# The Second-Maintainer Playbook: Bus Factor of One to a Vetted Co-Maintainer

*Compute your absence factor per activity, rank 15 to 30 candidates, run a dated triage-review-release trial, and finish with a second maintainer and a factor of two.*

- Canonical URL: https://www.refolk.ai/guides/second-maintainer-playbook
- Pillar: Engineering and open source
- Format: Playbook
- Published: 2026-11-15
- Last reviewed: 2026-11-15
- Reading time: 21 min
- Keywords: how to find a co-maintainer for my open source project, recruit new maintainer open source package, reduce bus factor open source project, contributor ladder triager to maintainer timeline, seeking new maintainer issue template, vet new maintainer after xz backdoor

## Key takeaways

- Compute the contributor absence factor separately for commits, PR reviews, issue triage, and releases, because a commit CAF of 2 can hide a review and release CAF that is still 1.
- CHAOSS says smaller projects may feel comfortable with a contributor absence factor of 2, so treat 2 on reviews and releases as the floor for this job, not the target.
- Tenure gates do not stop a patient attacker: the xz Utils attacker took about 8 months to reach self-merge, double DevStream's 4-month ladder total.
- Only 127 US profiles in Refolk's index put 'open source maintainer' in their headline, so source candidates from repository activity and downstream dependents first.
- Grant release power last and through a staged publish the incumbent approves, then rebuild from the git tag and diff it against the published artifact before granting registry rights.
- OpenTelemetry Collector contrib bars any promotion that leaves more than 25% of maintainers with the same employer; in a two-person group, apply it as 'do not share an employer'.

This playbook is for the solo maintainer who is the only person able to review, merge, and release a project other people depend on. It is also for the engineering manager, DevRel lead, or sourcer helping them fix that. It starts from a measured contributor absence factor of one and runs through a ranked candidate list and a dated triage-to-review-to-release trial, with identity and employer checks at every gate. It ends with a named second maintainer and a re-measured factor of two or more. Expect this to take months: the three trial gates alone run four weeks, then one to three months, then one to two release cycles.

## What is the contributor absence factor, and how do I measure it?

The contributor absence factor (CAF) is the smallest number of contributors who together account for 50% of a project's contributions. A project where one person does most of the work scores 1. CHAOSS, which publishes the metric definition, notes that it was previously called the bus factor. It is the number you are trying to move when you set out to reduce the bus factor of an open source project.

To compute it for one activity type:

1. Pick a window. Use the last 12 months.
2. Count each person's contributions of that type inside the window.
3. Sort people by count, highest first.
4. Add the counts down the list until the running total reaches 50% of the overall total.
5. The number of people you added is the CAF.

### Why you need four numbers, not one

CHAOSS says to calculate CAF for different contribution types, such as commits, issues, comments, or documentation. For a second-maintainer search, compute four of them: commits, PR reviews, issue triage (labels, closes, reproductions), and releases. Drive-by contributors push the commit number up, while reviews and releases stay with you.

A commit CAF of 2 proves that two people wrote half the code. It misleads when the second "person" is a dependency-update bot or a single large one-off PR, and it says nothing about who can approve or ship. I have found no published comparison of CAF across commits, reviews, triage, and releases on the same project, so I can't tell you how far apart they usually sit. Your own four numbers are the evidence. Record them as four integers with the window dates. That is the "before" half of the deliverable.

### What number to aim for

The CHAOSS starter project health example shows a CAF of 5. Its project viability model says smaller projects may feel comfortable with 2. Treat 2 as the floor and not the target, because large consumers of your package will expect more. Solo maintenance is common: in rOpenSci's package collection, 36% of packages listed only one person. If you need to score concentration across many dependencies rather than fix one project, use the [maintainer concentration risk guide](https://www.refolk.ai/guides/scoring-maintainer-concentration-risk). This playbook is about fixing a single project.

> **Tip:** Count releases by who published
>
> For the release CAF, count who actually ran the publish, not who wrote the changelog. If only your registry account has ever published, your release CAF is 1 whatever the commit log says.

## How long should a contributor ladder take from triager to maintainer?

Published contributor ladders put one to three months at each stage before someone becomes a maintainer, and DevStream's version adds up to four months. Those minimums set the pace at which trust builds. They don't establish who someone is, and the xz Utils attacker waited out all of them.

| Case | Contributor stage | Reviewer stage | Total |
|---|---|---|---|
| CNCF template | TODO | TODO | unset |
| DevStream | 1 mo | 3 mo | 4 mo |
| HAMi | 3 mo | 10 reviews | 3 mo+ |
| xz (Jia Tan) | Oct 2021 to Jun 2022 | n/a | about 8 mo to self-merge |

Notes on reading the table:

- The CNCF contributor ladder template leaves its numbers as "[TODO: Number]". If you adopt it, choose the numbers deliberately.
- HAMi's reviewer stage also requires organization membership. Membership needs 5 accepted PRs, 5 PRs reviewed, or 3 issues resolved and closed, plus 2FA.
- OpenTelemetry's membership guide says "Reviewer of the codebase for at least 1 month". The excerpt I have doesn't show which role that applies to, so read the document before you cite it.

The xz case is the one to calibrate against. The JiaT75 account was created on 26 January 2021. Its first xz Utils contribution, a minor .editorconfig patch, came on 29 October 2021, and it began merging its own commits in June 2022. Sources disagree on its first release. One gives 13 December 2022, and another says the first solo release was 5.4.2 in March 2023. Either way, it took about 8 months to self-merge and more than a year to release, which is double DevStream's whole ladder.

> **Watch out:** Tenure is not trust
>
> A patient attacker waits out every ladder minimum. In this playbook, each promotion also requires identity corroborated outside the project, and the release gate requires an artifact checked against the git tag.

## Where do second-maintainer candidates come from?

Candidates come from four lists built from repository activity and your dependents. Those are repeat issue reproducers, reviewers on other people's PRs, non-maintainer committers, and engineers at your heaviest downstream dependents. Searching for people who call themselves maintainers only supplements the fourth list. It shouldn't be where you start.

| List | What it proves | When it misleads |
|---|---|---|
| Repeat issue reproducers | Runs your code for real; can isolate a bug | The same broken setup filed many times, or a new account pushing for change |
| Reviewers on others' PRs | Already doing the scarce work | Approvals with no comments |
| Non-maintainer committers | Can change code to your standard | One big PR, then silence |
| Engineers at heavy dependents | Employer has a reason to fund their time | Shares an employer with your dominant contributor |

The first three lists come from your own issue tracker and PR history, and you can build them in an afternoon. The fourth takes the most work and gives the most. An engineer whose employer depends on your package can often get paid time for maintenance, which is what keeps a second maintainer around after the novelty wears off. Building it means going from the dependents in the public GitHub graph, to the companies behind them, to the engineers at those companies, to their GitHub accounts. [Refolk](/) handles that chain from a plain-English request across GitHub, LinkedIn, and the open web, so the list takes hours to build.

### What a headline search finds

A headline search finds very few people. In Refolk's index of professional profiles, few people describe themselves as open source maintainers at all.

| Segment | Profiles | Top employer count in sample |
|---|---|---|
| United States, all | 127 | 1 |
| Germany, all | 49 | 2 (Cargoboard) |
| United States, Senior/Strategic | 50 | 1 |

**127 - US profiles in Refolk's index with "open source maintainer" in the headline**

Germany has 49, so most real candidates will not surface from a headline search and must be sourced from repository activity.

The US pool is 2.6 times Germany's, and Senior/Strategic profiles make up 39% of it. The useful property is how spread out the employers are. In the top-25 US sample, no employer appears more than once, and named employers include Microsoft, AWS, Okta, Dynatrace, Docker, and CNCF. That makes the headline pool a good place to find someone who won't create an employer-concentration problem. It's too small to be your main source.

#### From long list to second maintainer

| Stage | Figure | Note |
| --- | --- | --- |
| Sourced | 15 to 30 | four lists, about a week |
| Pre-checked | 5 | affiliation noted |
| Agreed to trial | 1 to 3 | after 2 to 4 weeks of outreach |
| Promoted | 1 or more | re-measured CAF of 2+ |

*A working search narrows 15 to 30 sourced names to a top five, then to one to three trial candidates, then to at least one promoted maintainer.*

I ran this search: `Senior engineers in Germany who maintain Python packaging or build tooling projects` - [see the full result list](https://www.refolk.ai/s/ymkjwsqn7z).

*Returns named engineers with their employers and public code activity, ready to check against your dependents list and your employer cap.*

## How do I rank and pre-check the shortlist?

Rank on activity type first, then tenure, then employer, and corroborate identity outside the project before anyone gets a role. The output is a top five, each with an employer affiliation written next to the name. It takes two to three days.

### Ranking order

1. **Activity type.** Rank people who review others' PRs highest, then repeat reproducers, then committers. You are filling the review and release gates, so rank the people already doing that work first.
2. **Tenure in the project.** Use your ladder minimums as a floor and not a verdict. Someone who already meets HAMi's bar of 3 months and 10 reviews is further along than a one-month contributor.
3. **Employer.** OpenTelemetry Collector contrib bars any promotion that leaves more than one-fourth (25%) of the maintainers with the same employer. In a two-person group, anyone with an employer would break the literal cap, so apply its intent instead: the second maintainer shouldn't share an employer with you or with your dominant contributor. Also compute the Elephant Factor, the minimum number of companies whose employees perform 50% of the commits. The [Elephant Factor guide](https://www.refolk.ai/guides/employer-concentration-elephant-factor) covers that calculation.

The employer cap cuts both ways. It blocks capture by one company, and it also limits how many maintainers one downstream company can supply. Recruit across dependents rather than deep inside one.

### Identity corroboration

Run these checks before the Triage role, and repeat the relevant ones at each later gate:

- **Account age against first contribution.** JiaT75 was created about nine months before its first xz patch. A young account isn't disqualifying. A young account with no activity anywhere else is a warning sign.
- **Activity outside your project.** Look for other repositories, reviews in other projects, and issues filed elsewhere.
- **Presence outside GitHub.** Look for an employer page, conference talks, or a professional profile that matches the GitHub identity in name, timeline, and work.
- **A live conversation on camera.** OpenTelemetry Collector contrib requires candidates to have attended at least one SIG meeting on camera.

Passing these checks shows there is a persistent, accountable person behind the account. The checks can still be fooled when every outside profile is thin, or when the profiles were all created around the same time as the GitHub account. Matching a GitHub handle to a professional identity is the slowest part of the pre-check. Refolk searches GitHub, LinkedIn, and the open web together, which makes it a lookup.

#### Activity depth against identity corroboration

Horizontal axis runs from Identity thin to Identity corroborated. Vertical axis runs from Shallow project activity to Deep project activity.

| Quadrant | What it means |
| --- | --- |
| Patient stranger | Keep at Triage until identity is corroborated outside the project |
| Trial now | Offer the trial first and move to direct outreach |
| Not yet | Thank them, keep them on the list, revisit at re-measure |
| Invite in | Ask for reproductions or reviews before any role change |

*Deep project activity without corroborated identity is the xz shape, so hold it at triage until identity checks out.*

## The second-maintainer procedure, step by step

The procedure has eight steps in order: baseline, source, rank, outreach, triage gate, review gate, release gate, and re-measure. Each has a duration and a condition that marks it done. Don't open a gate until the previous one is done.

#### Bus factor of one to a vetted co-maintainer

1. **Baseline the absence factor four ways** - Compute CAF separately for commits, PR reviews, issue triage, and releases over the last 12 months. Done is four integers with window dates, in about a day.
2. **Source four candidate lists** - Pull repeat issue reproducers, reviewers on others' PRs, non-maintainer committers, and engineers at your heaviest downstream dependents. Done is 15 to 30 named people, in about a week.
3. **Rank and pre-check identity and employer** - Rank on activity type, tenure, and employer, and corroborate identity through account age and presence outside GitHub. Done is a top five with affiliations noted, in 2 to 3 days.
4. **Run direct outreach and a public call** - Message the top five, pin a Seeking co-maintainer issue, and post to ecosystem channels. Done is 1 to 3 candidates agreed to a trial, within 2 to 4 weeks.
5. **Open the triage gate** - Grant the GitHub Triage role for 4 weeks and spot-audit what the candidate labels and closes. Done is real triage that passes your audit.
6. **Open the review gate** - Grant Write plus a CODEOWNERS entry; the candidate reviews and you merge. Done is at least 10 reviews over 1 to 3 months.
7. **Open the release gate** - The candidate cuts a release through trusted publishing as a staged publish you approve, then gets registry maintainer rights with 2FA. Done is a release with a provenance link and a tag that matches the tarball, over 1 to 2 release cycles.
8. **Re-measure and record the result** - Recompute CAF on all four activity types and update MAINTAINERS with names and affiliations. Done is CAF of at least 2 on reviews and releases.

### The trial calendar

| Stage | Duration | Gate passes when |
|---|---|---|
| Baseline, source, rank | about 2 weeks | Top 5 with affiliations |
| Outreach | 2 to 4 weeks | 1 to 3 agree to a trial |
| Triage gate | 4 weeks | Spot-audit passes |
| Review gate | 1 to 3 months | 10 reviews done |
| Release gate | 1 to 2 release cycles | Tag matches tarball |
| Re-measure | at promotion | CAF 2+ on reviews and releases |

Put the actual dates for each gate in the trial issue on the day the candidate agrees. When the gates are dated, a stalled trial is easy to see, and pressure to skip ahead is easier to refuse.

### Where sources disagree on order

rOpenSci runs the public call first. CNCF ladders assume nominations come from people already contributing. I run direct outreach to the top five and pin the public issue in the same week. The direct message reaches your strongest candidates before the public call draws strangers, and the public issue catches people your lists missed.

## What should I grant at each gate?

Grant the smallest role that lets the candidate prove the next skill. That means Triage first, then Write plus a CODEOWNERS entry for reviews. Release rights come only through a staged publish that the incumbent approves. Admin and registry ownership come last, if they come at all.

| Power | GitHub role | Registry control |
|---|---|---|
| Label/close issues | Triage | none |
| Push, Actions secrets | Write | none |
| Settings, no destructive actions | Maintain | none |
| Manage access, delete | Admin | npm org owner |
| Ship a release | Write (unverified) | npm stage-only plus incumbent's 2FA approval |

How to read it:

- **Triage** is for people who need to manage issues, discussions, and pull requests without write access. It is the safe first gate.
- **Write** can create, update, and delete GitHub Actions secrets. Before you open the review gate, list every secret the repository holds and decide whether the candidate should be able to reach each one.
- **Maintain** manages the repository without sensitive or destructive actions. **Admin** can delete the repository, transfer ownership, and manage access, so keep it with you until well after promotion.
- **The release row is unverified for GitHub.** Whether Write alone can ship a release depends on how your workflows are wired, so check your own release pipeline.
- **On npm,** the most secure trusted-publisher setup uses stage-only permissions: the pipeline can run a staged publish but not a final publish. The second human approves the staged publish, so a release can't ship on one person's say-so.
- **On PyPI,** 2FA is mandatory for every account. I haven't verified how its Owner and Maintainer roles split powers, so read PyPI's current documentation before granting either.

Trusted publishing, where the registry accepts releases from a named CI workflow instead of a stored token, is available on PyPI, RubyGems, npm, and crates.io. Registry rules change, so check each registry's documentation when you reach the release gate rather than relying on what you remember.

#### Checks at each gate

1. **Triage** - Grant role; spot-audit closes; check account age and outside presence
2. **Review** - Write plus CODEOWNERS; audit Actions secrets; confirm employer against the cap
3. **Release** - Staged publish you approve; rebuild from the tag and diff the tarball
4. **Promote** - Registry rights with 2FA; audit every security contact
5. **Re-measure** - CAF four ways; update MAINTAINERS with affiliations

*Every promotion pairs a new power with an identity, employer, or artifact check, so tenure alone never unlocks the next role.*

> **Rule:** Release rights come last and staged
>
> Never grant registry publish rights until the candidate has shipped a staged release you approved and the published artifact rebuilds identically from its git tag.

> Merge power can be reviewed after the fact. Release power reaches every downstream user before anyone looks.

## What to send: the direct message and the Seeking co-maintainer issue

Message your top five directly first, then pin a public Seeking co-maintainer issue and post it to your ecosystem's channels in the same week. The direct message reaches your best candidates, and the public call catches the people your lists missed.

**6 of 6 - rOpenSci packages that found new maintainers through its ecosystem call**

This is the only published record, and no comparable figure exists for a standalone repository's pinned issue.

Ecosystem channels have the one published success record. rOpenSci highlights packages that need maintainers in a newsletter that comes out every two weeks. Find the equivalent for your language or framework, such as a foundation newsletter, a community forum, or a packaging working group, and post there. Candidate response counts and time-to-add aren't published for any of these programs, so track your own.

**Direct message to a shortlisted candidate**

```
Subject: Would you co-maintain <project>?

Hi <name>,

I maintain <project> on my own. I noticed <your reproductions on #<n> / your reviews on #<n> / your merged PRs>, and <company> depends on <project> in <where>.

I'm looking for a second maintainer, and you're on my shortlist of five. The path is staged and dated:
- Four weeks with the Triage role
- Then reviews with Write access and a CODEOWNERS entry, aiming for 10 reviews
- Then a release you cut and I co-sign, before any registry rights change hands

If you're interested, or if your team could give you time for this, could we have a short video call? If not, no problem. I'd welcome the name of someone who might be.

<your name>
```

*Name the specific activity that put them on the list; a message that could go to anyone will be ignored.*

**Seeking co-maintainer issue (pin it)**

```
<project> has one person who can review, merge, and release. That is a risk for everyone who depends on it, including me.

Who I'm looking for
- Someone who uses <project> in real work
- Willing to start with triage, then reviews, then releases
- Able to give regular time, with employer time welcome

How the trial works
- Triage gate, 4 weeks: label, reproduce, and close issues. I spot-check.
- Review gate, 1 to 3 months: Write access and a CODEOWNERS entry. You review, I merge. Goal: 10 reviews.
- Release gate, 1 to 2 releases: you cut a release through trusted publishing and I approve the staged publish.
- Then: registry rights with 2FA, and your name and affiliation in MAINTAINERS.

What I will ask for
- A link to your presence outside GitHub
- Your current employer, so the maintainer list stays balanced
- A video call before the first role change

Interested? Comment here or email <address>. Nominations are welcome.
```

*Fill in the bracketed fields and keep the checks section; saying it publicly discourages pressure campaigns.*

## How does this go wrong? Failure modes and false positives

Most failed searches either measure the wrong thing, mistake tenure or pressure for trust, or let release and contact powers move without anyone noticing. These are the eight to check for, and each one has a check you can run.

1. **CAF computed on commits only.** *What it looks like:* the commit CAF reads 2 and you declare victory, while reviews and releases still sit at 1. *Check:* compute CAF for each activity type, as CHAOSS recommends, and require 2 or more on reviews and releases specifically.
2. **Tenure treated as trust.** *What it looks like:* a candidate clears every ladder minimum, so promotion feels earned. The xz attacker cleared every minimum in the ladder table above. *Check:* corroborate identity outside the project at each gate.
3. **Pressure mistaken for demand.** *What it looks like:* accounts you've never seen complain about slow releases and urge a handover. In xz, those accounts were Jigar Kumar, Dennis Ens, and the Jia Tan persona itself. *Check:* look at account age, and whether the accounts have any activity anywhere other than your project. Pressure from accounts with no other history is a reason to slow down.
4. **Tag and tarball divergence.** *What it looks like:* the release looks clean on GitHub, but the published artifact contains code that isn't in the repository. That is how the xz backdoor shipped. *Check:* rebuild from the git tag and diff the result against the published artifact before you grant registry rights.
5. **Quiet contact changes.** *What it looks like:* nothing visible changes. In xz, the primary contact email in Google's OSS-Fuzz was switched to the attacker's. *Check:* audit every security contact (fuzzing services, SECURITY.md, registry contacts) at every gate.
6. **Single-employer capture.** *What it looks like:* the best candidate works at the same company as your dominant contributor, so one employer now controls the project in practice. CNCF flags maintainer lists with external names but contributions 80% or more from one organization. *Check:* apply the 25% cap's intent and compute the Elephant Factor before promotion.
7. **Trusted-publishing downgrade.** *What it looks like:* someone removes trusted publishing from the package and goes back to a token that one person controls. *Check:* use registry settings that require an interactive 2FA challenge to edit trusted publishing configuration (npm now does), and review that configuration at every gate.
8. **Headline search as sourcing.** *What it looks like:* the search returns a tidy list of self-described maintainers, none of whom use your code. Only 127 US profiles in Refolk's index carry the label. *Check:* build the first three lists from repository activity before you search by headline.

The xz backdoor was found by Andres Freund, an outside engineer, after it had shipped. The gates in this playbook exist so that the second person on your project catches that kind of problem before release.

## After promotion: how do I re-measure and keep it current?

Recompute CAF for all four activity types over the same 12-month window you used for the baseline, and compare. Success means at least 2 on reviews and releases. Then write down who holds which power and who they work for, so the next audit can be done quickly.

Keep the result current:

- **Update MAINTAINERS** with both names, current affiliations, and the date of promotion. Affiliations change, so ask both maintainers to update theirs whenever they change jobs.
- **Adopt an emeritus rule.** OpenTelemetry Collector contrib allows moving an approver or maintainer to emeritus status if they don't respond within two months. Use that kind of rule to remove powers from people who have gone quiet, including yourself.
- **Re-run the baseline on a schedule.** A project can drift back to a CAF of 1 without anyone noticing as the second maintainer's attention fades. Recompute all four numbers at least as often as you cut major releases.
- **Keep the list warm.** Candidates from the "Invite in" and "Not yet" quadrants are your source for a third maintainer.

#### Before you call the second maintainer done

- [ ] CAF is recomputed for commits, reviews, triage, and releases, each recorded with window dates.
- [ ] Review CAF and release CAF are both 2 or more.
- [ ] The second maintainer shipped at least one release through trusted publishing as a staged publish you approved.
- [ ] That release was rebuilt from its git tag and diffed against the published artifact with no unexplained differences.
- [ ] Both maintainers' registry accounts require 2FA, and trusted publishing configuration needs an interactive 2FA challenge to change.
- [ ] Every security contact, including fuzzing services and SECURITY.md, lists the people you expect.
- [ ] The two maintainers do not share an employer, and the Elephant Factor is recorded.
- [ ] Identity corroboration notes for the new maintainer are saved where a future maintainer can find them.
- [ ] MAINTAINERS lists names, affiliations, promotion dates, and a written emeritus rule.

## Frequently asked questions

### How do I find a co-maintainer for my open source project?

Start with your own repository activity, not a public appeal. Pull four lists: people who repeatedly reproduce issues, people who review other contributors' PRs, non-maintainer committers, and engineers at the companies that depend on you most heavily. Rank them, message the top five directly, and pin a Seeking co-maintainer issue in the same week. rOpenSci's ecosystem call found new maintainers for six of six packages, so post to your ecosystem's channels as well.

### How long should it take to go from triager to maintainer?

Published ladders suggest about four months at minimum. DevStream uses 1 month as contributor and 3 months as reviewer, and HAMi requires 3 months as contributor plus 10 reviews. This playbook's gates run 4 weeks of triage, then 1 to 3 months of reviews, then 1 to 2 release cycles. Treat these as a pace for building trust, not as proof of it.

### What should a Seeking New Maintainer issue say?

It should say the risk plainly, describe the staged path, and list the checks you will run. Name the gates (triage, then reviews with Write access, then a co-signed release), the review target, and the fact that you will ask for an employer and a presence outside GitHub. Candidates who know the process up front self-select, and saying it publicly makes it harder for pressure accounts to rush you.

### How do I vet a new maintainer after the xz backdoor?

Corroborate identity outside the project, not just tenure inside it. Check account age against first contribution, activity in other repositories, a consistent professional presence, and hold a live call on camera. At the release gate, require a staged publish you approve, rebuild from the git tag, and diff it against the tarball. Audit security contacts at every gate, because xz's attacker quietly took over the OSS-Fuzz contact.

### Can my co-maintainer work at the same company as me?

Avoid it. OpenTelemetry Collector contrib blocks any promotion that leaves more than 25% of maintainers with one employer, and CNCF flags projects whose contributions come 80% or more from one organization. With only two maintainers the literal cap can't be met, so apply its intent: the second person should not share an employer with you or with your dominant contributor.

### What contributor absence factor is good enough?

Two is the floor, not the goal. CHAOSS notes that smaller projects may feel comfortable with a contributor absence factor of 2, and its starter health example shows 5. For this job, success means at least 2 on both reviews and releases, measured separately. A commit-only CAF of 2 doesn't count, because reviews and releases can still sit with one person.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/second-maintainer-playbook*
