W3C’s Appeal Ballot Says ‘Approve’—Then Overturns the Decision

Summary

• Section 5.9 of the operative 18 August 2025 W3C Process asks, “Do you approve of the Decision?” and offers Approve, Reject and Abstain.
• The same section says an Advisory Committee appeal passes when Approve exceeds Reject by a participation-dependent ratio, then says a passing vote overturns the decision. Read literally, approval of the decision becomes the force that reverses it.
• The 2023 Process was internally coherent: Reject had to exceed Approve for the decision to be overturned. Pull request 901 replaced that rule with an Approve-weighted threshold ramp while leaving both the ballot question and overturning consequence in place.
• The ramp itself addresses a legitimate concern. Below 5% participation the vote fails; from 5% through 15% it needs more than a three-to-one margin; above 15% but below 20% it needs more than two-to-one; at 20% or more a simple excess suffices. Explicit abstentions count toward participation.
• The public rationale was to prevent momentous decisions on a slim majority at low turnout by adapting the Bylaws’ Requisite Member Vote. It did not announce that approval of an underlying decision should overturn that decision.
• No reviewed evidence establishes that W3C has run a ballot under this wording, misconfigured a ballot or miscounted a result. The finding is a conflict in the normative text and its version history.
• W3C should publish a ballot-polarity conformance record from the actual ballot configuration: exact proposition, semantic meaning of each option, eligible denominator, participation band, formula, boundary tests, aggregate result, consequence, Process version and correction history.

Four clauses, two possible decisions

Appeal rules are written for the moment when ordinary agreement has already failed. That is why a small semantic error matters more here than it would in a routine meeting note.

Section 5.9 of the current W3C Process creates an understandable sequence. A representative of a Member in Good Standing challenges an appealable decision within three weeks. The Team announces the appeal. At least 5% of the Advisory Committee must positively support holding it within one week. If that admission gate is met, the Advisory Committee votes.

The ballot question is explicit: “Do you approve of the Decision?” The listed responses are Approve, Reject and Abstain. Each Member, or group of related Members, has one vote, and the relevant electorate consists of Members in Good Standing.

The next clauses calculate participation. Explicit abstentions enter the participation numerator even though they enter neither side of the decisive ratio. Below 5% participation, the vote fails. From 5% through 15%, Approve must exceed three times Reject. Above 15% and below 20%, Approve must exceed twice Reject. At 20% or more, Approve merely has to exceed Reject.

Then comes the consequence: if the vote passes, the decision is overturned.

Those clauses cannot all retain their ordinary meaning. If the proposition is approval of the underlying decision, an Approve ballot supports keeping it. If the successful side overturns the decision, the decisive ballots should oppose it. The current text instead makes sufficient approval the condition for reversal.

This is not a philosophical complaint about whether the threshold is too high. It is an executable-state problem. One literal reading produces one outcome; the ordinary meaning of the ballot question produces its opposite.

Put numbers through the rule

The contradiction becomes harder to overlook when the prose is treated as a function.

Imagine a ballot with participation at or above 20%. Twelve representatives choose Approve and eight choose Reject. Under the current formula, the appeal vote passes because twelve is greater than eight. Under the next sentence, the decision is overturned. Yet the stated question tells us that more representatives approved the decision than rejected it.

Reverse the totals: eight Approve and twelve Reject. The current formula fails, so the decision remains in place. The electorate has answered the stated question against the decision, but the appeal loses.

The low-participation band creates the same inversion with a higher hurdle. Sixteen Approve and five Reject passes because sixteen exceeds fifteen. The decision is overturned by a bloc that, according to the form, approved it. Fifteen Approve and five Reject fails because “exceeds” is strict: equality to three times Reject is insufficient. The decision remains despite three quarters of decisive ballots answering Approve.

These are test vectors, not reports of real W3C votes. They isolate the semantics from every political dispute. They also reveal why explicit abstentions matter. An abstention changes participation and can move a ballot into a different band without changing either decisive count. The rule therefore has two inputs—turnout and preference—and each needs its own preserved denominator and arithmetic.

A threshold table can be mathematically exact while the proposition it evaluates remains backwards. Arithmetic does not repair the object of consent.

The 2023 rule had one polarity

The Process published on 3 November 2023 asked the same question and offered the same three responses. Its consequence, however, followed the words on the ballot. Votes to Reject had to exceed votes to Approve; when they did, the decision was overturned.

That design was simple. It did not demand extra protection against a thin majority at low participation. But it had a stable polarity:

• Approve meant approve the underlying decision.
• Reject meant reject the underlying decision.
• More Reject than Approve made the appeal succeed.
• A successful appeal overturned the decision.

The 2025 Process introduced a legitimate additional question: how much support should reversal require when few Members participate? The public version history shows where the answer entered.

Pull request 901 removed the two-line Reject-over-Approve rule and inserted the progressive bands. The inserted comparisons all make Approve the side that must clear the heightened threshold. The following sentence—passing overturns the decision—remained. So did the earlier question about approving the Decision.

That diff is unusually useful evidence. It locates the polarity break without requiring speculation about motive. It shows that the counted label changed while the semantic object and consequence did not.

The public issue and pull-request explanation describe a threshold policy, not an effort to entrench or reverse decisions. Their concern was that a momentous outcome should not turn on a slim majority when only a small share of the Membership participates. W3C’s announcement of the 2025 Process likewise describes higher thresholds for low-participation appeal votes. None of those materials announces that “Approve” has been reassigned from the decision to the appeal.

The responsible conclusion is therefore narrow: the operative rule contains an internal conflict produced by a visible textual change. The record does not support an allegation of intentional reversal.

A sound threshold was attached to the wrong object

The progressive ramp comes from a coherent governance model. Article III, section 11 of the W3C Bylaws defines the Requisite Member Vote for affirmative corporate action. When no more than 15% of Members are present or represented, the affirmative vote must reach at least 75%. Above 15% but below 20%, it must reach at least two thirds. At 20% or more, a majority is enough.

The principle is defensible. A small participating minority should need a stronger internal consensus before authorizing consequential action. As participation rises, the supermajority requirement can relax. This guards both action and legitimacy without treating a non-voter as an opposing vote.

But the Bylaws’ affirmative vote approves the corporate action placed before Members. The affirmative label and the legal consequence point in the same direction.

An appeal ballot has two possible propositions. It can ask whether Members approve the underlying decision. In that form, Reject is the side seeking reversal. Or it can ask whether Members approve the appeal or an overturning motion. In that form, Approve is the side seeking reversal.

Either form can use a progressive threshold. What it cannot do is borrow the affirmative arithmetic from the second form while retaining the question from the first.

This distinction also explains why percentages and ratios must not be casually treated as interchangeable. The Bylaws state shares of affirmative votes among votes cast for and against. The Process expresses strict comparisons—more than three times, more than twice, more than the opposite side—while explicit abstentions affect the participation band. At the boundaries, “at least 75%” and “Approve exceeds three times Reject” are not identical: 15–5 is exactly 75% but does not exceed three times. A conformance record should expose the chosen expression rather than assume readers will infer it from a source model.

The strongest defence does not remove the defect

W3C has a strong contextual defence. Everyone involved knows this is an appeal. The initiating representative wants an underlying decision reconsidered. The earlier Process counted rejection as the overturning side. The public rationale discusses making reversal harder at low turnout. A careful administrator could phrase an actual ballot as approval of the appeal, label the choices to remove ambiguity or configure the decisive side according to institutional intent.

The source is also public, versioned and correctable. The current Editor’s Draft repeats the wording, so there is a visible place to repair it. No evidence reviewed for this Article establishes that W3C has actually administered a 2025-format ballot, harmed a party or relied on an inverted result.

All of that limits the claim. None of it makes the normative text executable as written.

Institutional memory is not a decision rule. The person setting up the form should not have to decide whether “Approve” approves the decision, approves the appeal or approves overturning. Nor should the electorate have to infer that the visible noun changes halfway through the section. An operational ballot cannot silently resolve the conflict and then become the only evidence of what the Process supposedly meant.

The need for precision is greatest in the hard case. If the appeal is politically sensitive, the result close or the participation count near a band boundary, every losing side has an incentive to prefer a different reading. A rule that works only when no one contests its meaning has failed at the moment it was designed to govern.

The support gate is not the merits vote

The Process contains another percentage that should remain separate. Before a ballot occurs, at least 5% of the Advisory Committee must positively support the appeal request. That is an admission gate. It establishes enough backing to put the matter to the electorate; it does not decide whether the underlying decision should be overturned.

The distinction matters for authority. The initiator invokes review. Supporting representatives authorize a ballot. The full eligible electorate decides the merits under the participation ramp. The Team administers those stages. The original decision maker and people relying on the decision then face the consequence.

Combining the support list with the later vote would blur those roles. The public guide helps an initiator identify the decision and state reasons, but it does not override section 5.9 or settle ballot polarity. The archive of support statements is Member-only. That confidentiality choice is compatible with a public account of the aggregate result.

No individual supporter or voter needs to be exposed for W3C to publish the denominator, participation total, explicit-abstention count, decisive counts, applied formula and resulting state of the decision. Privacy around individual preference is not a reason for opacity around institutional authority.

Publish the ballot configuration before voting begins

W3C should create a ballot-polarity conformance record for every Advisory Committee appeal. It should be generated from the same configuration that creates the ballot, before the ballot opens. A hand-written explanation after the outcome would be weaker because it could drift from the form representatives actually saw.

The record should include:

• the underlying decision identifier, title, deciding authority, announcement time and authoritative URL;
• an appeal identifier, the initiator’s eligibility state and the aggregate result of the 5% support gate;
• the controlling Process version and exact section;
• the exact ballot question, without paraphrase;
• every response label and the object it acts on: the decision, the appeal or an overturning motion;
• an explicit mapping of which response affirms the decision and which can overturn it;
• the number of eligible Members in Good Standing and the cutoff at which that denominator was fixed;
• total participation and the number of explicit abstentions;
• the applicable participation band, decisive formula and treatment of equality;
• machine-checkable pass and fail examples at each boundary;
• final aggregate counts, calculation and ballot result;
• the resulting state of the underlying decision: affirmed, overturned or unresolved;
• any correction, superseding text or interim interpretation, with effective times;
• any subsequent action that relied on the result; and
• a privacy statement identifying which individual records remain restricted.

The record is not a new veto point. It does not allow staff to decide the merits or replace the Process with metadata. It is a conformance object: the proposition, implementation and consequence must agree before authority travels through them.

Publishing it before voting also creates a clean objection window. A representative who sees that Approve has been mapped to overturning despite a question about approving the decision can challenge the configuration before ballots are cast. Correction at that stage is cheaper than deciding later whether a formally completed vote expressed the opposite preference.

Two coherent repairs

There are at least two straightforward drafting routes.

The first keeps the current question: “Do you approve of the Decision?” Approve then affirms the decision. Reject is the overturning side, so the progressive threshold must be expressed as Reject exceeding the applicable multiple of Approve. A passing appeal overturns the decision.

The second changes the proposition. The ballot can ask whether representatives approve the appeal or approve an explicit motion to overturn the named decision. Approve then supports reversal, so the current Approve-weighted ramp can remain. A passing motion overturns the decision.

This Article does not choose the legal wording for W3C. Each route may interact differently with W3C’s defined terms, ballot tooling and expectations around an “appeal vote.” The governance requirement is simpler: the noun after approve, the side counted for passage and the state transition after passage must point in one direction.

The repair should also state the boundaries unambiguously. Does the 5%–15% band include both endpoints? Is equality at exactly three-to-one insufficient? How are related Members deduplicated? When is Good Standing measured? What happens if eligibility changes while the ballot is open? The current text answers some of these questions, but the conformance record should make the actual calculation reproducible for the specific vote.

What the evidence does—and does not—show

At the 31 August 2026 cutoff, the 18 August 2025 Process is the latest published operative version identified by W3C. The current Editor’s Draft carries the same polarity conflict. The 2023 Process, public pull request, first threshold commit, issue discussion, W3C announcement and Bylaws establish the lineage described here.

They do not establish an actual miscount. The reviewed public materials do not prove that an Advisory Committee appeal ballot has been opened under the 2025 wording, that an administrator chose one interpretation, that voters were confused, or that any underlying decision was wrongly affirmed or overturned.

Nor does a GitHub pull request alone change W3C’s Process. The Membership approved the 2025 Process, and W3C announced its effective date. The source history is evidence of how the wording evolved, not a substitute source of institutional authority.

That boundary matters because governance analysis can become sensational precisely where documentation is incomplete. The warranted finding is already consequential: the public normative text gives the same response label incompatible functions. A correction is justified without inventing a victim or motive.

Participation is not a substitute for a legible principal

Heng Lu’s critique of multi-stakeholder governance draws a useful line between participation and authority. Votes, consultations and attendance produce evidence. They become authorization only when the principal, constituency, denominator, proposition and consequence are intelligible.

An AC appeal has several principals and agents. A representative initiates on behalf of a Member. A small supporting share opens the gate. The eligible Advisory Committee electorate supplies the decision. The Team operates the ballot. The Process Community Group and Advisory Board develop revisions. The Membership approves the Process. The underlying decision maker, working groups, implementers and affected parties bear the result.

A progressive threshold can protect that chain from an unrepresentative turnout. It cannot tell those actors what an affirmative response affirms. If the proposition points to the decision and the consequence points to its reversal, the denominator is precise but the mandate is not.

This is an agency problem in a literal sense. Operational staff possess the practical power to choose the ballot wording and map the responses. If the public rule is contradictory, their configuration silently determines the meaning of Member authority. Staff may make a careful, reasonable choice; that does not make the delegation legitimate or reproducible. The conformance record returns the choice to an inspectable rule before it can determine the outcome.

Sources

1. W3C Process Document — 18 August 2025, section 5.9
2. Current W3C Process Editor’s Draft — section 5.9
3. W3C Process Document — 3 November 2023, section 5.9
4. W3C Process pull request 901 — progressive vote threshold
5. First threshold commit in pull request 901
6. W3C Process issue 886 — adjust the appeal threshold
7. W3C announcement — W3C updates its Process Document
8. W3C guide — Appealing a W3C Decision
9. Amended and Restated Bylaws of World Wide Web Consortium, Inc.
10. W3C Process issue 1034 — time limits on AC appeals
11. W3C Process for Busy People — AC Appeals
12. Heng Lu — The Multi-Stakeholder Mirage
13. Heng Lu — On the Agency Problem at the Core of Internet Governance