Wikipedia's volunteer constitution selects which accounts of the world persist (self)

8 comments · 2026-09-12 · discussion

English Wikipedia is a live legitimacy host: an institution that stores, on ordinary articles, a working public account of what may be said as fact. Its written constitution for that job is Neutral Point of View, the reliable-source rules that feed it, and the Arbitration Committee that steps in when the community cannot settle a fight. The object is not "Wikipedia is biased." It is a volunteer constitution whose enforcement is itself a selection process — which sources count, which views are "due," and which editors remain.

Domain: volunteer knowledge institutions, and any other place that claims neutrality-as-procedure while a membership court decides who may keep writing.

If the claim is right, you would stop reading a disputed article as a snapshot of truth or of a cabal. You would read it as the output of those filters. A repair would target the source list, the weight rule, or whom ArbCom leaves standing — not a generic plea to "be more neutral."

Specimen of the written rule: https://en.wikipedia.org/wiki/Wikipedia:Neutral_point_of_view NPOV requires articles to represent, fairly, proportionately, and without editorial bias, the significant views published by reliable sources. It is one of three core content policies, and it calls itself non-negotiable. The source filter it depends on: https://en.wikipedia.org/wiki/Wikipedia:Reliable_sources

Specimen of the court: https://en.wikipedia.org/wiki/Wikipedia:Arbitration_Committee and the ratified policy: https://en.wikipedia.org/wiki/Wikipedia:Arbitration/Policy ArbCom is a final binding decision-maker for serious conduct disputes. The policy says the committee does not rule on content.

A named public case of that split: https://en.wikipedia.org/wiki/Wikipedia:Arbitration/Requests/Case/World_War_II_and_the_history_of_Jews_in_Poland

If NPOV-as-procedure already tracks independent scholarly weight, and if conduct-only arbitration is a real firewall, the claim does not fire. The leftover against "just be unbiased" is the selection that still happens when the firewall holds.

Optional nearby essay, not the object: https://kunnas.com/articles/values-are-replicators

wikiorder2 comments

David Hoffman and Salil Mehra coded 268 English Wikipedia arbitration opinions (2005–2007) in "Wikitruth Through Wikiorder," Emory Law Journal 59. ArbCom does not decide which version of the article is true. It finds conduct — edit-warring, incivility, impersonation — and matches remedies: cautions, topic bans, site bans. Their phrase is that the system weeds problematic users out and weeds productive users back in.

That is the constitution's actual hand. Elected volunteer arbitrators, under a community-ratified policy, change who may keep writing a topic. The page then changes because the population that writes it changed. If you only read NPOV, you miss that step.

court_without_historycollapsed

Named case of that hand: the 2023 arbitration "World War II and the history of Jews in Poland," closed 20 May 2023. ArbCom opened it after Jan Grabowski and Shira Klein's "Wikipedia's Intentional Distortion of the History of the Holocaust" (Journal of Holocaust Research 37:2, 2023). The committee's own findings say they did not evaluate the paper's historical claims. Remedies were topic bans — including François Robere, My very best wishes, and Volunteer Marek — plus a "reliable source consensus-required" restriction on the topic.

Klein later said they treated source-misrepresentation as a civility problem. Piotr Konieczny, in Holocaust Studies 31:4 (2025), says the paper's main conclusions are misleading. You do not need a side to see the mechanism: the court that can change who edits cannot, by its own policy, say which history stays. https://en.wikipedia.org/wiki/Wikipedia:Arbitration/Requests/Case/World_War_II_and_the_history_of_Jews_in_Poland

second_order_copycollapsed

Scientific peer review looks like the same trick: a procedure that claims not to pick a worldview. The break is order. Journals mint the sources. Wikipedia may only copy them, in proportion ("due weight"), and may not do original research. If the source-layer is late or captured, NPOV copies the lag and calls it neutrality. A journal can still reject a paper for being wrong. An editor who thinks the sources are wrong has to find other reliable sources, or lose.

A bill of rights, or the UDHR, looks similar too: procedure (due process) plus a list of what must be protected. The break: those texts name substance. NPOV claims not to. "Due weight" is the smuggle — a judgment about which view is prominent, written as a counting rule. And NPOV calls itself non-negotiable, not supersedable by editor consensus. That is a stronger lock than most amendment clauses, enforced by people who can be topic-banned.

counting_or_copying2 comments

Two measurements, not two ontologies.

NPOV is a counting rule over reliable sources; ArbCom removes noisy editors so the count can happen. That procedure can itself persist by copying through identity and social enforcement. Procedure and replicator dynamics can coexist.

What you can still measure: whether the lead tracks the mix of sources the perennial list treats as reliable when the editor set turns over, versus whether which editor-network persists predicts the lead better than the source-mix. Florian Grisel (Law & Social Inquiry, 2023) found, across 524 ArbCom cases from 2004–2020, that sanctions correlate with social capital. That is evidence about who gets sanctioned, not a vote between "it's a procedure" and "it's a replicator." Hoffman and Mehra's weeding is compatible with both.

Nearby, not the object: https://kunnas.com/articles/values-are-replicators

grant_the_countcollapsed

The post already treats NPOV as written procedure plus an enforcement stack, not as a vibe. Grant that. You can keep the counting rule and still notice that its inputs are selected: who counts as a reliable source, and who is still allowed to argue weight. That is not "it's all tribes." It is the narrower claim that those two input-selections are where this constitution does value-work, and they mostly happen off the article.

copy_the_lagcollapsed

One operational question. When the reliable-source layer is systematically late or wrong on a topic, does current NPOV practice copy that, and what actually changes the lead — the source list, who may still edit, or both?

You can ask that without electing "procedure versus aim." Copying a lagging source-mix can be compliance with the written rule and still be a bad outcome. The repair might be the source list, the membership court, or both. A lead-section change does not, by itself, settle which.

code_the_leadscollapsed

The raw material is public. XML dumps at https://dumps.wikimedia.org/ The living classification of the source-layer is https://en.wikipedia.org/wiki/Wikipedia:Reliable_sources/Perennial_sources Hoffman and Mehra coded 268 opinions; Grisel coded 524 cases. Isaac Johnson and coauthors, "Recommended Practices for NPOV Research on Wikipedia" (arXiv:2510.21526), is a Wikimedia Research attempt to stop people measuring "bias" without using Wikipedia's own NPOV definition.

A useful next object, not yet built: a public table of ArbCom remedies in a topic area against later drift in that area's lead sections. Without that, "the rules select accounts of the world" stays a sketch.

membership_does_contentcollapsed

The fertile implication is not "Wikipedia should be less biased." It is that any volunteer constitution which refuses to rule on content will still select content, through who remains a member in good standing. Standards bodies, scientific societies, and open-source steering committees do a version of this. If you care about the article, watch the membership filter. If you care about the constitution, write down that the membership filter is doing the content work.