Community Smells in Software Communities

• Community smells are sociotechnical anti-patterns that incur social debt by disrupting communication and collaboration in software teams.
• They are detected through methodologies like social network analysis and machine learning, using metrics such as centrality, churn rates, and co-occurrence patterns.
• Mitigation strategies include organizational restructuring, process interventions, and real-time dashboards to monitor and manage community health.

Community smells are sociotechnical anti-patterns manifesting as sub-optimal structures or recurring dysfunctional patterns in the social and organizational dynamics of software development communities. Analogous to code smells, which signal maintainable risks in source code, community smells denote sources of “social debt”: invisible coordination breakdowns, knowledge silos, communication inefficiencies, and forms of latent conflict that accumulate cost and degrade both teamwork and technical outcomes. Their identification and management are critical for sustaining software quality, team cohesion, and project longevity (Cabllero-Espinosa et al., 2022, Annunziata et al., 24 Apr 2025, Lambiase et al., 19 Feb 2026).

1. Formal Definition and Taxonomy

Community smells are defined as sets of organizational and social circumstances—emergent from interactions, structural decisions, and cultural patterns—whose persistence engenders stress, conflict, reduced team performance, and additional project costs, collectively termed “social debt” (Cabllero-Espinosa et al., 2022). Over 30 distinct community smells have been described; the most prominent include:

Smell Name	Taxonomic Group	Characteristic Description
Organizational Silo	Structural	Strong intra-group, weak inter-group links
Black Cloud	Communication	Unstructured or overloaded communication
Radio Silence/Bottleneck	Process/Gatekeeping	Single intermediary blocks information flow
Prima Donna	Authority	One/few resist changes, block collaboration
Sharing Villainy	Knowledge Flow	Poor, outdated, or unreliable dissemination
Solution Defiance	Conflict	Factional divides on technical direction
Truck Factor	Knowledge Risk	Critical dependence on very few contributors
Unhealthy Interaction	Social Dynamics	Low engagement, terse or slow exchanges
Toxic Communication	Conflict	Hostile, negative, or exclusionary language
Organizational Skirmish	Competence/Process	Clashing expertise or process misalignment

These patterns are codified with variants and extensions, reaching a full taxonomy of 30 unique smells characterized by emergent context, triggers, and process/product/people impacts (Cabllero-Espinosa et al., 2022, Annunziata et al., 24 Apr 2025, Lambiase et al., 19 Feb 2026).

2. Emergence, Properties, and Evolution

Community smells arise through diverse mechanisms, including contextual constraints (e.g., distributed teams, lack of onboarding), poor communication infrastructure, asymmetries in expertise distribution, weakened collaboration norms, and persistent unresolved conflicts (Cabllero-Espinosa et al., 2022). Their properties exhibit the following:

• Dynamic Progression: Smells can intensify or subside as team composition, policies, or workload shifts (“C_i(t)” as the time-varying intensity of smell i).
• Latent Effects: Smells may incubate before causing noticeable degradation in cooperation, coordination, or product quality (often crossing critical thresholds to propagate more widely).
• Interdependent Causality: Certain smells co-occur, exhibiting positive prevalence odds ratios (POR), indicative of “domino effects” (e.g., silos often co-occur with prima donnas or radio silence).
• Stage Evolution: The Community Smell Stages Framework models five sequential phases: induction, local emergence, team-level spreading, organizational spreading, and progressive impact on external stakeholders (Cabllero-Espinosa et al., 2022). Metrics such as prevalence P_i(t) and cumulative social debt D(t) = ∑_i ∫_0^t C_i(τ)dτ formalize this evolution.

3. Detection Methodologies and Metrics

Detection approaches operationalize socio-technical metrics derived from repository mining, communication artifacts, and social network analysis. Key methods include:

• Static and Dynamic Social Network Analysis: Construction of intra-team interaction graphs (collaboration and communication edges). Metrics include betweenness centrality (for bridge/bottleneck detection), closeness centrality, module contribution entropy (for silos), and churn rates (for assessing truck factor risks) (Gote et al., 2023, Lambiase et al., 19 Feb 2026).
• Machine-Learning-based Classification: Tools such as csDetector and CADOCS utilize >50 socio-technical features (e.g., node centrality, sentiment polarity, truck-factor indices) and threshold-based or learned classifiers (AUC ≈ 0.93, F1 ≈ 0.83 for CADOCS) to flag the presence of each smell (Annunziata et al., 24 Apr 2025, Cynthia et al., 18 Jun 2025).
• Temporal and Path-based Models: Higher-order network centralities (MOGen), which incorporate process memory, allow localization of smells within development workflows, outperforming static models in predicting influential/bottleneck nodes (Gote et al., 2023).
• Prevalence and Co-Occurrence Metrics:
  • Prevalence of smell X:

  $P(X) = \frac{|\text{repos with }X|}{N}$ - Prevalence Odds Ratio between A, B:

  $$\mathrm{POR}(A,B) = \frac{A_{\cap B}\times\overline{A}_{\cap\overline{B}}}{A_{\cap\overline{B}}\times\overline{A}_{\cap B}}$$

This models the pairwise statistical association among smells.

4. Empirical Prevalence, Interactions, and Impact

Surveys of open-source and ML-enabled projects reveal pronounced ubiquity and characteristic patterns:

• Prevalence: In ML-enabled systems, prima donnas, sharing villainy, and solution defiance dominate (e.g., PDE: 92.6%, SV: 83.5%), with truck factor, toxic communication, and organizational silos also elevated (Annunziata et al., 24 Apr 2025, Cynthia et al., 18 Jun 2025). Quantum software projects similarly exhibit high rates of authority and knowledge centralization (PDE, TF: 94%) (Lambiase et al., 19 Feb 2026).
• Co-Occurrence: Strong positive PORs (e.g., PDE–OSE = 4.34; RS–OSE = 3.19) indicate that certain smells cluster, amplifying mutual harm (Annunziata et al., 24 Apr 2025, Lambiase et al., 19 Feb 2026). Negative POR values highlight cases where the presence of one smell appears to mitigate another.
• Temporal Trends: Some smells fluctuate sharply over project life cycles (e.g., SV declines as projects mature, OSE and RS rise with scale or organizational complexity). Spikes in community smells and technical debt often precede major releases and are followed by corrective activity (Annunziata et al., 24 Apr 2025, Cynthia et al., 18 Jun 2025).
• Consequences: High occurrence of community smells robustly correlates with increased self-admitted technical debt (SATD), particularly in code, design, and requirements categories. Communication and authority-based smells are the strongest predictors of sustained or escalating technical debt (Cynthia et al., 18 Jun 2025, Cabllero-Espinosa et al., 2022).

5. Mitigation Strategies, Management Approaches, and Tools

Effective management of community smells involves both organizational structuring and process interventions:

• Organizational Strategies: Rotation of code/module ownership, flattening hierarchies, instituting liaison roles, routine knowledge-sharing sessions, and onboarding scaffolds directly target high-prevalence smells (Annunziata et al., 24 Apr 2025, Lambiase et al., 19 Feb 2026, Cabllero-Espinosa et al., 2022).
• Process and Governance Mechanisms: Scheduled retrospectives with explicit agenda for detecting and refactoring smells, assignment of explicit stewards, and integration of smell metrics into team dashboards support ongoing mitigation (Dantas et al., 25 Feb 2025, Bueno et al., 17 Jul 2025).
• Tool Support: csDetector, CADOCS, and CodeFace4Smells operationalize automated and semi-automated smell detection over sociotechnical data streams. Supplementary visualization (e.g., co-occurrence heatmaps) and hybrid dashboards enhance the early warning capabilities (Annunziata et al., 24 Apr 2025, Bueno et al., 17 Jul 2025, Cabllero-Espinosa et al., 2022).
• Formal Protocols: Communication plans, knowledge bases, and codes of conduct, as well as inclusive decision-making (RFC) procedures, undergird technical prevention of authority and conflict-based smells (Lambiase et al., 19 Feb 2026, Cabllero-Espinosa et al., 2022).

6. Research Directions and Cross-Domain Extensions

Research on community smells is rapidly expanding but still under-theorized in several domains:

• Scope Expansion: Most studies validate a subset of the 30 taxonomy smells, often focusing on ML, quantum, and large open-source systems. Systematic cross-cultural, cross-domain, and large-scale empirical validation is an open agenda (Lambiase et al., 19 Feb 2026, Cabllero-Espinosa et al., 2022).
• Coupling with Technical Smells and Debt: The causal linkage and evolutionary dynamics between community smells, code/architecture smells, and various technical debts (especially in ML and quantum projects) is an active area (Cynthia et al., 18 Jun 2025, Bueno et al., 17 Jul 2025).
• Model Refinement: Integrating multi-modal data (commits, issues, chats, code metrics) and refining dynamic accumulation models (e.g., D(t)=∑∫C_i(τ)dτ) are recommended directions (Cabllero-Espinosa et al., 2022, Gote et al., 2023).
• Socio-Technical Dashboarding: Real-time instrumentation of community health metrics into CI/CD pipelines and developer feedback loops is advocated to bridge detection and actionable prevention (Annunziata et al., 24 Apr 2025, Bueno et al., 17 Jul 2025, Cynthia et al., 18 Jun 2025).
• Impact Evaluation: Empirical Sankey diagrams show that ~45% of identified effects relate to breakdowns in cooperation, with communication factor effects also dominant (Cabllero-Espinosa et al., 2022). How these empirically translate to project outcomes and costs remains to be quantified at scale.

7. Broader Implications and Practical Recommendations

Community smells, by accumulating social debt, undermine the resilience and efficiency of software teams across domains. Early detection (via lightweight metrics and automated tools), targeted interventions (mentoring, liaisons, knowledge-sharing), and systematic governance (rotating ownership, structured onboarding, codes of conduct) constitute the core of evidence-based mitigation (Cabllero-Espinosa et al., 2022, Annunziata et al., 24 Apr 2025, Cynthia et al., 18 Jun 2025, Lambiase et al., 19 Feb 2026). Longitudinal monitoring and intervention effectiveness studies are recommended to evaluate remedial actions over time and tailor strategies to project size, domain, and organizational culture.

References:

• (Cabllero-Espinosa et al., 2022) Community Smells—The Sources of Social Debt: A Systematic Literature Review
• (Annunziata et al., 24 Apr 2025) How Do Communities of ML-Enabled Systems Smell?
• (Lambiase et al., 19 Feb 2026) Socio-Technical Well-Being of Quantum Software Communities
• (Bueno et al., 17 Jul 2025) Socio-Technical Smell Dynamics in Code Samples
• (Cynthia et al., 18 Jun 2025) How Do Community Smells Influence Self-Admitted Technical Debt in Machine Learning Projects?
• (Gote et al., 2023) Locating Community Smells in Software Development Processes Using Higher-Order Network Centralities
• (Dantas et al., 25 Feb 2025) The Role of the Retrospective Meetings in Detecting, Refactoring and Monitoring Community Smells