Size	Public Hero %	Enterprise Hero %
Small	72%	68%
Medium	85%	80%
Large	95%	93%

Enhancement Resolution Ratios

Type	Group	Median $E_r/E_t$	Scott–Knott Rank
Public	Hero	0.42	Rk:2
Public	Non-Hero	0.52	Rk:1
Enterprise	Hero	0.68	Rk:2
Enterprise	Non-Hero	0.52	Rk:1

These statistical findings reinforce that hidden heroes—whether infrastructural packages or principal contributors—are systemic phenomena. They provide scaffolding for the productivity and reliability of research and development undertakings, necessitating deliberate policies for their cultivation and retention (Nesbitt et al., 2024, Agrawal et al., 2017).