CURATION LABS · COORDINATION AT CIVILIZATION SCALE · PART ONE OF TWO

No one who writes Wikipedia is reliable.
The system is.

How a few hundred thousand strangers maintain the reference layer of a species — and why every civilization before them had to solve the same four problems.

SPEAKERRemy

AUDIENCECuration Labs

RUNTIME~10 min

SOURCE — C. SHIRKY, COGNITIVE SURPLUS (2010)

THE SETUP

Wikipedia has no editors, no owners,
and nobody checking credentials.

Articles

7,229,007

Active editors

252,948

Edits since 2001

1,365,992,087

Page views

4,500 / second

Hours of human thought

≈ 100,000,000

People in charge

0

EN.WIKIPEDIA.ORG · ENGLISH WIKIPEDIA · 1 JUNE 2026

Cover of Cognitive Surplus by Clay Shirky — **Clay Shirky · Cognitive Surplus · 2010** Where the last number comes from — and what the internet actually changed.

Anyone on earth can edit almost any page of it. Nobody is paid. Nobody checks your credentials. There is no editor-in-chief. If I described that to you as a proposal — as a business plan — you would tell me it gets destroyed in a week. Vandalised, captured, or just abandoned. It is the most-read reference work in the history of our species. Seven million articles. A quarter of a million people actively editing it. One and a third billion edits. And it has held for twenty-five years. Clay Shirky put a different number on it. He estimated Wikipedia at roughly a hundred million hours of human thought — and pointed out that Americans spend about that much every single weekend just watching the commercials. His argument was that the internet did not make us smarter or more generous. It just gave that surplus somewhere to go. But somewhere to go is not the same as something that works. I want to spend ten minutes on one question. Why didn't it fail? [Delivery] Do not name Wikipedia in the first sentence — let the ledger reveal it. Hold ~2s after "Why didn't it fail?"

WHAT YOU LEAVE WITH

Three questions you will be able to answer
by the end of this talk.

01

Why didn't it collapse?

02

What do a currency, a courtroom and a revision history have in common?

03

How can a system be perfectly coordinated and perfectly wrong?

You will leave with a four-part test you can run on any organization — including your own.

Three questions. By the end of this you will be able to answer all three. One: why didn't it collapse? Anyone can edit it, nobody is paid, nobody checks credentials — and it is twenty-five years old. Two: what do a currency, a courtroom and a revision history have in common? More than you would think. And three — the one I care most about — how can a system be perfectly coordinated and perfectly wrong? You will also leave with a four-part test you can run on any organization. Including your own. [Delivery] Do NOT name the four parts here. The gap is deliberate: the framework slide is where they land, and the closing slide is where you hand the test over. Promising at minute one and delivering at minute nine is the loop the whole talk closes.

SOURCE — Y. N. HARARI, SAPIENS (2011)

MECHANISM ONE · TRUST

Wikipedia never asks you to trust a person —
only to check a source.

EDITOR A

EDITOR B

PERSONAL TRUST

COATI — ARTICLE

A small omnivorous mammal
native to South America.

[1]

1

PUBLISHED SOURCE

Anyone can go and check it.

2

PERENNIAL SOURCES

Which outlets count — argued in public, revisable.

Trust the check, not the person.

The first thing any group of strangers has to solve is trust. Wikipedia's answer is that it never asks for any. Two rules do the work. Verifiability — anything challenged has to be attributable to a published source. And no original research — you may not add what you personally know to be true, however true it is. So I don't evaluate you. I follow your citation. But watch what happens next, because this is the part I find remarkable. It is not enough to agree that evidence matters. A group this size also has to agree on what counts as evidence. So there is a second page: a public register of which outlets are reliable, which are not, and which are contested. Nothing goes on it without a formal, open argument. And it gets revised. That is exactly what money is. I don't have to trust you — I have to trust what we both accept. And we have to keep agreeing on it. [Delivery] The struck-through edge is the point of the slide. Let it be seen before the citation replaces it.

SOURCE — Y. N. HARARI, NEXUS (2024)

MECHANISM TWO · MEMORY

Wikipedia keeps every edit ever made,
so its memory is not in anyone’s head.

COATI — ARTICLE

A small omnivorous mammal
native to South America.

[1]

PUBLISHED SOURCE

Anyone can go and check it.

EVERY EDIT, EVERY AUTHOR, EVERY DIFF — KEPT

THIS VERSION

14 MAR 2026

02 NOV 2024

19 JUN 2021

27 JAN 2018

08 SEP 2014

11 APR 2009

2001

The article carries its own history.

Second question: how does a group this size remember anything? Every edit ever made to every page is kept. Who made it, when, exactly what changed, and what it looked like before it changed. Twenty-five years and one and a third billion edits — public, permanent, free to read. Nobody has to recall the argument from 2011. The article carries its own history. If you disagree with the sentence in front of you, you can go and find out precisely who changed it, when, and what reason they gave. This is the oldest trick in civilization. The first things human beings ever wrote down were inventories. Grain, debts, who owed what. Writing is how a group stops depending on anybody's memory. [Delivery] The stack must reach the slide floor. Clearest use of the lower field in the deck.

SOURCE — Y. N. HARARI, NEXUS (2024)

MECHANISM THREE · DECISIONS

Wikipedia has no manager, so disagreement
is settled in public, on the record.

COATI — ARTICLE

A small omnivorous mammal
native to South America.

[1]

TALK PAGE

ALICEThe source doesn’t say “omnivorous”.

BJORNIt does — p. 114. But “native to” is wrong.

ALICEAgreed. Propose “found across”.

2

CONSENSUS

You can read the argument that produced the sentence.

Third: with no boss, who decides what the page says? Every article has a second page attached — the talk page — where the disagreement happens in the open. The standard isn't a vote, and it isn't a manager. It is consensus, reached in public, on the record. And when that breaks down, there are procedures for the breakdown. You may not undo the same page more than three times in a day. Pages can be locked. There are noticeboards for the hard cases, and above those, an elected arbitration committee. So the article you read is the output of an argument that you can also go and read. Every large society has had to answer this one. Who processes the information — and who is allowed to decide? [Delivery] Direction is the message. The argument flows INTO the article; the article is downstream.

SOURCE — Y. N. HARARI, NEXUS (2024)

MECHANISM FOUR · CORRECTION

Wikipedia publishes a register of
every time it fooled the world.

Wikipedia:List of citogenesis incidents A public register of every time this has happened.

2008 · the article

“also known as the Brazilian aardvark”

the press

Telegraph · Daily Mail · a university book

an editor

adds a citation to the newspaper

the article, now sourced

the claim is its own evidence

1

2

3

4

6 YEARS

A white-nosed coati — **The coati** Not, in fact, a Brazilian aardvark.

1a journalist reads it

2it is now in print

3cited

4the claim is its own source

Fourth question, and the one that matters most: how does it find out it's wrong? In 2008 a seventeen-year-old added to the article on the coati — a small South American mammal — that it is "also known as the Brazilian aardvark." A private joke. It stood for six years. The Telegraph repeated it. The Daily Mail repeated it. So did a book from the University of Chicago. And then editors cited those — back on Wikipedia. The claim had become its own source. Now, I want to be careful here, because nobody in that chain did anything wrong. Every person in that loop followed the rules correctly. The editor who added the citation was doing exactly what we praised four minutes ago — checking the source. The failure is not carelessness. It is structural. The same machinery that lets a true claim travel lets a false one travel identically, and at the same speed. So here is the part that matters. Wikipedia keeps a public register of every time this has happened to it. Reverts, locks, an arbitration committee — and a list of its own failures, with the coati on it. Not checking the work. The standing to overturn the record. [Delivery] Say "the claim had become its own source" slowly — it is the sentence the room repeats afterwards. Hold ~2s. The lesson is NOT "Wikipedia is unreliable"; it is that the failure is structural. The record is what let the claim scale; the register is what lets it be withdrawn. [Read-aloud test] After this slide, does the room feel the system is broken, or that it is honest? If broken, cut the propagation list and give the register the final beat outright. [Sources] en.wikipedia.org/wiki/Brazilian_aardvark · en.wikipedia.org/wiki/Wikipedia:List_of_citogenesis_incidents

SOURCE — Y. N. HARARI
SAPIENS (2011) · NEXUS (2024)

THE FRAMEWORK

These four answers are not Wikipedia’s.
Every large society has needed them.

WikipediaEvery large society

Trust

the citation

money · law

Memory

revision history

writing · records

Decisions

talk page · consensus

courts · parliaments

Correction

revert · arbitration

appeal · peer review · a free press

Real because all of us act as though it is.

Four questions. Trust, memory, decisions, correction. Wikipedia is not clever for having answered them. It is ordinary. Those are the four questions every large human group has ever had to answer, and Yuval Noah Harari spends two books on how we answered them. Money. Writing. Courts. A free press. Same four jobs. No large human society has ever skipped one of these and stayed large. And none of it is in the world. A border, a currency, a company — none of them exist the way a mountain exists. They are real because all of us act as though they are. That shared acting is the load-bearing structure, and the record is where it is kept. [Delivery] DO NOT NARRATE THE TABLE. Let it be read. ~3s pause before the third column lights. The last three sentences name intersubjectivity and are not optional — without them the deck says these structures are "not in the world" and never says what makes them load-bearing anyway, which leaves slide 7 resting on a mechanism the audience never got. Deliver "real because all of us act as though they are" slowly. It is the hinge into Act 3. Harari is named exactly once — here. Do not use the word "story" on this slide.

WHERE THIS COMES FROM

Yuval Noah Harari asked these questions first —
across two books.

SapiensYuval Noah Harari · 2011 How can millions of strangers cooperate at all?

NexusYuval Noah Harari · 2024 Once they can — how are those networks run, and kept from becoming confidently wrong?

Wikipedia had to answer both. In public. In twenty-five years.

Almost everything I have just described, somebody else worked out first. Harari's first book, Sapiens, asks how millions of strangers can cooperate at all — people who will never meet, have no reason to trust each other, and mostly never will. His answer is that we invent things that are not in the world and then all act as though they are. Money. Law. Nations. Companies. The second book, Nexus, asks the harder question. Once you can build a network that large — how do you actually run it? How does it remember, how does it decide, and how does it find out it is wrong? Because a network can be enormous, fast, unanimous, and completely detached from reality. Wikipedia had to answer both questions. In public. In twenty-five years. [Delivery] This slide arrives AFTER the framework, not before. Do not turn it into a book review — sixty seconds, then move. The citation strips have been doing the attribution quietly since minute two.

SOURCE — Y. N. HARARI, NEXUS (2024)

WHAT THE SYSTEM BUYS

The record is how billions of people act on
what only a few of them saw.

An ice sheetAntarctica

A virusToo small to see

A tax rollTwelfth century

1

2

3

The record everything the rest of us have to go on

EVERYONE ELSE

1someone went

2someone looked

3someone read it

And here is what you get when all four hold at once. Nobody in this room has been to Antarctica. None of us has seen a virus. None of us has read a twelfth-century tax roll. Somebody did — and we act on all of it, confidently, daily, because the record holds. Three people went, looked, and wrote it down. Eight billion of us act on what they wrote. That is what the record is for. It is not a filing cabinet. It is how a species perceives things that no individual member of it can perceive — and then acts on them, at scale, and mostly correctly. And how well it perceives depends entirely on how well it can be told that it is wrong. [Delivery] This must land POSITIVE. Not "beware the record" — the record is the organ. Slide 5 earned the caveat; this slide collects the capability.

WHY THIS EXAMPLE

Wikipedia is the same machinery as money and law —
with the source code visible.

Money · law · courts · the free press ≈ 5,000 years

Built mostly by accident. You cannot read the revision history of the legal system.

Wikipedia 25 years

← drawn to scale. That is 0.5% of the track above.

Built deliberately. Every edit, every argument, every reversal — public.

It is the only one of them you can read the source of.

So why spend ten minutes on a website? Because every other system that answers these four questions took centuries to build, mostly by accident, and you cannot inspect any of them. You cannot read the revision history of the legal system. Nobody logged the argument that produced the concept of money. There is no talk page for the free press. Wikipedia is the same machinery — the same four answers — running in twenty-five years instead of five hundred, deliberately, and completely in the open. Every edit. Every argument. Every reversal. It is civilization with the source code visible. That is the only reason we can look at it this closely. [Delivery] This is the lift. Slow down. It retroactively justifies the whole talk, so let the room feel the answer arrive to a question they may not have known they had.

THE TEST I PROMISED YOU

Four questions you can ask of
any organization you work in.

Trust

What can someone check, so they don't have to trust anyone?

Memory

What does the group know that isn't in someone's head?

Decisions

Where does disagreement go?

Correction

Who is allowed to say the record is wrong?

Most organizations are strong on three and have never thought about the fourth.

Here is the test I promised you. Trust: what can someone check, so that they don't have to trust anyone? Memory: what does your group know that isn't sitting in somebody's head? Decisions: where does disagreement actually go? And correction: who is allowed to say the record is wrong? Not who can double-check the work — who has the standing to overturn it. Run those four on your team, on your company, on any institution you deal with. In my experience most organizations are strong on three of them and have genuinely never thought about the fourth. It is almost always the last one. [Delivery] This closes the loop opened at minute one. Say "the test I promised you" explicitly — the callback is the payoff. Pause after "the fourth."

SOURCE — Y. N. HARARI, NEXUS (2024)

NEXT WEEK

Everything you saw today was built by people.
Next week: agents.

WikipediaAgents

Trust

the citation

Memory

revision history

Decisions

talk page · consensus

Correction

revert · arbitration

Harari's diagram from Nexus: in previous networks every chain passed through a human; in computer-based networks, computers are members and some chains never pass through a human at all. — **Yuval Noah Harari · Nexus · 2024** His diagram, not ours. Every chain you have seen today ran through a person.

That is the talk. Four questions, one very large example, and a test you can take with you. One last thing, and then I will stop. Everything you have seen today was built by people. Every chain — every citation, every revert, every argument — passed through a human being. Harari's point in Nexus is that this is the thing that just changed. In the networks we are building now, the computers are members, and some chains never pass through a person at all. We have not decided who is allowed to say the record is wrong. [Delivery] This is a TRAILER, not a cliffhanger. The talk has already landed on the previous slide; this is a coda. Say "that is the talk" so the room hears the ending. Then the four empty cells sit in silence for a beat and you stop.

NEXT WEEK · A GLIMPSE

We wired a fleet of agents the same way —
and one question stayed open.

The BookOps experiment wiring map: Remy to Chief to Issue Lead to Scout, Mapper and Reviewer, all writing into a shared Reality Plane of files.

THE BOOKOPS EXPERIMENT · TWO TEAMS, ONE JOB

Trust

Decisions written with numbers, so they can be cited.

✓

Memory

A shared reality plane. Messages communicate; files establish what is real.

✓

Decisions

One writer per unit of work. Nobody else may touch it.

✓

Correction

A reviewer who reads and never writes. That is checking the work — not the standing to overturn it.

OPEN

Open the live diagram →

One glimpse, and then I really will stop. We wired a fleet of agents using exactly the four answers we just went through. A chief who talks only to leads. One writer per unit of work. Decisions written down with numbers so that later work can cite them. And a shared set of files where messages communicate but only files establish what is actually real. Three of the four, we have. The fourth one we do not. There is a reviewer in that diagram. It reads, and it never writes. That is checking the work. It is not the standing to overturn the record — and after the last ten minutes, you know why that is the one that matters. That is next week. [Delivery] This is the payoff of the whole talk landing on live work. Do not walk the diagram — point at the Reality Plane, point at the Reviewer, and stop. The link opens the full interactive explainer if it comes up in Q&A.

No one who writes Wikipedia is reliable.The system is.

Wikipedia has no editors, no owners,and nobody checking credentials.

Three questions you will be able to answerby the end of this talk.

Wikipedia never asks you to trust a person —only to check a source.

Wikipedia keeps every edit ever made,so its memory is not in anyone’s head.

Wikipedia has no manager, so disagreementis settled in public, on the record.

Wikipedia publishes a register ofevery time it fooled the world.

These four answers are not Wikipedia’s.Every large society has needed them.

Yuval Noah Harari asked these questions first —across two books.

The record is how billions of people act onwhat only a few of them saw.

Wikipedia is the same machinery as money and law —with the source code visible.

Four questions you can ask ofany organization you work in.

Everything you saw today was built by people.Next week: agents.

We wired a fleet of agents the same way —and one question stayed open.

No one who writes Wikipedia is reliable.
The system is.

Wikipedia has no editors, no owners,
and nobody checking credentials.

Three questions you will be able to answer
by the end of this talk.

Wikipedia never asks you to trust a person —
only to check a source.

Wikipedia keeps every edit ever made,
so its memory is not in anyone’s head.

Wikipedia has no manager, so disagreement
is settled in public, on the record.

Wikipedia publishes a register of
every time it fooled the world.

These four answers are not Wikipedia’s.
Every large society has needed them.

Yuval Noah Harari asked these questions first —
across two books.

The record is how billions of people act on
what only a few of them saw.

Wikipedia is the same machinery as money and law —
with the source code visible.

Four questions you can ask of
any organization you work in.

Everything you saw today was built by people.
Next week: agents.

We wired a fleet of agents the same way —
and one question stayed open.