A Decade of Lessons from Apache Incubator Release Votes (2014–2025)

Introduction

This report summarises ten years of Apache Incubator release-vote activity to identify long-term patterns in community behaviour, process efficiency, and mentoring outcomes. It documents how the ASF’s culture of openness and collaboration translates into measurable improvement over time. The goal is transparency and shared learning, rather than evaluating individual communities or mentors.


Executive Summary

Over ten years of Apache Incubator release votes, it has evolved from rule enforcement to a culture of mentoring and shared accountability. Analysis of about 33,000 emails across 1,425 vote threads shows a clear pattern: fewer rejections, faster collaboration, and a stronger, more confident generation of ASF communities.

Key changes

The Incubator’s history demonstrates that mentoring, openness, and consensus are effective practices, showing how ASF culture strengthens through open collaboration.


1. Overview and Context

This analysis was conducted to understand how Incubator release practices have evolved and to identify any measurable signals of community health or mentorship engagement. It covers all public general@incubator.apache.org release-vote threads between January 2014 and September 2025, spanning votes on approximately 180 Podlings, including both now-graduated and retired projects.

During that period, approximately 33,000 emails, 1,400 vote threads, and 1,000 distinct individuals participated in release votes.

Oversight depended on a rotating core of 15–20 active IPMC members and mentors who contributed over two-thirds of all votes at any given time. Although individuals changed, this steady renewal kept institutional knowledge alive and ensured continuity across generations of Podlings.


2. Cultural and Process Evolution

2014–2017 – Policy-driven

Compliance language dominated: “This violates ASF policy.”
Mentors taught through correction, and threads ran long.

2018–2020 – Standardised

Checklists and reproducible builds became common.
Tone softened to “Please check …”, and re-votes declined.

2021–2025 – Collaborative

Mentors and Podlings worked together: “Let’s fix this before the next RC.”
−1 votes became more educational rather than blockers.


3. Mentorship and Continuity

About 200–240 active mentors participated directly in release-vote discussions, with a core of 20–25 carrying sustained load.

Even mentors who participate only occasionally play an essential role in ensuring Podlings reach the three binding +1 votes required for release approval. Since 2020, clearer expectations for engagement, along with the expansion of onboarding and best-practice resources, have strengthened participation and continuity.


4. Release Cadence and Review Efficiency

Today, most Podlings release regularly and with fewer RCs. About 60% of releases pass on the first RC, and mature Podlings average a new release every six months.

Most discussions reach consensus within two or three days. Formal vote closure remains at a minimum of 72 hours as required, with most threads reaching a clear outcome within that window. Pre-vote checklists and clearer expectations made the review faster and more constructive.


5. Oversight Quality and Issue Trends

Decline in negative votes

From 2016 to 2025, threads with at least one −1 fell to an average of about 9% of release-vote threads, with a high point near 15% in 2018 and a low point near 5% in 2023.
The decline reflects fewer procedural errors and more pre-vote preparation.

Nature of issues

The majority of early −1 votes were caused by missing or incorrect LICENSE and NOTICE attributions, unbuildable source packages, and artifacts staged outside the official distribution area. By the early 2020s, these issues had declined sharply as checklists, policies, and clearer documentation became standard practice. Problems involving signatures, checksums, and disclaimers have nearly disappeared, and source conting binary artifacts are now rare. More concerns today are legal or provenance-related, i.e. issues that require human judgment rather than mechanical checking.


6. Policy Reforms and Their Effects

YearReformImpact
2015–2017Incubator Release ChecklistStandardised expectations; procedural errors dropped sharply.
2019Distribution PolicyClarified “source-only”; most signing issues disappeared.
2020WIP Disclaimer UpdateEnabled early Podlings to release safely.
2020–2025Mentor Onboarding and Best-Practice MaterialsGradually expanded training resources; positive community feedback.

Each reform was followed by fewer recurring mistakes and smoother votes, suggesting that clearer documentation supports more consistent practice.


7. Release Rhythm and Graduation

Healthy incubation has a rhythm: regular releases, timely feedback, and visible improvement.


8. Governance and Values in Practice

Release oversight remains anchored in peer-to-peer, community-driven governance. Policy guidance flows laterally through mentors and community consensus rather than through formal intervention.


ASF ValueHow It Appears in Release Votes
Community over CodeBroad mentor and IPMC participation; success defined by collaboration, not perfection.
ConsensusFewer disputes and faster convergence to +1 votes.
Merit through ContributionPodling members increasingly initiate and close votes themselves.
TransparencyDiscussions remain public, modelling accountable governance.

9. Early-Warning Indicators for Struggling Podlings

IndicatorTypical ThresholdMeaning
No release for 9–12 monthsStrong predictor of declineCommunity slowing
Repeated unresolved −1 votesUnaddressed provenance or governance issuesNeeds mentor focus
No mentor reply within 3 daysSign of disengagementSupport gap
Fewer than three votersWeak community reachRisk of isolation

Tracking these signals enables the IPMC to intervene early, preventing momentum from being lost.


10. Methodology

Source: general@incubator.apache.org archives (Jan 2014 - Sept 2025).
Processing: Automated detection of release-vote threads and −1 issue classification.
Scope: The analysis covers only public release-vote threads on general@incubator.apache.org. Private mentoring, off-list collaboration or podlings dev lists is outside this dataset.
Limitations: Private mentoring is not visible; mentor-training effects are still emerging.
Threads were identified by subject lines containing [VOTE] and [RESULT] for the same release candidate.