Provides a statistical framework for the 'Bus Factor' and identifies non-code activity (reviews, meetings) and recency-bias as primary signals for current knowledge distribution.
The bus factor is identified as a critical metric for project resilience, with 63% of engineers reporting work on projects at high risk of stalling due to maintainer turnover. Analysis of 1,932 open-source projects showed that 16% faced the departure of all key engineers, and only 41% survived after such an event. Research finds that knowledge is generated through multiple modes beyond just code: commits (MRR 0.560), code reviews (0.403), and issue tracker activity (0.316) are primary signals. Crucially, the 'forgetting curve' means knowledge decays quickly, with a reported median halving-time of 4 months. The study introduces a multimodal algorithm that incorporates reviews and meetings, achieving improved accuracy in identifying key developers compared to traditional VCS-only methods which often overlook the knowledge contributions of senior technical advisors who commit less code.
16% of 1,932 open-source projects failed due to the departure of all key knowledge owners across their history, with only 41% surviving the transition.
sudden project stalling or death due to the loss of a single point of human failure
historically low contributor diversity is the definitive predictor of project existential risk, as projects that never attracted broad participation are most vulnerable to maintainer departure.
Standard algorithms focus solely on data from version control systems... knowledge is shared and created not only by writing code... medians halving time was 4 months.
Risk Guard evaluates technical signals but does not account for 'Knowledge Decay' over time or non-code contributions (like reviews or meetings) in its bus factor score.
Risk Guard would be better if it applied a 'Recency Bias' to contributor activity and factored in code review volume to more accurately estimate current knowledge distribution.