Summary
- Humming was documented in IETF working-group guidance by 1994 and remains useful because it is a quick, deliberately non-electoral signal. It can reveal whether a room leans, whether a question is understood and where a chair should continue discussion; it does not measure an authorised electorate.
- RFC 7282 says a hum should begin a conversation rather than end it. Volume cannot show whether a technical objection was answered, whether entities heard the same proposition, whether remote and absent experts were represented, or whether independent implementations support the chosen direction.
- Reproducibility should mean reconstructability, not a demand to recreate the sound. The final record should preserve the exact questions, sequence, chair's immediate reading, objections, reasons, meeting evidence, list confirmation, later revisions and remaining dissent so another competent reviewer can assess the consensus finding.
A ritual designed not to become a ballot
The IETF hum is memorable because it looks unlike the formal decision devices used by legislatures, companies and standards associations. A chair asks entities who support one direction to hum, then asks for the competing direction. The room produces two indistinct acoustic signals. No roll is called. No corporate delegation casts a weighted vote. The chair listens and decides what the signal means for the discussion.
That informality is not an accident. The IETF has no fixed voting membership. Anyone can participate as an individual technical contributor. Attendance changes by meeting and by issue. The same person may understand one draft deeply and another only generally. Counting every body in the room as an equal authorised voter would create false precision and invite recruitment for numerical advantage.
Humming resists that transformation. The sound is collective but not readily attributable. It gives a chair a sense of direction without generating a tally that entities can mistake for a binding majority. It can show that a proposition has substantial interest, that opposition exists, or that the room is too uncertain to proceed. It is fast enough to guide a live agenda.
The problem begins when the very lack of precision that protects rough consensus is used to shield the final judgment from scrutiny. “There was a strong hum” can become a conclusion whose factual basis no one outside the room can assess. A later entity may not know the exact question, whether both directions were tested, what uncertainty was expressed, or how the chair connected sound to technical issues.
The appropriate reform is not to eliminate the ritual or turn it into a vote. It is to separate the hum's exploratory function from the evidence required for a final consensus finding. The hum can remain informal. The decision must become reconstructable.
The documented practice reaches back to the early IETF
RFC 1603, published in 1994, described working groups as open communities operating through rough consensus. Its account of session management allowed consensus to be determined by a show of hands, humming or another method accepted by the group. The chair decided whether rough consensus had been reached.
The document placed that discretion inside a wider architecture. Working groups conducted much of their business by electronic mail. Sessions required minutes recording discussion and decisions. The standards ideal included multiple independent interoperable implementations and operational experience. Area Directors and the IESG provided oversight. Humming was one instrument within a sequence of development, review and testing.
RFC 2418 replaced RFC 1603 in 1998 and retained humming. It rejected 51 percent as rough consensus and warned that dominance did not mean volume or persistence. It required meeting decisions on new or materially changed questions to be reviewed on the mailing list. The chair's room judgment therefore never stood alone as the only available evidence.
This history matters because it rebuts two simplified stories. Humming was not invented as a scientifically calibrated measurement whose accuracy later deteriorated. Nor was it merely social theatre without procedural context. It was a low-resolution aid in an institution that expected open discussion, written review, independent implementation and appeal.
The period since the 1990s has changed the environment. Meetings are larger, hybrid and globally distributed. Remote tools can produce explicit response counts. Repositories hold much of the textual debate. Audio and video can be recorded. Those changes increase the ability to preserve evidence but also increase the temptation to mistake a tool output for consensus.
The durable question is not whether the sound is traditional. It is whether the institutional record still lets people outside the room understand why the chair proceeded.
RFC 7282 narrows the legitimate use of a hum
RFC 7282 gives the clearest modern account. Its central instruction is that humming should start a conversation, not end one. The chair should use the signal to identify what needs explanation, not treat the louder sound as a winning vote.
Rough consensus concerns issues. A group may proceed when all material objections have been addressed, even if some objectors remain unsatisfied. A small minority with a valid unresolved technical concern can prevent a consensus finding. Conversely, a small set of entities may carry the technically sound position when the larger room prefers an option that does not work.
A hum cannot communicate these distinctions. It conveys direction and perhaps intensity, but not the reason for either. A entity may hum against a proposal because of congestion, privacy, editorial clarity, commercial preference or incomplete understanding. Another may support it because code exists, because a deadline approaches, or because the alternative appears worse. Acoustic volume collapses every reason into one signal.
RFC 7282 therefore expects the chair to ask what lies behind the sound. If the hums are close, discussion is obviously needed. If one is much louder, discussion can still be needed because a quiet objector may hold the decisive technical issue. If nobody hums, the room may be confused, indifferent or unprepared. The chair's judgment is about the issues, not decibels.
This principle supplies the reproducibility standard. A later reviewer need not hear precisely what the chair heard. The reviewer must be able to inspect the proposition, objections, responses and technical evidence that justified the chair's conclusion. The sound can disappear once its informational content has been converted into reasons.
Reproducibility is not replaying the acoustics
Scientific reproducibility often means that another investigator can repeat a method and obtain a comparable result. A meeting hum cannot satisfy that literal model. The same people cannot be returned to the same state of knowledge. Room acoustics, microphone placement, remote delay, question order and social cues cannot be perfectly recreated. Even a high-quality recording changes the listener's position and may not capture relative sound accurately.
Governance needs a different form: decisional reconstructability. An informed reviewer should be able to recover what decision was being considered, what evidence was available, which objections remained, what the chair inferred from the hum, what discussion followed and why the final conclusion met the IETF's consensus standard.
This is similar to reproducibility in operational incident analysis. Investigators cannot rerun a global outage under identical conditions. They can preserve logs, configurations, timelines, observations and hypotheses well enough for another team to test the explanation. The record does not recreate the event; it makes the causal judgment reviewable.
For humming, the irreducible observation can be modest: the chair heard stronger support for one proposition, mixed signals, or limited public evidence response. The record should not convert that impression into invented numbers. It should surround the observation with facts that are reproducible: exact wording, draft version, entities present, remote access available, expressed objections, links to discussion, implementation evidence, and subsequent list confirmation.
A reproducible consensus finding is therefore not an objective acoustic measurement. It is a transparent exercise of expert discretion. Another competent chair might have weighted the evidence differently. The record should make that disagreement intelligible and appealable rather than pretending it cannot exist.
The wording of the question is part of the evidence
Small changes in a hum question can change the signal. “Should this become a working-group document?” differs from “Is this technical approach the group's preferred basis?” Adoption can mean willingness to work on a draft, confidence in its architecture, or rejection of competing drafts. Entities may hum according to different interpretations.
Compound questions are worse. A chair may ask whether the group supports a mechanism “with the proposed security changes” when those changes contain several contested elements. A entity who supports the mechanism but rejects one element has no accurate hum. The resulting sound cannot be mapped to one proposition.
Negative phrasing can create confusion: “Do you oppose not adopting?” Option order can prime the room. An initial strong hum may discourage a later dissenting sound. A request made immediately after a persuasive presentation measures a different state from one made after opponents respond. A question asked at the end of a session may capture departure schedules as much as technical judgment.
The minutes should preserve the exact or substantively exact wording of every consequential hum, the order of options and the document version under discussion. If the chair reframes the question after clarification, both formulations matter. The record should say whether an abstain or limited public evidence-information option was offered.
This requirement imposes little burden. Chairs already prepare slides and agendas. A displayed question can be copied into minutes. Remote tools can preserve prompts. The benefit is large: later readers can distinguish agreement on direction from approval of final text.
Reproducibility starts before the sound. A decision cannot be audited if the proposition itself is reconstructed only from memory.
The room is a selected sample, not the working group
An IETF meeting room is open, but it is not a representative sample in the statistical sense. Attendance reflects travel, registration, time zones, employer support, visa access, topic scheduling and personal workload. Entities may enter for one agenda item and leave before another. Remote entities can face latency or queue disadvantages. The people present are valuable contributors, not an electorate drawn from a known population.
This is why humming cannot establish a participation percentage or mandate. There is no stable denominator. The loudness of a hum depends on how many people choose to sound, how loudly they do so, where they sit and how the chair hears them. A hundred silent attendees are not necessarily abstentions; some may lack context or simply be listening.
The meeting record should nevertheless disclose the setting. Approximate attendance, remote participation availability, known audio failure, unusual schedule constraints and the stage of the agenda help a reviewer calibrate the observation. This is descriptive context, not a voter roll.
Selection becomes more important when the subject affects entities who rarely attend. A mechanism may impose operational burden on small networks, library maintainers or regional deployments absent from the room. Their absence does not give them a veto, but a strong hum among protocol authors cannot prove that operational concerns were considered.
Chairs should ask a separate coverage question: which expertise and deployment interests did the room contain, and which were missing? A follow-up request can seek the missing review on the list or through a relevant directorate. This converts demographic uncertainty into a technical evidence task.
The room can legitimately guide the work. It cannot authorise a claim that “the working group agreed” until the wider review surface has had a meaningful chance to test the result.
Remote humming reveals the measurement problem
The shift to online and hybrid meetings made the mechanics explicit. An acoustic room hum does not transmit cleanly through conferencing systems. Muted microphones, noise suppression, unequal latency and audio mixing destroy the collective signal. Digital tools therefore offer buttons, hand-raising or response categories.
Those tools can improve access and preservation. They can show how many connected entities responded, capture the exact prompt and include remote people who would otherwise be inaudible. They can also make the result look like a vote. A display of 35 to 30 invites numerical threshold reasoning even when the IETF's method says issues, not counts, decide.
The IESG's 2020 narrative minutes record this tension during discussion of virtual meeting tools. Entities considered whether numbers, anonymity, intensity and question design would preserve or distort the function of humming. One expressed concern was precisely that chairs should not convert a close count into a declaration of consensus.
An experimental virtual-hum requirements draft proposed downloadable summaries of questions and responses for meeting minutes. The proposal did not become a binding rule, but it illustrates the useful distinction: tools can preserve observations while RFC 7282 continues to govern their meaning.
Digital response data should therefore be retained as contextual evidence, not treated as the decision. The record may say that a stated number of connected entities selected each prompt, subject to tool limitations. The chair must still identify technical objections, determine whether they were addressed and confirm the result through the working group's broader channels.
Remote technology does not make humming reproducible by turning sound into numbers. It makes the inputs easier to preserve, which is valuable only if the institution preserves the issue-based judgment as well.
Intensity is informative and dangerous
A hum appears to carry intensity. A loud, immediate response feels different from a hesitant murmur. Chairs can use this information to judge whether a direction has energy, whether entities are confused or whether opposition deserves focused time.
Intensity is not strength of evidence. A entity may hum loudly because the issue affects a product, because they are confident, or because they enjoy the ritual. A quiet entity may hold packet traces showing a failure. Cultural comfort with public expression differs. Remote response tools may remove intensity altogether or replace it with a scale that falsely quantifies conviction.
The chair should record intensity descriptively only when it affected the next step. “The room response appeared strongly directional, but one technical objection prompted further discussion” is more useful than “overwhelming consensus.” The first statement separates the observation from the conclusion. The second transforms sound into authority.
Intensity can also help detect a legitimacy problem. If a strong room response collapses when the exact proposition is posted on the list, the earlier signal may have reflected wording, audience or momentary enthusiasm. The divergence is evidence requiring explanation, not an embarrassment to conceal.
The proper hierarchy is clear. Technical evidence determines whether an objection is valid. Reasoned discussion determines whether it is addressed. Intensity helps the chair manage time and decide where clarification is needed. It does not reverse that order.
This discipline preserves the human value of humming. Chairs are not required to ignore the room's mood. They are required to avoid presenting mood as a reproducible finding about protocol correctness or community authority.
A hum should trigger questions, not closure
After each material hum, the chair should announce what was heard and invite correction. If support appeared stronger, ask opponents to state the unresolved issue. If opposition appeared stronger, ask supporters whether a narrower proposition or new evidence answers the concern. If the signal was weak, ask whether people understood the question and read the relevant draft.
This immediate statement performs three functions. Entities can correct a misheard room. The minutes gain a contemporaneous interpretation rather than a later reconstruction. The chair exposes whether the hum is being used to guide discussion or to terminate it.
The strongest opposing argument should then be captured. RFC 7282 does not require every preference to be accommodated, but it does require objections to be addressed. A chair can summarise the issue and ask the objector whether the summary is fair. The group can examine evidence instead of debating the apparent volume.
Closure can follow when the answer is clear. The chair might conclude that the objection identifies no failure, that a revision resolves it, or that implementation evidence supports one tradeoff. The decision statement should include that reason. If the answer requires work, the hum has done its job by locating the next task.
This sequence reduces procedural theatre. Entities know that humming is not their only chance to influence the outcome. Quiet experts can speak after the signal. Chairs gain permission to move forward once an issue is answered rather than repeatedly sampling the room.
The practical cost is minutes, not months. A consequential hum followed by a two-minute reasoned exchange produces a better record than several unstructured hums whose meaning must later be debated.
Minutes need a minimum evidentiary grammar
Meeting minutes vary from transcripts to sparse bullet points. For a consequential consensus test, the record should contain a minimum set of fields even if the rest of the session is summarized lightly.
First, identify the draft version, section or concrete proposition. Second, preserve the questions and their order. Third, describe the room and remote response without pretending to precision the method did not have. Fourth, record the chair's immediate interpretation. Fifth, state the strongest material objection and the response. Sixth, identify the action: further discussion, revision, list confirmation, adoption call, last call or no conclusion.
If implementation evidence was mentioned, link it. A packet trace, interoperation report, code issue or deployment account is Network-resource evidence supporting the judgment. If evidence was disputed, the minutes should say so. A statement that “implementations exist” is less valuable than a link showing which behavior was tested against which revision.
Minutes should record dissent by issue rather than by dramatic quotation. Personal attribution can be useful where the contributor wants responsibility for a technical claim, but the governance record primarily needs the concern. A concise statement protects against the later assertion that the room was unanimous.
The chair should review the consensus section promptly, ideally before memories fade, and post it to the list. Entities can correct factual errors. Corrections should remain linked rather than silently rewriting the historical record.
This grammar is small enough for volunteer working groups. It makes a later consensus finding dramatically more reproducible without requiring full transcription or acoustic analysis.
Recordings help, but they cannot carry the judgment
Audio and video recordings can resolve disputes about wording, sequence and whether an objection was voiced. They preserve tone and presentation context that minutes omit. In a hybrid meeting, they may show whether a remote entity attempted to enter the queue.
Recordings are poor final decision records. They are time-consuming to review, difficult to search, sensitive to audio quality and fragile as the only link between a document and its rationale. A person evaluating one consensus call should not have to watch a two-hour session to discover the proposition and outcome.
Acoustic reproduction is also unreliable. Microphones compress sound, rooms have uneven pickup, and remote listeners receive a mixed channel. A recording can establish that humming occurred; it may not reproduce what the chair perceived from the front of the room.
The minutes should therefore cite the recording with a timestamp and extract the decisional facts. The recording is supporting evidence. The written statement is the authoritative explanation of what the chair concluded and why. If they conflict, the discrepancy should be corrected publicly.
Long-term access matters. Stable Datatracker records and mailing-list messages are more useful than links tied to a temporary meeting platform. The working group should preserve enough text that later maintenance does not depend on proprietary playback or a vanished format.
The existence of a recording must not excuse weak minutes. More data can reduce accountability when the institution transfers the burden of interpretation to every future reader. Reproducibility requires organised evidence, not merely stored media.
Mailing-list confirmation is a substantive second test
RFC 2418 requires review on the mailing list when a meeting reaches a decision on a new or materially changed issue. RFC 7282 reinforces that final decisions are expected on the list. Confirmation is not a ritual announcement that nobody must answer. It gives absent entities the proposition, rationale and opportunity to raise a material issue.
A useful confirmation message should state what the room hum suggested, not exaggerate it into final consensus. It should reproduce the exact decision proposed, link the relevant draft and minutes, summarise the principal objection and answer, identify implementation evidence, and set a reasonable period for response.
Silence on that message has greater meaning than silence after a bare “please entity.” Readers can see what they are confirming. Former entities can identify a missing assumption. Implementers can compare the current text with their code. Remote contributors can correct an incomplete account of the meeting.
If the list raises no new issue, the chair can publish a final conclusion that integrates room and list evidence. If a new objection appears, the chair must address it. The meeting hum is not invalidated; it becomes one part of a wider discussion. A serious flaw can properly reverse the room direction.
Substantive changes after confirmation require another check. An editor cannot resolve dissent by quietly changing text and relying on the earlier hum. The chair should decide whether the change implements the agreed answer or creates a new proposition. The record should link the revision and any follow-up call.
List confirmation makes consensus reproducible because it converts a local acoustic event into a written, globally inspectable claim. Its legitimacy depends on the quality of the statement and response, not merely on the fact that an email was sent.
Dissent must survive the louder sound
A humming test can erase minority evidence in two ways. Entities may decline to hum against an apparently dominant room. Minutes may describe only the chosen direction, leaving no trace of the reason opposition existed. Both failures conflict with issue-based consensus.
The chair should preserve material dissent even when the group proceeds. The record can state that an objector predicted a specific failure, that the group examined named evidence, and that the chair considered the issue answered for a stated reason. The objector need not agree with the conclusion.
This protects good-faith minority participation. It also constrains repetitive obstruction. Once the strongest form of the issue is recorded and answered, later repetition does not create additional weight. New evidence can reopen the question; volume alone cannot.
The dissent statement should distinguish feasibility, architecture, security, operations and preference. If a entity simply prefers option B, the group can choose A after understanding the tradeoff. If B's advocate shows that A cannot interoperate with deployed systems, the issue requires a technical answer. If the concern is that remote entities were excluded from the hum, the remedy is wider confirmation.
Where uncertainty remains, the document can be narrowed. A claimed universal mechanism may proceed as experimental. A risky feature may be optional. A deployment assumption can be stated explicitly. Preserving dissent allows the status and language to reflect evidence honestly.
Institutional legitimacy is not demonstrated by a room sounding unified. It is demonstrated when a decision can acknowledge a continuing objection and still explain why proceeding was responsible.
Independent reviewers make the chair's inference testable
The chair listens from inside the working group's history. That expertise is an advantage, but it can also make shared assumptions invisible. Independent review tests whether the proposition survives outside the people who formed it.
RFC 4858 expects document shepherds to assess review depth and breadth, key non-working-group review, specialist needs, controversy and implementation. These questions provide a natural check on an enthusiastic room hum. A shepherd can ask whether anyone independent interpreted the text and whether the relevant operational perspectives were present.
Independence has several forms. A reviewer unaffiliated with the authors may expose organizational assumptions. A second implementation written from the specification can expose ambiguity. An operator outside the dominant deployment environment can expose cost or failure. A security reviewer can test claims that ordinary discussion accepted.
The record should state what the independent reviewer examined and what changed. A name alone is weak evidence. A review that identifies no defect still matters when its scope is clear. If all implementations derive from one codebase or all reviewers share one design team, the limitation should be disclosed.
Independent review does not overrule the working group. It supplies evidence to the consensus caller. A technically strong review may validate the room direction, reveal a missing issue or show that apparent opposition rests on an obsolete draft.
The practical reproducibility test is whether a later reviewer can follow this chain: hum suggested direction; dissent identified issue; independent examination tested issue; revision or reason answered it; list confirmation found no remaining material objection. That is far stronger than a claim that the chair heard the room clearly.
Implementers convert preference into network evidence
Humming is especially vulnerable when entities compare appealing designs before implementation. The room can express a preference based on architecture, simplicity or familiarity. Independent code can reveal that the preferred option has ambiguous parsing, incompatible state, unsafe fallback or unacceptable resource cost.
Implementation evidence should be linked to the version and question under discussion. A prototype of an earlier design does not validate a late security change. Two wrappers around one library are not fully independent interpretations. A successful demonstration of the normal path does not test recovery.
Useful evidence includes interoperation matrices, packet captures, conformance results, issue reports, performance under realistic conditions and operational rollback. Negative findings belong in the record as much as success. The group may choose to proceed despite a limitation, but the hum should not conceal it.
Implementers do not receive extra votes. Their evidence receives weight because it bears on whether an objection is valid. This preserves open participation while respecting the IETF's engineering mission. A newcomer with a reproducible failure can change the direction of a room full of senior supporters.
The chair can use humming to ask a different question: who has read the draft, who intends to implement, or which option deserves prototyping? These are resource and interest probes, not consensus decisions. Clearly labelling them prevents later minutes from turning willingness to work into approval of technical content.
Network-resource evidence is the strongest antidote to acoustic confidence. The sound tells the chair where the room leans. Implementations show whether the network can follow.
The chair's reasoned statement is the decisive artifact
No collection of fields can remove judgment from rough consensus. The chair must decide whether objections were understood, whether evidence answered them and whether further delay would improve the result. Accountability therefore centres on a reasoned statement.
The statement should identify the decision and scope. It should describe the hum as one observation, not the result itself. It should list material objections, their disposition and remaining uncertainty. It should summarise independent review and implementation evidence. It should state how meeting and list input were integrated. It should identify any conflict or unusual participation limitation relevant to the judgment.
The level of detail can vary. A straightforward adoption call may need a paragraph. A controversial security mechanism may need an issue table. A close room signal followed by substantial revision needs a clear chronology. Proportionality keeps the obligation realistic.
The statement should avoid unsupported adjectives such as “overwhelming” unless the term is clearly descriptive of the room and not used as proof. “The first hum was audibly stronger” is an observation. “The group overwhelmingly approved” implies a constituency and ballot that did not exist.
Publishing reasons also helps the chair. Entities can correct a misunderstanding quickly. Area Directors and shepherds can assess the conclusion. Future chairs can see which issues were settled and why. Appeals can focus on evidence rather than recollection.
The decisive artifact is not the audio, response count or meeting slide. It is the chair's inspectable mapping from observations and technical issues to the consensus standard.
Appeals need a record they can actually review
IETF procedure allows escalation and appeal because chairs and steering bodies can err. An appeal cannot function well if the challenged decision is “the chair heard a strong hum.” The reviewer would have to defer completely or rerun an impossible acoustic event.
A reconstructable record changes the question. The appeal body can ask whether the proposition was clear, whether material evidence was omitted, whether an objection was accurately described, whether the group was given list confirmation, whether substantive changes followed without review, and whether the chair applied issue-based rather than numerical reasoning.
The appeal should not become a second vote. Nor should it replace technical judgment merely because another outcome was possible. Review can be deferential while still checking reasons and procedure. A chair who considered the relevant evidence and explained a plausible conclusion should not be reversed because the hum was close.
Effective remedies can be narrow. The appeal body may require a corrected record, a focused list call, an independent review or reconsideration of one issue. It need not discard years of work. Where a missing perspective could materially change deployment, a limited delay may protect the standard's credibility.
The public record also deters weak appeals. A entity cannot plausibly claim that dissent was ignored when the issue, answer and follow-up test are linked. Transparency supports closure as much as challenge.
Reproducibility therefore strengthens institutional authority. The chair remains empowered to decide, while the institution retains the ability to distinguish judgment from arbitrary assertion.
Capture can hide behind an apparently spontaneous room
Humming reduces some capture risks because it does not reward organized head-counting with an official vote total. It does not eliminate agenda power, attendance concentration, employer coordination or social pressure.
A dominant design team may frame the alternatives, schedule the hum after its presentation and fill the room with people already committed to implementation. A smaller competitor may lack travel support. Entities may hear a strong first response and suppress their own dissent. None of this requires bad faith; shared context can create the same effect.
The safeguards are procedural and evidentiary. Publish the question before the session where practical. Present the strongest alternative fairly. Invite conflicts and implementation interests to be visible. Take opposing and limited public evidence-information signals. Ask for reasons. Confirm the proposition on the list. Seek independent review when the room and authors substantially overlap.
Repeated humming can itself become a pressure device. A chair should not keep sampling until the preferred direction sounds stronger. If the signal is unclear, identify the unresolved issue and collect evidence. Record every consequential test rather than only the final one.
Remote participation can mitigate or worsen capture. It widens access but may give on-site speakers richer context and faster queue access. The record should note material technical failures or participation asymmetries and give the list a genuine opportunity to correct them.
Consensus capture is best detected by asking whether the decision would still have a defensible evidentiary basis if the loudest cluster were set aside. Independent review, implementation results and reasoned issue closure should remain.
Membership accountability in a community without members
The IETF's lack of formal membership is a reason not to vote, not a reason to ignore participation evidence. The institution should be precise about whose signal it observed. “The room leaned toward A” is accurate. “The IETF approved A by hum” is usually too broad before wider confirmation and formal progression.
Membership accountability here means that every entity can understand how voice becomes decision. The rules should not privilege people who know the ritual's hidden conventions. A newcomer should be told that humming is not a ballot, that reasons matter, and that an objection can be raised after the sound. Remote and absent entities should know where confirmation will occur.
Affiliations matter for context, even though contributions are individual. A room containing several people from one employer may still reach a technically correct conclusion. Disclosure helps the chair recognise missing implementation independence or concentrated consequence. It should not turn entities into instructed delegates.
Accountability also protects those who do not hum. Silence should not be described as consent by every attendee. The record can state approximate response and uncertainty without assigning positions to individuals who made no sound. Digital tools should avoid publishing identity unless the purpose and expectations are clear.
The institutional claim should expand in stages. The room produced a directional signal. The working group considered and confirmed an issue resolution. The shepherd found adequate review. The IESG approved progression. Each stage has its own evidence and authority. Compressing them into the hum inflates a ritual beyond its proper role.
A practical protocol for reproducible humming
Before the session, chairs should publish the decision sought, draft version, alternatives and known issues. They should identify whether the hum will test interest, choose a direction, assess readiness or locate uncertainty. Entities should have a reasonable chance to read the relevant text.
During the session, the question should be displayed or read precisely. Each meaningful alternative should receive a separate prompt, with an limited public evidence-information option where useful. The chair should avoid compound or confusing negatives. Remote entities should have an equivalent response path or the limitation should be stated.
Immediately afterward, the chair should describe what was perceived and ask for correction. Material objections should be elicited and summarised. The group should identify the evidence needed to answer them. If no decision is reached, that outcome should be recorded without stigma.
The minutes should preserve the questions, sequence, setting, chair's reading, objections, evidence and action. A recording timestamp and digital response summary can support the text. The chair should not invent a numerical estimate for an acoustic hum.
The list confirmation should present the proposed conclusion and rationale, not merely announce that humming occurred. It should invite absent technical evidence, link the exact revision and state the response period. Later substantive changes should receive proportionate reconfirmation.
The final consensus statement should integrate every venue, disclose remaining dissent and known coverage gaps, and explain why the issues were considered addressed. The shepherd should independently assess review depth, implementation and rough points. High-impact or disputed cases should preserve a concise issue ledger.
This protocol does not make every meeting bureaucratic. Most exploratory hums need only a note. The full record applies when a hum materially contributes to a claim that a working group adopted, rejected or progressed technical work.
What should never be claimed from a hum alone
A hum alone cannot establish that a fixed percentage supported a proposition. It cannot establish that silent attendees agreed. It cannot establish that people outside the room received notice. It cannot establish that the draft was read, implemented or operationally tested. It cannot establish that an objection was answered merely because the opposing sound was quiet.
It also cannot establish independence. Ten audible supporters may share one employer, codebase or design assumption. Nor can anonymity prove freedom from coordination. Humming obscures identity; it does not dissolve interests.
A hum cannot carry a conclusion across substantive revision. If the text changes the mechanism, threat model or operational requirement, the earlier sound addresses the earlier proposition. The group must assess whether new review is needed.
Finally, a hum cannot substitute for reasons in an appeal. The chair's personal sensory confidence is relevant but not dispositive. The institution must be able to show how the observed direction related to addressed technical issues.
These limits do not make humming weak. They define its strength. It is an efficient, low-stakes sensor for a chair in an open, fluid community. Sensors become dangerous when their output is interpreted beyond what they measure.
The strongest case for keeping the hum
Eliminating humming could produce worse governance. Formal electronic polls would invite thresholds and strategic recruitment. Shows of hands could expose entities to employer or peer pressure. Endless microphone discussion could allow persistence to masquerade as consensus. Chairs need a quick way to test the room.
Humming also signals cultural humility. The group knows it cannot identify a stable electorate or reduce technical judgment to arithmetic. The ritual reminds entities that the chair listens for rough direction while reasons remain decisive.
The practice can welcome uncertainty. A weak or divided sound tells the chair that the room is not ready. An limited public evidence-information prompt can reveal that too few people read the draft. A strong sound followed by a serious objection can demonstrate why consensus is more than popularity.
The hum's future therefore depends on refusing to overclaim it. If it remains an input to documented reasoning, its informality is an asset. If it becomes the unreviewable proof of final authority, its informality becomes a legitimacy gap.
The IETF does not need to choose between tradition and auditability. It can preserve the sound and improve the record. The combination is truer to rough consensus than either acoustic mystique or digital voting.
Conclusion
Humming has survived because it solves a real institutional problem. An open technical community has no defensible voter roll, yet a chair needs to know whether a room is converging, divided or confused. A collective sound offers directional information without pretending to count an electorate.
That information is limited public evidence as final evidence. Sound does not preserve the proposition, identify the technical reason, reveal missing expertise, answer dissent, prove implementation or include absent entities. A bare statement that “the hum was strong” asks every later reviewer to trust an irreproducible sensory judgment.
The remedy is decisional reconstructability. Preserve the exact questions and sequence. Record what the chair heard. State the material objections and responses. Link independent review and Network-resource evidence. Confirm the result on the mailing list. Trace later changes. Publish the reasoned conclusion and retain an appealable record.
This discipline keeps the chair's judgment rather than replacing it with arithmetic. It also keeps the minority issue visible rather than letting volume erase it. The hum can guide the moment; the record must justify the decision.