Summary
- For the 2025–2026 cycle the IETF printed a 192-name pool in a fixed order, four public lottery draws, the canonical key string, every MD5 value against its divisor, the resulting positions and the literal
./pickinvocation — enough for a stranger to rebuild the ordering. - The rule that shapes who actually sits, no more than two volunteers sharing a primary affiliation, comes from BCP 10 and not from the 2004 selection RFC, and is applied to the computed ordering by the chair afterwards, in prose.
Start with the document a stranger can use. The results email of 11 July 2025, signed by that cycle's chair, Benno Overeinder, states that the RFC 3797 algorithm run against the pool and the lottery draws “produces the following list”, and then shows its work (results announcement). Four seed sets: NZ Lotto 20 28 32 36 38 39 24; US Powerball 5 9 25 28 69 5; Croatian Loto 7 2 4 9 12 16 23 31; EuroMillions 1 8 9 18 50 1 5. The canonical key string those numbers were flattened into, printed in full: 20.24.28.32.36.38.39./5.5.9.25.28.69./2.4.9.12.16.23.31./1.1.5.8.9.18.50./. An entropy figure of about 72.5 bits. A table pairing each MD5 value with its divisor — 192, 191, 190 and down — and the list position that came out. And the invocation itself: % ./pick -d -s 2025.seeds -n 2025.names -c 15.
That is not a summary of a draw. It is the draw, in a form a reader can execute. Give the same seed strings and the same ordered pool to a hash function and a remainder operation and the same fifteen positions fall out.
What the procedure asked for
The IETF's own overview page describes the committee as a chair, ten voting volunteers, four to five liaisons and an advisor, and says the ten voting members “are selected through a complex, random process detailed in RFC 3797” (overview). That RFC is Informational, dated June 2004, authored by D. Eastlake 3rd; it obsoletes RFC 2777 and began life as an individual submission rather than a working-group document (record).
Section 2.1 puts the volunteer pool with the selection administrator, passes the names through the IETF Secretariat to check eligibility, and requires the eligible list to be published early enough for disputes. Section 2.2 requires the exact algorithm, including public future sources of randomness, to be published before the randomness is revealed. Section 2.3 requires the resulting values plus a summary of the algorithm's execution to be announced so that anyone can verify both, requires more names to be drawn than are needed, as ordered alternates, and requires a complaint cut-off. Section 1 is candid about motive: the purpose is to protect “the selection administrator (currently, the appointed non-voting NomCom chair)” from suspicion of bias (RFC 3797).
The machinery in section 4 is plain enough to reimplement on a napkin: each source's numbers written as decimals, sorted ascending, concatenated and slash-suffixed; the source strings joined in a pre-specified order and ASCII-coded; an MD5 hash of that string with a two-byte counter — zeroes, then 0x0001, then 0x0002 — each read as one big number. Select the first by dividing the first number by the pool size and taking the remainder plus one; select the second by dividing the second number by the pool size minus one, with the first selectee removed; and so on down (algorithm). Section 5.1 keeps uncertain-eligibility people in the published pool, freezes the numbering, and provides that a selectee later found ineligible is skipped and the count continues. Section 5.2 resolves an ambiguous randomness source only by “a reasonable choice” of the selection administrator, on good faith (RFC 3797).
The cycle that printed its inputs
The 27 June 2025 announcement set the terms: “Volunteer selection will follow the process outlined in RFC 8713 and use the random selection method defined in RFC 3797”, with four lottery sources named by date and result URL, about 180 volunteers registered at that point, and registration closing on 30 June (27 June). A separate announcement on 3 July listed the fifteen positions the committee would fill — one IETF LLC Board seat on a three-year term, one IETF Trust seat, six IAB seats, seven IESG seats — with incumbents named (positions). Fifteen positions is not ten volunteers; the two numbers measure different things and the page does not join them.
The pool arrived on 8 July, with the sentence that makes the whole exercise possible: “The list is presented in the exact order that will be used for the RFC 3797 selection process” (pool). Then the draws, then the table.
The rule that lives elsewhere
In the text of RFC 3797 examined for this piece, no limit on organisational affiliation appears. The limit is in BCP 10. RFC 9389, April 2023: “There are restrictions to ensure that no more than two volunteers with the same primary affiliation are chosen”, attributing that two-organisation limit to section 4.17 of RFC 8713, and noting both that publishing names and affiliations lets the community review affiliation claims and that a challenge under that section can concern the qualification data rather than the randomisation (RFC 9389).
The 8 July announcement supplies the working detail: subsidiaries count as the same affiliation as the parent company; where a conflict is clear the chair disqualifies the selection and moves to the next eligible volunteer; where affiliation is ambiguous the person is kept provisionally and replaced only if the inquiry disqualifies them (pool). The 11 July email then reports the two things that happened after the table: Andrew Campling declined, David Schinazi was taken instead, and the chair applies the same substitution “when more than two candidates are from the same organisation” (results).
The immediate authority for the cap, section 4.17 of RFC 8713, was not obtained for this article. What is known here about it is what RFC 9389 says about it.
What the sentence covers
Here is the boundary, stated exactly. A reader holding these pages can re-derive the ordering. A reader holding these pages cannot audit the composition of the seated ten, because the pages reviewed do not print the affiliation determinations — which candidate was compared with which, on what evidence, and at what point in the ordering the substitution took hold. The declination and the substitution rule are reported; the determinations are not.
Nor does any page reviewed for this article name a certifier, witness or auditor of the draw distinct from the committee chair (results).
The seed that changed
One more detail belongs beside the arithmetic rather than in a footnote. The 27 June announcement named four lottery sources. On 1 July the chair replaced them: “I have adjusted the schedule so that all seed sources now come from after the contest period, which ends on 7 July” (1 July). The replacement draws all fell after the volunteer challenge window closed, which is what section 2.2 asks for in substance. The record documents the change and prints the reason. It does not reconcile the two sets of sources with each other.
Codification, still pending
Someone has noticed the soft joint. draft-eastlake-rfc3797bis-06, an individual Internet-Draft rather than an adopted standard, would formalise elimination categories for pool members later found ineligible or unavailable, freeze the numbering after a cut-off, and put hard deadlines on challenges, stating that a person later determined ineligible “can be skipped and subsequently selected persons used” (draft). It does not bind the committee. Its existence is the clearest available sign that the step the 2025 chair handled in a sentence is understood, inside the community, to need a table.
What a relying reader needs
Five things, and this record supplies four of them plainly. The two instruments and their dates, and which supplies which rule — RFC 3797 for the ordering, BCP 10 for the composition. The split between what is recomputable and what is reported — the pool, seeds, key string, hashes, divisors, positions and invocation on one side; the declination and affiliation substitution on the other. The provenance of the cap, which RFC 9389 gives second-hand. The absence of a named certifier apart from the chair. The fifth is the missing column.
This is a stronger record than most appointment bodies produce, and it is worth saying so precisely rather than warmly: the ordering is engineered to be checked, and it can be. The seating is not. Print the determinations, or encode the cap as a step in the tool with affiliation data as published input, and the same stranger who rebuilt the table can rebuild the committee. Until then the draw belongs to anyone with a terminal, and the ten seats belong to the chair's published account of what happened after it.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






