When the maintainers leave the room

Three communities that rarely share a mailing list have, in the same stretch of months, run into the same wall. The Nixpkgs core team dissolved itself after ten months, citing health and an unworkable load. The TLS working group chairs placed D. J. Bernstein under thirty days of moderated posting for repeated disruption under long-standing IETF process. On emacs-devel, a NonGNU ELPA submission for an LLM agent frontend turned into a multi-day tangle of free-software doctrine, bot-generated replies, and explicit requests to stop flooding the list. On Ethereum Magicians, an issuance EIP drew more than two hundred comments laced with personal attack and AI accusations before its authors pulled it from the next fork. The through-line is not ideology. It is governance under fatigue: high-stakes volunteer projects still lack durable models for who holds power, who absorbs conflict, and what happens when the people doing that work burn out.

The Nixpkgs announcement is the clearest statement of the cost. Writing for the departing core team, qyliss described pride in concrete work reforming committer onboarding, extending merge automation, re-establishing contact with GitHub after a security incident, and drafting an initial automation and AI policy. Then the diagnosis: the role "has sadly not turned out to be the lightweight role compatible with active technical contribution that we had originally hoped it would be," and "stepping down is necessary for our health." The team called continuation unsustainable, pointed to attrition, and noted that outreach had produced only one active applicant. The critique went upward as well as inward. Their experience, they wrote, was that the Steering Committee "as an institution lacks a native instinct for the delegation envisioned" when the core team was created. In the thread that followed, Philip Taron, tasked in loose terms with restarting something like the group, put the recruitment problem without romance: the job means "finding folks who are willing to be hard workers for zero money and high responsibilities and that are broadly trusted." It is "not open casting call, really." The names that previously held the role set a "High bar." Elsewhere, suggestions that an external dispute-resolution offer might help met the reply that "that ship has sailed long time ago." The technical substance of Nixpkgs (package delegation, merge bots, security advisories, AI policy) never left the frame; what failed was the human layer meant to keep that work from collapsing into private escalation.

The IETF case looks different on the surface and rhymes underneath. TLS chairs, after consulting the area director, announced that postings by D. J. Bernstein would be held for moderation for thirty days "due to continued disruptive behavior," following a string of public warnings. The action sits on BCP 9 and RFC 3934, with the chairs choosing moderation rather than a full suspension of posting rights. Released messages must be confirmed on-topic and not disruptive, with delay of up to two working days. Bernstein is not an anonymous troll; he is a cryptographer whose technical contributions and combative list style are both long-running facts of protocol work. The chairs' move is process applied to a famous participant, not exile of a nobody. Supporters of strict list discipline read it as overdue protection of the working group's attention. Critics of heavy moderation in standards bodies read any throttle on a known expert as a risk that difficult technical dissent gets reclassified as disruption. Either reading leaves the same structural point: the chairs, not an infinite volunteer pool, absorb the cost of deciding message by message what crosses the line, and they do it under rules written for an earlier traffic pattern.

Emacs-devel supplied a third variant of the same pressure. Thanos Apollo proposed hermes, an Emacs frontend for a free Hermes agent, with dashboards, streamed replies, approval prompts, and session tools. The thread immediately widened into GNU criteria: whether packaging the frontend effectively steers users toward nonfree services or SaaSS providers the agent surfaces, how far maintainers must chase dependency graphs before accepting a package, and what counts as a clear answer from a submitter. Richard Stallman argued that the project must not guide users toward the Hermes app if that path involved nonfree sales or platforms. Eli Zaretskii pressed for precision about what any cited page actually said, then repeatedly asked participants to stop mixing issues and flooding the list while he worked the question with Stallman. Jean Louis kept returning to whether hosting the package encourages SaaSS even when both hermes.el and the agent are free. Jacob S. Gordon, declining to litigate the package itself, drew a separate line: "I feel like a line was crossed in the latest LLM-automated exchange, and that our community would be worse off if that were normalized." The technical object (an ELPA package talking to a local or remote agent) became secondary to process load: who may speak, how much repetition is tolerable, and whether automated replies are a legitimate participation mode or an attack on the list's human bandwidth.

Ethereum Magicians made the failure mode loud. EIP-XXXX (discussed as EIP-8363), a tapered issuance burn that would partially burn validator rewards to reshape the ETH issuance curve, became one of the most commented proposals in forum history. Authors rooted the design in prior research on stake ratio, LST dynamics, and the claim that a very high staking ratio threatens security, neutrality, and ETH's role as money. Opponents called it a tax on engaged holders, a rewrite of the post-Merge deal, and a decentralization story that had not shown solo stakers would survive lower yields. The tone collapsed. Commenters accused one another of missing the question, of AI-assisted wall-of-text replies, and of "AI fuelled handwavy" abstraction divorced from how real stakers behave. Sock-puppet cleanup and personal attacks joined the technical dispute. The authors withdrew the EIP from consideration for the Hegotá fork, agreeing that a fork-scoping exercise was the wrong venue for issuance policy and promising a dedicated process. They stood by the motivation while listing open categories still unresolved: security effects of a lower stake ratio, industry impact on yield-based products, whether the curve is the right instrument, and who remains staking under the new equilibrium.

Across these threads the pattern is stable. Authority is either under-delegated (Nixpkgs core versus Steering Committee), exercised as scarce chair attention (TLS moderation), doctrinal and exhausting to apply (NonGNU and free-software boundary cases), or procedurally mismatched to the decision (issuance policy inside a fork CFI). Care work (triage, onboarding, warning problem users, reading dependency ethics into a package) is still treated as a side quest compatible with deep technical contribution until the people doing it say it is not. Conflict tools remain blunt: disband the team, moderate the poster, tell the thread to stop, withdraw the EIP. Money, where it appears at all, is insufficient to change the recruitment equation Philip Taron described.

Nothing here is resolved. Nixpkgs faces a Steering Committee election and an empty core-shaped hole with a trust bar few will clear for free. The TLS moderation window is time-bounded; the underlying dispute style is not. Hermes.el's acceptance path still turns on how GNU draws lines around agents and services that did not exist when the rules were drafted. Ethereum issuance has been promised its own process after a thread that demonstrated how quickly monetary politics burns the forum. The recurring crisis is not that open projects attract difficult people or hard topics. It is that they still design power and care as if burnout were a personal failure rather than a predictable output of the institutions they have.