Summary

  • The proposed SLA replaces an old target that was met in only 42.5% of measured quarters since 2016 and introduces size-adjusted, rolling-year measures for the time under RPC control.
  • Its boundary is reasonable for holding the RFC Production Center accountable, but it does not by itself tell an author or reader how long the complete accepted-to-published journey takes.
  • Governance improves if the bounded SLA is paired with an attributable handoff ledger: every state change should show when it began and ended, who owns the next action, which clock is running, and why a correction was made.
  • The consultation remains open until 6 September 2026. The proposed targets and ratchets are not yet an adopted service commitment.

A green measure can coexist with a long journey

Service levels are supposed to convert a complicated operation into a promise that can be tested. The danger is that the promise becomes narrower than the experience it is taken to represent. In RFC publication, an editor receives a document after approval, performs substantive and copy editing, coordinates questions, works through references and registry actions, conducts final author review and prepares the canonical publication. Some elapsed time belongs to the RPC. Some belongs to authors, the IESG, IANA or another participant. An honest accountability system needs the boundary. A trustworthy public account also needs the whole trip.

The IETF Administration LLC’s consultation makes the first problem explicit. It says the existing agreement is no longer fit for purpose because the work has changed and because the old measurement has important gaps. Across the measured quarters since 2016, the old SLA was met only 42.5% of the time. That figure does not merely signal a service struggling against a target. It also signals that a target which is routinely missed eventually stops distinguishing an exceptional failure from ordinary operating conditions.

The historical measure could also misstate where time sat. Before a measurement change in the second quarter of 2025, a document could appear to remain on the RPC clock while the production center was waiting for an author. There was no separate state for waiting to be assigned to an editor. Portions of AUTH48—the final review phase in which authors confirm the edited text—could remain outside the measured on-clock time. A chart built from those states could therefore be arithmetically correct and operationally ambiguous.

The proposal answers that ambiguity by measuring the time the RPC controls. This is a sound accountability choice. A production center should not fail its contractual measure because an author takes weeks to answer a question, because the IESG must resolve a technical issue, or because an IANA action is outstanding. At the same time, a document author does not experience “RPC-controlled days.” The author experiences one calendar: acceptance, editing, questions, final review and publication.

That distinction matters because the current aggregate is long. The consultation estimates that a typical RFC of 30 to 50 pages now takes about 245 calendar days from receipt to publication. It attributes roughly 115 days of that condition to the existing backlog. Without the backlog, the estimate is about 130 days. It also reports that time outside the RPC’s control averages around 97 days and is broadly independent of page count. These numbers are modelled or aggregate views of overlapping service conditions; they should not be added together as though they were three separate blocks.

Page count changes the portion the RPC can influence. The consultation estimates about 0.34 additional day of RPC work for each page. That is why a single processing target would reward a queue rich in short documents and punish one carrying unusually long work. Size bands are a sensible correction. They are not, however, an explanation of why any individual document stood still.

What the replacement agreement would measure

The proposed arrangement uses a rolling year rather than judging one volatile quarter in isolation. It combines three kinds of signal: throughput, the maximum active backlog and size-adjusted processing percentiles. For its first-quarter baseline, 75% of documents would have to remain within RPC-controlled limits of 130 days for documents of 15 pages or fewer, 170 days for documents of 16 to 40 pages, and 190 days for documents above 40 pages.

The agreement would then tighten in two directions. The required share rises by one percentage point per quarter until it reaches 85%. After the first year, the maximum processing times fall by 5% each year. One ratchet asks the service to cover more documents; the other asks it to do so faster. This is materially stronger than publishing a static aspiration which becomes easier merely because work changes around it.

There is also a backlog objective. With editing resources and the scope of work assumed to remain unchanged, the proposal models an approximately 20% reduction in the backlog over three years and elimination over seven to twelve years. That forecast is a baseline, not a law of nature. A different arrival rate, a change in document complexity, staff turnover, a stream-specific surge or added review obligations can alter it. The consultation itself asks whether zero backlog warrants short-term investment and whether the community would accept less work, different priorities or other trade-offs to move faster.

The proposal therefore contains a real operating theory: preserve the current service and resource base, improve incrementally, use a measure the RPC can influence, and allow the backlog to decline over a long horizon. The governance question is whether the community wants only an enforceable boundary for the contractor or also a legible account of the publication system it collectively owns.

AUTH48 shows why ownership must travel with time

RFC 9280 describes the RFC Editor Model and separates policy, approval and production roles. RFC 8711 places the relevant administrative responsibility with the IETF Administration LLC. The authors’ publication guidance, queue documentation and working-group chair material explain a production process with multiple stops: initial checks, editing, coordination over questions, reference dependencies, IANA actions and AUTH48. These are not defects in themselves. Deliberate review protects the archival quality of the series.

AUTH48 is an especially useful example. The edited document is close to publication, but authors and designated reviewers must approve the final text. A delayed answer should not count as RPC underperformance. Yet making that waiting time disappear from the service narrative would be equally misleading. It remains a real delay, it may have a known owner, and it may be capable of escalation. The correct response is attribution, not omission.

The same logic applies when a document awaits editor assignment. A queue state can be fully under the service’s institutional responsibility even when no editor is actively working on it. If the model records only “active editing,” the queue looks cleaner by excluding the waiting room. The consultation’s recognition of the missing state is important because a named queue is governable in a way an unclassified pause is not.

An attributable handoff ledger would complement rather than weaken the SLA. For each document state, it would retain the entry time, exit time, responsible role, next actor, clock basis, stated reason, escalation status and any later correction. Public reporting could aggregate these fields without exposing confidential editorial exchanges. A document’s history could then produce at least three honest totals: RPC-controlled time, externally controlled time, and end-to-end calendar time.

Corrections require special care. When a state or owner is reclassified after the fact, the record should preserve what changed, when, and under whose authority. Otherwise a performance series can improve because historical attribution moved rather than because service improved. An audit trail does not forbid correction; it distinguishes correction from silent metric management.

Authority without mandate laundering

RFC 9280 assigns RFC Series policy development to the RFC Series Working Group and its approval function to the RFC Series Approval Board, while the RFC Production Center implements publication work. That RSWG-and-RSAB authority governs RFC Series policy; it does not absorb the separate authority by which IETF standards are developed and approved. RFC 8711 gives IETF Administration LLC administrative, operational and financial responsibility, not authority over the technical merits of those standards. These are separate powers, not stages in one chain of command.

The handoff ledger would not alter that allocation or grant any participant a new mandate. It would give each institution evidence that a document entered or left its segment, identify the next action and preserve later corrections. Visibility and attribution can cross an institutional boundary without the administrator, production center or metric acquiring the power that remains on the other side.

A measurement system for more than one audience

The RPC needs a measure it can manage. IETF Administration LLC needs a measure it can use for oversight and contracting. Authors need to know where a document is and who must act. Stream managers need to see whether a particular workflow generates avoidable waits. The community needs to decide whether resourcing and scope produce an acceptable publication outcome. Readers and implementers ultimately care when the stable RFC appears.

No single number serves all of those audiences. The proposed SLA is strongest when read as a contractor-control instrument. Trouble begins only if its result is presented as the complete condition of RFC publication. A green result on the RPC-controlled percentile can coexist with a growing author-controlled wait, an unresolved IANA dependency, a stalled external review or a long pre-assignment queue if those conditions sit outside or at the edge of the bounded measure.

The alternative is not to charge every delay to the RPC. That would make the measure unfair and would create an incentive for the production center to pressure authors or other bodies merely to protect its score. It could also conceal the value of careful review. The better design keeps responsibility granular: the party able to change a delay owns that segment, while the end-to-end total remains visible as a shared system outcome.

Individual responses already illustrate the range of community judgment. Mirja Kühlewind’s message asks about the meaning and usefulness of the proposed limits, while Acee Lindem’s response addresses the pace and practical expectations from another perspective. They are individual interventions, not IETF consensus. Their value is that they test the assumptions before a metric hardens into the vocabulary through which performance is discussed.

What the consultation still leaves open

The consultation asks whether the proposed speed is right, whether different RFC streams need different treatment, whether zero backlog should be purchased more quickly, how author satisfaction should be measured, whether quality is sufficiently protected, and whether the service should do less. Those questions should not be reduced to a vote for faster or slower publication. Each one identifies a control variable.

Different streams may carry different editorial complexity or urgency, but separate rules can also fragment accountability. A satisfaction score can expose friction, but it can reward speed at the expense of editorial challenge. A short investment can remove accumulated work, but without an arrival-rate and staffing model the queue can return. Reducing scope can improve measured speed while transferring work to volunteers or lowering archival quality. Every improvement therefore needs a declared cost and a named recipient of that cost.

Quality belongs beside time. The RFC Series exists to preserve a durable technical record, not to maximize document completions. The proposed consultation is right to ask about doing less because a time target can induce invisible scope reduction. Public reporting should show rework, material post-publication corrections, author disputes and other quality indicators separately from elapsed time. A fast passage through a weakened edit is not the same service.

The most useful decision before 6 September is thus not whether the RPC should have a bounded SLA—it should. It is whether adoption will include a second, non-punitive view of the complete journey and a durable state history. The first creates enforceable accountability. The second keeps the institution from confusing its contract boundary with the experience it is responsible for governing.

Sources

  1. Consultation on a Service Level Agreement for RFC Editing and Publication
  2. IETF Administration LLC announcement of the consultation
  3. RFC 9280: RFC Editor Model (Version 3)
  4. RFC 8711: Structure of the IETF Administrative Support Activity, Version 2.0
  5. RFC publication process for authors
  6. RFC Editor queue: how the publication queue works
  7. RFC Production Center process information for working-group chairs
  8. RFC Series discussion of publication timing and process states
  9. Mirja Kühlewind’s individual consultation response
  10. Acee Lindem’s individual consultation response