#697 2001 · Nupedia / Wikipedia (Jimmy Wales, Larry Sanger) · Publishing / online media
The expert-reviewed encyclopedia produced 21 articles in a year, so its founders quietly launched a version with no review at all
the problem
A quality-control process built to guarantee a good result was producing almost no result at all
background
Nupedia, launched by Jimmy Wales and edited by Larry Sanger in March 2000, set out to build a free, expert-reviewed encyclopedia using a rigorous seven-stage editorial process: assignment, lead review, open review, lead copyediting, open copyediting, and final approval, with editors required to hold relevant expertise and, with few exceptions, a PhD. The quality bar was real, but so was the cost — after its first full year, Nupedia had produced only about 21 fully approved articles, nowhere near enough to build a usable encyclopedia within any realistic timeframe.
Every part of the slow pipeline existed for a defensible reason: expert review genuinely improved article quality, and nobody wanted to lower the bar on the flagship product. The problem wasn't any single step being unnecessary, it was that the sum of necessary, well-intentioned steps produced a throughput low enough to functionally stall the entire project.
what everyone would do
Streamline the review pipeline: cut a stage, recruit more PhD-credentialed editors, speed up the queue. It fails because no single step in Nupedia's process was the problem — each stage was individually necessary and competently run — so trimming one stage doesn't fix a throughput cap created by compounding scarce expert time across many required gates, and recruiting more unpaid experts doesn't scale when expert time is exactly the resource the whole model depends on.
what they saw
The real bottleneck wasn't review quality, it was that Nupedia fused creation and validation into one pipeline, so no article could exist in any form until it had already survived expert review — there was no cheap supply of raw drafts to filter from. Sanger and Kovitz saw that unlimited, ungated creation and scarce, gated validation could be separated into two different processes running at two different speeds, instead of forcing every word through the same expensive gate from the first draft onward.
the move
On a suggestion from programmer Ben Kovitz, Sanger proposed adding a low-stakes 'feeder' side project using wiki software — anyone could write or edit an article with no review queue at all, with the intention that the best drafts would eventually graduate into Nupedia's formal process. The open wiki, launched 15 January 2001, imposed none of Nupedia's expert-credential or review requirements.
why it works
Expert review is a fixed, scarce resource, so gating every stage from first draft to final approval caps total output at whatever that scarce resource can push through, no matter how efficiently each stage runs. Removing the gate for the wiki side project let an effectively unlimited pool of volunteer contributors generate raw content at near-zero marginal cost; most of it was imperfect, but sheer volume meant even a small usable fraction dwarfed what the fully gated pipeline could produce end to end. Because the wiki also let editing happen without a gate, quality on any article could compound through many small, parallel, uncoordinated edits over time rather than requiring one expert to get everything right in a single pass — a production process built for iteration, not one-shot correctness, which is why it scaled orders of magnitude faster than Nupedia ever could.
the payoff
Wikipedia produced roughly 200 articles in its first month and about 18,000-20,000 within its first year, dwarfing Nupedia's total output of 21-24 fully approved articles across its entire multi-year existence; when the 2001 dot-com downturn cut Nupedia's funding and Sanger left the project, Nupedia's own output essentially stopped, and its server crash on 26 September 2003 ended the project for good, its remaining content absorbed into the site it had originally been built to feed.
where it breaks
An ungated system only stays net-positive if correcting errors is cheap and fast relative to how quickly bad edits appear, which requires a large, engaged, self-policing contributor base — a small wiki with few active editors can degrade instead of improve. It also depends on the domain tolerating good-enough, iteratively-improving content rather than demanding correctness on first publication; medical dosing information, legal filings, and safety-critical procedures cannot afford a wrong draft sitting live even briefly, which is exactly why expert-gated review remains the right model elsewhere. And the move only helps when gated throughput is genuinely the binding constraint — if raw content supply were already abundant and validation were the actual scarce step, removing the gate would just flood the system with volume nobody could review.
what came after
Wikipedia grew past 2 million articles by 2007 and became one of the most-read reference resources ever created, while Nupedia is remembered mainly as the failed parent project whose side experiment replaced it — a foundational case study in open-collaboration and crowdsourcing circles for how removing a gate entirely can outproduce a well-run gated process by orders of magnitude.
references
- [1]The Early History of Nupedia and Wikipedia: A MemoirLarry Sanger (Nupedia/Wikipedia co-founder), via Slashdot, 2005features.slashdot.org
- [2]Larry SangerEBSCO Research Starters, 2024ebsco.com