It is tempting to read a contributor distribution as a failure. The pattern is real and severe. In an analysis of Apache Software Foundation projects, 2,786 contributors — about 59.8% of the population — produced only 2.27% of all commits [7]. The 90–9–1 rule of thumb, in which roughly one percent of users create most content, has been observed across online communities for decades [8]. But a large passive tier does not by itself make a project fragile. What makes it fragile is the mirror image of that distribution: the small set of people who hold the operational knowledge, review the changes, and carry the invisible coordination work. If they leave, the project stalls.
The bus factor names that risk precisely. It is the minimum number of people whose sudden unavailability would cause a project to stall, and it is a concentration measure, not a headcount measure — a project with hundreds of lifetime contributors can still have a bus factor of one [1].
- How concentrated projects actually are
The empirical distribution of bus factors is sobering. A study of 133 popular GitHub projects found that 46% had a bus factor of one and another 28% had a bus factor of two, meaning nearly three quarters depended on at most two people [1]. Separate empirical work on project abandonment treats the truck factor — the same concept under another name — as a measurable signal of knowledge concentration rather than an anecdote [2].
A more recent, larger-scale analysis reinforces the point from a network perspective. Drawing on roughly 25 years of Cloud Native Computing Foundation data, over 100,000 contributors and more than four million actions, it reports that the top 1% of contributors control about 40% of total influence, with a Gini coefficient near 0.85 — higher than the roughly 0.50–0.65 range typically cited for income inequality in highly unequal societies [3]. The distribution fits a power law (a Kolmogorov–Smirnov test gives p = 0.34, failing to reject the hypothesis), and the authors attribute the persistence to preferential attachment, not purely to merit [3].
A caution on provenance: that network study is a preprint and has not been peer reviewed, and its "influence" is a specific PageRank-derived construct. The exact Gini depends on the chosen metric, so the number should be read as evidence of strong concentration under a defensible method, not as a universal constant.
- Why concentration turns into burnout
Concentration becomes a failure mode when the people at the center are overloaded and under-resourced. In the 2024 Tidelift survey, 60% of maintainers reported being unpaid, nearly 60% said they had quit or considered quitting a project they maintained, and almost half described themselves as solo maintainers [4]. A peer-reviewed interview study of free and open-source maintainers finds that the work extends well beyond code into largely invisible interpersonal and organizational labor, with maintainers describing a "constant bombardment of issues and notifications" [5]. That workload is difficult to see in a commit graph, which is exactly why it is dangerous.
The recent shift toward AI-generated contributions has made the load heavier rather than lighter. The Linux Foundation's 2026 research reports that 59% of maintainers see growing contribution volume — rising to 69% in large projects — while 25% of incoming pull requests are judged low-quality "AI slop", climbing to 31% in large projects [6]. Notably, 71% of maintainers agree the tools lower barriers to entry, but 53% do not believe they have improved community trust, and only 35% of projects have a formal policy on generative AI use [6]. For high-complexity tasks, 21% of respondents report taking more time with the tools [6]. Lower barriers, in other words, can expand the review burden on precisely the small group already near capacity.
- The counter-argument worth taking seriously
Concentration is not purely a defect. Core-led projects often benefit from consistent design judgment, fast decisions, and a coherent quality bar; forcing broad ownership too early can dilute review standards and spread knowledge thinly across people who lack context. The high influence of long-tenured contributors partly reflects genuine accumulated expertise, and the network study itself finds that code-review activity predicts centrality more strongly than raw commit count (R² = 0.74) [3]. The goal is therefore not to level everyone down. It is to ensure that essential knowledge is held by enough people that no single departure is fatal, while preserving a strong core.
- A diagnostic and intervention framework
Signal | What it indicates | Intervention |
|---|
Bus factor of 1–2 | Single-point knowledge failure | Pair on critical subsystems; require co-review of core changes |
One person merging most PRs | Review bottleneck and concentration | Add and mentor maintainers; define merge rights by area |
Rising issue/PR volume with stable maintainer count | Load growing faster than capacity | Written contribution policy; contributor-run triage rotation |
Key decisions made in private channels | Knowledge not captured | Record decisions in durable, searchable places |
Core contributors show disengagement signals | Early burnout | Redistribute work before they leave; fund or staff the role |
The common thread is deliberate redundancy. Bus factor improves only when a second person genuinely understands an area, which means pairing, shared on-call or triage rotations, and writing down decisions that currently live in someone's head — not merely adding names to a file.
The bus-factor concept is well established but its algorithms differ across studies, and recent work is still trying to unify the definitions, so cross-project numbers should be compared with care. The quantified centrality findings come from either a non-peer-reviewed preprint or industry surveys whose underlying samples are partly unpublished. The strongest qualitative claim — that maintainers perform invisible, burnout-prone coordination work — is supported by peer-reviewed interview research but rests on self-reported experience rather than controlled measurement. Read the specific percentages as indicators of a consistent direction across independent sources, not as interchangeable precision.
Next steps: compute the bus factor for each critical area rather than for the project as a whole, identify the single-person subsystems, and assign a second maintainer to each through pairing and review. Track whether the core group's share of merges falls over time; if it does not, the redundancy is nominal rather than real.
References
- Beyond Code Contributions: Network Position, Temporal Bursts, and Code Review Activities in Large-Scale Open Source Ecosystems, arXiv:2602.06426 (preprint, not peer reviewed) — https://arxiv.org/pdf/2602.06426
Loading article