10 years of the Incubator

What Has Changed in 10 Years (2015–2025) - A data-driven view of how incubation actually works today

This guide summarises ten years of Incubator data (2015–2025) covering:

• podling lifecycles and time in incubation
• graduation and retirement patterns
• podling population and intake
• releases and time to first release
• committer and PPMC additions
• governance activity and oversight
• system capacity and behavioural change

The goal is not to define rules, but to replace anecdotes with observable patterns so podlings and mentors can interpret their own progress in context.

1. Structural Change: A Smaller Incubator

One of the largest changes over the decade has nothing to do with processes or behaviour, it's that the Incubator itself is smaller.

• Mid-2010s: 45–63 active podlings
• 2023–2025: 28–32 active podlings

Intake has also declined:

• fewer proposals are entering
• far fewer long-running “stuck” podlings accumulate
• cleanup and retirement of old backlog is largely complete

This structural shift must be accounted for in every interpretation of the data. A smaller Incubator naturally produces fewer messages, fewer threads, and fewer total graduations.

2. Time in Incubation: Graduation, Retirement, and Cohort Age

2.1 Graduation timing

The long-term pattern:

• High-podling years (45–63 podlings):
  graduation medians ≈ 20–42 months

• Low-podling years (28–40 podlings):
  graduation medians ≈ 13–18 months

Thus, incubation appears faster today partly because:

• there are fewer podlings (structural), and
• modern podlings move more consistently and predictably (behavioural).

Both effects matter.

2.2 Cohort-age effects

Recent cohorts (2019–2024) have not had time to accumulate:

• long-tail graduates
• long-tail retirees

Retirement typically happens 40–60 months after entry. Most modern cohorts are too young to fully reflect their distribution.

2.3 Graduation vs retirement probability

For fully matured cohorts (2015–2020 entries):

• Graduation rate: ~70–80%
• Retirement rate: ~20–30%

This has remained stable. There is no evidence that the Incubator has become “easier” or “harder”.

3. How Podling Population Affects Speed

Across 2015–2025, podling population strongly correlates with graduation speed:

• more podlings → slower median graduation
• fewer podlings → faster median graduation

In other words:

System size influences incubation speed. - When the Incubator is smaller, everything moves faster.

This structural factor must be considered when comparing eras.

4. Release Behaviour: Earlier, More Consistent, More Successful

4.1 Release health

Over the decade:

• first releases succeed more often
• technical issues decline (LICENSE, NOTICE, DISCLAIMER, naming)
• retries happen faster
• few releases require re-votes
• escalation is rarer

These improvements reflect real behavioural change.

4.2 Time to first release

Across all podlings:

• Median: ~5 months
• Mean: ~8 months (a few very slow outliers inflate the mean)

By cohort:

• 2015–2018: first release ~6–7 months median
• 2019–2024: first release ~4–5 months median

More podlings attempt releases earlier.

5. The First Release Effect

The first release is the strongest predictor of long-term success.

Podlings that release early:

• add committers earlier
• add PPMC members earlier
• show governance activity earlier
• attract more contributors
• almost always graduate

Podlings that delay their first release:

• rarely add new people
• struggle to build governance
• often retire

Those that release early almost always graduate.

6. Community Growth: First Governance Addition (Committer or PPMC)

To measure real community growth independent of bootstrap activity. (Note that all additions in the first 60 days of incubation are excluded.)
This section uses the earliest of:

• the first committer addition, or
• the first PPMC addition

This produces one consistent signal across all cohorts, avoiding historical differences in how podlings recorded roles.

6.1 Decade-wide pattern (2015–2024)

Across all podlings entering between 2015 and 2024:

• Median time to first governance addition: ~214 days (~7.0 months)
• Mean: ~350 days
• Range: 61–1,744 days

The wide range reflects both fast-moving communities and long-delayed cases, especially among older cohorts.

6.2 Early cohorts (2015–2018)

For podlings entering 2015–2018:

• Median: ~319 days (~10.5 months)
• Mean: ~464 days
• In many cases, the earliest addition is recorded as a PPMC member rather than a committer due to historical differences in how roles are handled.

These cohorts show slower and more variable early growth.

6.3 Modern cohorts (2019–2024)

For podlings entering 2019–2024:

• Median: ~148 days (~4.9 months)
• Mean: ~229 days
• The earliest addition is more often recorded as a committer.

These cohorts show faster, more consistent early growth with far fewer long delays.

6.4 Overall trend

Across the decade:

• first governance additions occur significantly earlier
• variance has decreased
• extreme long delays are now rare
• modern podlings typically add their first non-bootstrap contributor or governance member within 5–6 months, compared with 10–11 months in the mid-2010s

This earlier, more consistent signal aligns with other decade-long changes, such as earlier first releases, fewer stalled podlings, and more predictable progression.

7. Governance Formation and Activity

Graduating podlings show:

• regular governance threads on dev@
• shared responsibility
• visible design/review activity
• consistent release discussion
• growth in committer and PPMC membership

Retiring podlings:

• go silent for long periods
• are dominated by 1–2 people
• fail to make releases
• show little visible decision-making

These behavioural patterns are consistent across the decade.

8. Podling-Normalised Governance Load on incubator-general

Thread-level data from incubator-general shows a drop in raw message volume and total threads, but this closely matches the decline in podling population.

When adjusted for podlings:

• governance threads per podling remain stable (~7–12)
• release votes per podling are stable for most of the decade (~4–4.5)
• but increase sharply in 2023–2024 (~6.1–6.3)
• incubator-general is increasingly used for formal votes (vote-share rising from ~70% to ~75–85%)

Interpretation:

• IPMC oversight per podling has not declined.
• The Incubator-general list has become a little more vote-focused
• The recent rise in release-vote threads per podling aligns with earlier and more frequent release attempts.

9. The Two Eras of Incubation

9.1 The Heavy Era (2014–2019)

• high podling population
• long median incubation times
• many long-running stalled podlings
• slower release progression
• slower committer and PPMC growth
• heavier IPMC workload
• more governance bottlenecks

9.2 The Fast Era (2021–2025)

• fewer podlings
• shorter medians
• earlier first releases
• faster committer and PPMC growth
• fewer stagnation cases
• fewer bottlenecks
• more predictable progression

Some improvements are structural (fewer podlings), but many are behavioural.

10. Early Warning Signs for Mentors

Across all data sets, the strongest practical indicators of trouble are:

• No first release within 6–12 months
• No committer additions after 12 months
• No PPMC additions over entire incubation
• Long silences on dev@
• Extreme concentration of work

Podlings with these patterns overwhelmingly resemble historical retirees.

Podlings with the opposite patterns overwhelmingly graduate.

10.1 Governance Normalised by Podling Count

Raw dev@ governance traffic rises sharply in 2016–2020 and then falls after 2021. However, this is almost entirely explained by changes in the number of active podlings. When we normalise by podling count, a different picture emerges.

Over the decade, podlings have consistently generated roughly 25–37 dev@ governance threads per year. The overall level of dev@ governance per podling has been remarkably stable, and there is no evidence of a long-term decline in oversight or activity.

Class-level behaviour shows clearer trends:

• DISCUSS threads per podling have declined moderately since 2016.
• VOTE threads per podling have increased slightly, mirroring the rise in earlier, more frequent podling releases.
• REPORT and OTHER threads per podling have declined, reflecting fewer problematic podlings and fewer governance escalations.

2021 shows a temporary dip in all classes, but activity rebounds immediately in 2022 and remains stable thereafter.

Overall, after adjusting for podling population, dev@ governance activity per podling is steady across the decade, with a shift toward more focused, vote-driven activity and fewer overhead or crisis-driven governance discussions.

10.2 Graduation and Retirement by Cohort

When we group podlings by the year they entered the Incubator, a clear pattern emerges. Older cohorts (2015–2017) show the full lifecycle: a mixture of graduates and retirees, with retirement rates typically around 20–40%, matching the long-term historical average.

Cohorts from 2018 onward show far fewer retirements, but this is a structural effect of cohort age rather than a behavioural change. Historically, retirement occurs 40–60 months after entry, while graduation usually occurs within 12–24 months. As a result:

• podlings that are going to graduate tend to do so early
• podlings that are going to retire tend to do so much later

This means that recent cohorts (2021–2024) have had enough time to generate graduates but not enough time to generate their retirements. The apparent drop in retirements and the apparently higher graduation rates of modern cohorts are both explained entirely by cohort age.

There is no evidence that modern podlings retire less often; we have not reached the point in the cohort lifecycle when retirements commonly occur.

The one real change visible in the cohort data is the shortened time to graduation. This reflects both structural factors (a smaller Incubator with fewer bottlenecks) and behavioural improvements (earlier releases, faster community growth, and clearer guidance).

In short:

• graduation is faster,
• retirement rates are stable,
• recent cohorts appear healthier only because they are younger,
• and no long-term shift in retirement behaviour has yet been detectable.

11. Summary: What Has Actually Changed

Compared with a decade ago, the Incubator today is:

• Smaller (structural)
• Faster (behavioural)
• Cleaner (fewer long-running stagnation cases)
• More predictable (clearer signals of success and failure)

Structural improvements

• fewer podlings → lower systemic load
• fewer extreme outliers

Behavioural improvements

• earlier releases
• earlier committer/PPMC additions
• improved documentation and mentoring practice
• more consistent governance

Cohort-age effects

• newer cohorts have not yet produced their late retirements
• apparent gaps are due to age, not softened standards

And the key operational lesson:

First release + first committer + first PPMC growth tell you more about a podling’s future than any single numeric threshold.

12. Cohort Trends in Committer and PPMC Growth (2015–2025)

Committer and PPMC additions are the clearest indicators of real community growth. Across the decade, these signals show strong and consistent patterns, and the differences between cohorts help explain why modern podlings progress more predictably than earlier ones.

To avoid bootstrap noise, all measurements exclude additions made within the first 60 days of incubation.

12.1 Long-term shift: earlier and more consistent growth

Across the full dataset, we observe a clear shift:

• 2015–2018 cohorts:

  • first committer ~15 months median
  • first PPMC ~14–18 months
  • many podlings add nobody for the first 12–18 months
  • higher variance between projects
  • delays strongly correlated with later retirement

• 2019–2024 cohorts:

  • first committer ~6 months median
  • first PPMC ~10–12 months
  • early additions now the norm
  • variance dramatically reduced
  • the “healthy pattern” emerges earlier and more consistently

Modern podlings are faster, more consistent, and less variable in their early growth.

12.2 Cohort-level patterns: what changed when

Breaking down by cohort shows the shift clearly:

2015–2017 cohorts (Heavy Era):

• Large mix of older backlog and high-intake podlings
• Many podlings entered “quiet early phases” lasting 12+ months
• Committer additions were late and highly uneven
• Stalled podlings distorted system behaviour
• PPMC additions often happened only near the end (or not at all)

2018–2020 cohorts (Transition):

• Cleanup of long-running podlings begins
• Intake slows
• Commits begin appearing earlier
• Some podlings follow the new pattern, others lag
• Variance declines, but not fully stabilised

2021–2024 cohorts (Fast Era):

• Early committer additions become the norm
• Podlings with no additions in first year immediately stand out as high risk
• First PPMC addition typically within 10–12 months
• Very few podlings show extended “quiet” periods
• Growth curves cluster tightly around a stable pattern

This marks the most consistent period of podling behaviour in the Incubator’s history.

12.3 Graduation vs retirement: growth trajectories diverge early

The difference between graduates and retirees is visible long before the outcome.

Graduating podlings:

• first committer ~9 months median
• first PPMC ~12–13 months
• multiple additions over time
• steady governance activity
• release activity triggers further growth
• growth continues until graduation

Retiring podlings:

• may add zero committers
• often add zero PPMC members
• if additions occur, they tend to be very late
• post-setup growth is minimal or absent
• no release → no new contributors → no governance growth

The divergence happens within the first 6–12 months, not at the end.

12.4 Earlier releases → earlier people

Cohort-level analysis reinforces the “first release effect”:

Podlings that make early releases reliably add new people sooner.

The behavioural chain observed across cohorts: First release → new contributors → first committer → first PPMC → stable governance → graduation

This pattern strengthened from 2019 onward and became dominant by 2021. Podlings delaying their first release often never add anyone.

12.5 Reduced variance: modern cohorts behave more predictably

A decade ago, podlings varied widely:

• some added people early
• some added nobody for 2–3 years
• some grew quickly
• others stagnated quietly

This unpredictability made mentor judgment and IPMC oversight more difficult.

Modern cohorts show tight clustering around the healthy pattern:

• committer additions around 6–10 months
• PPMC additions around 10–15 months
• release cycles start within the first year
• very few long-delayed trajectories

This reduced variance is one of the clearest improvements in incubation.

12.6 Structural vs behavioural components

The shift is partly structural and partly behavioural:

Structural:

• smaller Incubator
• fewer large, single-vendor-heavy podlings that stagnate
• fewer long-running legacy projects
• cleanup of the backlog

Behavioural:

• clearer documentation
• stronger mentor onboarding
• better community development guidance
• podlings making earlier releases
• consistent expectations communicated across cohorts

Both forces produce earlier growth and more consistent outcomes.

12.7 Implications for mentors

Cohort-level patterns sharpen mentor judgement:

• A podling with no committer additions by month 12 matches historical retirees.
• A podling with multiple additions within a year matches graduates.
• A podling that adds nobody after its first release is unusual and needs investigation.
• A podling with early committer additions before first release typically shows strong internal momentum.
• A podling with no PPMC additions after 18 months is falling behind modern norms.

Mentors no longer need to rely on intuition as cohort comparisons provide context.

12.8 Summary of cohort trends

Across 2015–2025:

• Committer additions moved from late (15+ months) to early (~6 months)
• PPMC additions moved from late (14–18 months) to mid-range (~10–12 months)
• Graduating and retiring podlings diverge early in behaviour
• Cohort variance drops sharply after 2020
• Modern cohorts consistently follow healthy patterns
• Structural cleanup + behavioural improvements drive the shift

Cohort-level growth trends show that the Incubator has become faster, more predictable, and significantly healthier over the past decade.

13. Technical Release Issue Improvements (2015–2025)

One of the most striking decade-long improvements is the sharp decline in technical issues found during release votes. This pattern is highly consistent across podlings, release attempts, and cohorts. The change is both quantitative (fewer issues) and qualitative (different kinds of issues).

Even though the number of release votes per podling has increased, the number of technical blockers has decreased. This is one of the clearest signs of improved podling behaviour and clearer expectations.

13.1 What counted as a “technical issue”?

From release-vote subject lines, mentor/IPMC commentary, and archived review notes, the typical technical issues included:

• missing or incorrect LICENSE files
• missing or inaccurate NOTICE files
• missing DISCLAIMER or incorrect incubating disclaimer
• wrong source-release naming (missing “incubating”)
• unclear or incomplete build instructions
• missing signature or checksum files
• non-reproducible build outputs
• bundled dependencies without the correct license text
• provenance concerns in included assets
• unapproved binary artifacts

These issues formed the bulk of “-1 (binding)” reviews in the mid-2010s.

13.2 Long-term decline in technical failures

Across 2015–2025, clear trends emerge:

2015–2018 (Heavy Era):

• High rate of technical failures
• Multiple release attempts often required
• Frequent need for re-votes
• Licensing mistakes extremely common
• Many podlings unclear about source vs binary releases
• Documentation gaps caused repeated errors
• Mentors spent more time correcting basics than guiding behaviour

2019–2021 (Transition):

• Noticeable improvement
• Clearer documentation begins to reduce repeated mistakes
• New cohorts show fewer license/NOTICE errors
• First-release failure rates drop
• More podlings succeed on the first or second attempt

2022–2025 (Fast Era):

• Technical blockers become rare
• First-release pass rates improve significantly
• Re-votes become uncommon
• Very few podlings forget “incubating” in source-release names
• Licensing errors become exceptions rather than norms
• Most issues that do appear are fixed quickly
• The “technical problem” category shrinks dramatically

The decline is not subtle, it is clear, consistent, and significant.

13.3 Why technical issues declined

Multiple systemic improvements contributed:

Better documentation

• Distribution policy consolidated
• Board-approved guidelines published (2019 onward)
• Clearer explanations of LICENSE, NOTICE, and DISCLAIMER
• Stronger examples and templates
• Updated release guides
• Clear language about Maven being supplementary rather than an official distribution

Increased mentor familiarity

• Mentors now handle more similar cases
• Best practices have become widely shared
• Less confusion across podlings about ASF norms
• “First release walkthroughs” have become more frequent

Better podling behaviour

• Earlier first releases give more time for learning
• Modern podlings expect to iterate
• Teams more aware of ASF policies before entering incubation
• Reduced intake of “dropped over the wall” code dumps lacking provenance

Fewer special-case projects

• Fewer podlings entering with complex licensing burdens
• Fewer large monolithic codebases inherited from non-ASF ecosystems

Technical issues declined both because podlings are better prepared and because mentors/IPMC reviewers are more consistent.

13.4 Re-votes, retries, and remediation

Technical retries now:

• occur rarely
• are usually resolved in a single follow-up
• rarely require a full second vote
• rarely trigger policy escalations
• rarely block graduation

This is a dramatic contrast to the mid-2010s, when multi-cycle re-votes were common.

Modern podlings not only fail less often but they also recover faster.

13.5 Relationship with earlier releases

This pattern is deeply tied to the “first release effect”:

• podlings that release early learn earlier
• technical issues get resolved sooner
• later releases are cleaner, smoother, and faster
• podlings that delay releases never reach the “clean release” stage

The rise in release-vote volume per podling corresponds directly with:

• fewer technical failures
• more releases overall
• more reliable progression toward graduation

13.6 Summary of technical release improvements

Across the decade:

• Technical issues have declined sharply.
• First-release pass rates have increased.
• Re-votes have decreased.
• Most modern release problems are non-technical (e.g., governance or community concerns).
• Mentors provide more consistent early guidance.
• Documentation is clearer.

Taken together, technical improvements in release practices are among the strongest indicators that incubation today is faster and more predictable than it was a decade ago.

14. Cohort-Age Effects: Numerical Examples

Many apparent improvements in the last 3–4 years can only be understood when we account for cohort age. The simple fact is that recent podlings have not been in incubation long enough to produce long-tail graduates or long-tail retirees.

This section gives hard numerical examples to clarify the effect.

14.1 Retirement typically occurs 40–60 months after entry

From 2015–2020 cohort data:

• median retirement time: ~45–55 months
• shortest retirements: ~32–36 months
• longest retirements: 60+ months

This means:

Any cohort younger than ~3.5–5 years cannot possibly show its true retirement rate.

14.2 Recent cohorts have not yet matured

As of 2025:

• 2022 cohort: ~3–4 years old
• 2023 cohort: ~2–3 years old
• 2024 cohort: ~1-2 years old
• 2025 cohort: < 1 year old

These cohorts are far younger than the retirement window.

Thus:

• They can show early graduates
• They cannot yet show their retirements

Any apparent reduction in retirements is therefore a cohort-age artifact, not a behavioural change.

14.3 Time-to-graduation also depends on cohort age

Graduation tends to cluster much earlier than retirement:

• median graduation time: ~13–18 months in modern cohorts
• long-tail graduates: ~24–30 months
• rare delayed graduates: 36+ months

This explains why:

• recent cohorts show many graduates
• but still show zero or near-zero retirees
• because retirement sits on a different timescale

The two signals operate on different clocks.

14.4 Example: comparing a mature cohort vs a young one

2016 cohort (mature):

• ~47 podlings
• ~70–80% graduates
• ~20–30% retirees
• full lifecycle visible

2023 cohort (young):

• ~30 podlings
• some early graduates already
• 0 retirements because not yet in the 40+ month window

The difference is time, not behaviour.

14.5 Numerical illustration: why “zero retirements since 2021” is not meaningful

This statement is superficially true but misleading:

• podlings entering 2021 would reach retirement window around 2025–2026
• podlings entering 2022: 2026–2027
• podlings entering 2023: 2027–2028

Thus:

Of course recent cohorts show no retirements as they are too young.

A “missing” retirement signal for these cohorts is an expected outcome, not evidence of improved survival rates.

14.6 Why cohort-age matters for interpretation

Without adjusting for cohort age, the following misconceptions appear:

• “retirements disappeared after 2021”
• “modern cohorts graduate more successfully”
• “the Incubator is easier now”
• “standards must be lower”

When corrected:

• old cohorts show full distributions
• young cohorts show partial distributions
• nothing indicates reduced scrutiny or weaker graduation criteria
• the observed differences are structural, not policy-driven

14.7 Summary of cohort-age effects

• Retirement takes 3.5–5 years, so young cohorts cannot yet retire.
• Graduation takes 1–2 years, so young cohorts show fast graduates.
• Mature cohorts show full lifecycle patterns; young ones do not.
• The apparent decline in retirements is entirely explained by cohort age.

When interpreted with cohort age in mind, the decade shows consistency in outcomes, with the only real shifts being in speed, behaviour, and structural load, not in standards.

15. Confirming Stable Oversight Across Both dev@ and general@ (2015–2025)

After normalising all metrics by podling count, a single conclusion is unavoidable: ASF oversight per podling has remained stable throughout the decade.

This finding holds across both communication layers used during incubation, the podling’s own dev@ list (local governance) and the Incubator’s general@ list (ASF-level oversight).

Because raw message volumes track podlings numbers almost perfectly, only normalised analysis reveals the actual pattern.

15.1 dev@ governance is stable once normalised

Across ten years, podlings generate:

• ~25–37 dev@ governance threads per podling per year

This range is consistent across:

• the Heavy Era (2015–2019)
• the transitional years (2020–2021)
• the Fast Era (2022–2025)

Even when raw dev@ traffic rises or falls with podling population, the governance footprint per podling stays within the same narrow band.

This indicates:

• podlings are doing the expected amount of governance work
• mentors are guiding internal governance appropriately
• ASF’s expectations for community visibility remain unchanged

15.2 general@ governance is also stable once normalised

Normalised incubator-general activity remains:

• ~7–12 governance threads per podling per year

Again, this pattern holds across all eras.

Raw counts decline in later years simply because there are fewer podlings. Normalising removes that structural bias and reveals a steady oversight workload.

This confirms:

• oversight per podling has not declined
• the IPMC continues to perform the same role
• the shift in discussion patterns does not reduce governance activity
• quieter raw traffic is a structural effect, not behavioural

15.3 The mix of governance changed, but the amount did not

Across both lists:

• DISCUSS threads ↓
• OTHER threads ↓
• REPORT threads ↓
• VOTE threads ↑ (due to earlier/more releases)

This indicates:

• fewer confused podlings
• fewer escalations
• fewer stuck projects
• less meta-governance
• more formal decision-making
• more release progression
• more stable governance formation on dev@

The change is in quality, not quantity.

15.4 Oversight across eras: Heavy → Transition → Fast

Across the decade:

Heavy Era (2015–2019)

• higher raw traffic
• more noise
• many process clarifications
• many escalations
• larger systemic load

Transition (2020–2021)

• cleanup of long-running podlings
• temporary global disruption
• mixed cohort behaviour

Fast Era (2022–2025)

• clearer governance
• earlier releases
• more predictable podling progression
• cleaner dev@ vs general@ separation
• fewer escalations
• quieter raw traffic, but stable normalised oversight

Through all of this:

The number of governance interactions per podling remained stable.

15.5 Why this matters for interpreting the decade

A common misinterpretation is:

• “general@ is quiet, so oversight must be weaker now.”

Normalised data shows the opposite:

• podlings receive the same level of oversight
• governance simply happens in more appropriate places
• dev@ governance is healthy and consistent
• general@ is focused on votes and escalations
• increased release cadence increases vote activity

Incubator oversight is not declining, it is becoming more aligned with ASF norms.

15.6 Oversight quality increased even though oversight quantity remained stable

Additional indicators confirm this:

• fewer long-running stalled podlings
• earlier intervention by mentors
• clearer graduation criteria
• improved mentor onboarding
• better documentation
• more predictable podling trajectories
• fewer technical release issues
• more consistent release reviews
• higher vote-share on general@
• shorter response cycles due to reduced systemic load

The quantity of oversight per podling stayed the same. The effectiveness of oversight increased.

15.7 Summary: oversight has not declined the system improved around it

Across both dev@ and general@:

• governance volume per podling is stable
• oversight roles are unchanged
• early-warning signals are still visible
• expectations are the same as a decade ago
• governance quality improved even as noise decreased
• release activity increased
• podling behaviour improved

Quieter raw traffic reflects fewer podlings and fewer problems, not disengagement or relaxed governance.

The Incubator today applies the same level of oversight per podling, but podlings progress through incubation more quickly, make releases earlier and more often, and show fewer long-running issues, resulting in a more predictable overall process.