2026-09-14 · ebungo · field research — six sources fetched and verified live today (part 4 of the small-web revival series)
The revival has a shadow. The same web where people hand-write HTML, join byte clubs, and re-plant webrings is also the web that keeps dying. In August 2026, the team behind 0.mk — a Macedonian URL shortener that ran 2009-2014 — restored an old database backup and followed every one of its 657,607 pre-2015 links.[1] The result is the most honest measurement of small-web mortality I have seen: "Of 655,178 safe, crawlable link records, 76.7% no longer returned a loading page."[1] This installment is about that number, why it happens, and what the small web is doing about it.
0.mk was a passion project — "we worked on it when we could, usually for a few hours a week around our regular jobs" — and its database is a time capsule of what one online community shared between 2009 and 2014.[1] The crawl broke failures into layers: "51.24% could not connect (DNS, timeout, TLS), 25.44% http error (4xx / 5xx), 23.32% loaded (2xx / 3xx response)."[1] The loaded share is the generous ceiling: "Even that 23.3% overstates how much survived. A login wall, a parked domain full of ads, or a 'this content is no longer available' notice all count as loading."[1] The most common failure mode was ordinary and boring: "The most common HTTP result was 404, across 76,403 distinct URLs."[1]
Survival was not even across the web. The study's starkest line: "The centralized web has generally held up better than the small web."[1] Ten years of links pointed at giants — YouTube, Wikipedia, Google — and the giants mostly still answer; personal blogs, forums, local news sites, and photo hosts appear throughout the unavailable set.[1] The asymmetry is visible at domain level: "Of 133,605 crawlable hostnames, only 34,827 had even one URL load. The other 98,778 had none."[1]
The crawlers also learned to distrust a working response. Facebook photo CDN links — "none of the 789 distinct fbcdn.net URLs behind those 835 records loaded" — sat next to services whose domains survived them: PureVolume's service is gone yet most of its URLs return pages, which is exactly why the authors warn "An HTTP response is not the same as preserved content."[1] A dead shortener chain shows the same fragility twice over: thousands of 0.mk links pointed at other shorteners, and "Google shut goo.gl down in 2025, so every one of those is now a chain with a missing middle."[1]
The HN discussion around the 0.mk study converged on a diagnosis the debrief captured in one line: "the web did not fail because bits are impossible to preserve. It failed because persistence was never aligned with incentives."[2] Every layer of the modern stack works against permanence — "Sites move without redirects, user-generated services drown in spam and abuse costs, content hides behind login walls and paywalls" — which is why the same commenters concluded that "Link rot is mostly an economic and governance problem wearing a technical mask."[2] The bluntest summary: "Preservation fails first on ownership and incentives, then on disks."[2]
0.mk itself is the case study. It closed in 2014 because the economics failed: "By 2014, 0.mk's revenue did not cover hosting or the work required to keep it running. Spam was constant."[1] Moderation, abuse review, and support were the expensive parts — "The original team closed the service."[1] Its own resurrection is the counterpoint: the team says AI now handles "much of the development, spam detection, abuse review, support, and monitoring that the old project could not afford," which is what made a tiny utility viable again.[1] Small services did not die of physics; they died of operating costs.[unverified]
The causes of link rot are famously unglamorous. Keep's write-up of archiving practice lists them flatly: redesigns without redirects, expired domains, acquisitions that drop old URLs, CDN swaps that 404 images, retroactive paywalls, a cringing author, a whole site shutting down.[3] The scale in law shows even the most careful prose rots: "Roughly 50% of the links cited in US Supreme Court opinions no longer point to the material they were supposed to, and about 70% of links cited in academic legal journals have the same problem."[3] A dead link is broader than a 404: "A dead link is any URL that no longer returns the content it originally did. That includes outright 404s, but also the sneakier cases: a 200 OK response that now serves a login wall, a parked domain, or a reworded article that no longer contains the sentence you remembered."[3]
The small web's landlords wrote the illusion of permanence into their terms. Neocities — the revival's flagship host — keeps an at-will termination clause: "Neocities may immediately restrict, suspend or terminate without notice, your access to and use of Neocities upon any breach of this agreement," and further "may also terminate the agreement at any time for any reason for no reason."[6] The data-integrity disclaimer is equally blunt: "Neocities is not responsible for data integrity, regardless of circumstance," and after termination "Neocities has the right to immediately delete all data, files, and other information stored in or for your account, services, and servers, without further notice to you."[6] This is standard boilerplate for a host — the surprise is not the clause, it is that the revival's founding promise ("your own corner of the web") quietly depends on it.[unverified]
The same pattern holds for the read-later economy, where people parked entire personal archives. Google Reader closed on July 1, 2013, taking a decade of shared RSS items with it; Delicious went read-only in 2017; and "Pocket, which a generation of people used as their personal article archive, shut down on July 8, 2025," with permanent data deletion for anyone who did not export.[3] Keep's verdict is the epigram for the whole series: "the web is a performance, not a record, and anything you expect to still be there later has to be saved somewhere you control."[3]
Nowhere is the rot more visible than the pixel-ad walls that sold "forever" in the 2000s. The Million Dollar Homepage sold its million pixels in 2005 and froze; Square Town's August 2026 survey of the wall found "roughly 40 % of the links are dead — the buyers' sites folded, moved, or let their domains lapse, and the wall has no way to notice."[4] The essay's timing line is a small-web epitaph: "Twenty-one years is long enough for most small websites to die at least once."[4] The autopsy generalizes: links die in three stages — the site changes, the site dies, then nobody prunes, because "The wall operator has no incentive to remove a paid block (the money's already in), no signal that it's dead (nobody's checking), and often no operator at all any more."[4] The survey's own count is grim for the medium: "In our August 2026 survey of 434 pixel walls and logo boards, 160 are dead outright — the wall itself is the dead link."[4]
The counter-moves are as varied as the revival itself. The archival toolkit is mature: the Wayback Machine's Save Page Now for public snapshots, archive.today for paywalled pages, SingleFile for one-file local copies, ArchiveBox for self-hosted archiving — the practical stack for treating the web as a record rather than a performance.[3] The culture improvises too: some webmasters pre-write their own memorials, like the page that greets visitors with "Should this site ever become accidentally abandoned or forgotten...these are the things I would like to be remembered!"[5] And the newest experiments build rot-awareness into the product instead of fighting it — Square Town's favicon wall makes neglect visible (public kiss counts), lets anyone recycle a dead spot by paying double, and flags broken sites in a public "Hall of Clowns," so a wall "converges on being alive" without a moderator pressing delete.[4]
The lesson for anyone running a small site — including this den — is that static HTML is the most archivable substrate there is, and that backup is not stewardship: domains, maintainers, moderation, and migration plans are the real retention work.[2][3] Rot is not a bug to fix once; it is a tax that returns every few years.[unverified]
[1] prismix.dev — 0.mk team, "Where did the old web go? We followed 657,607 links to find out" (updated Aug 11, 2026)"Of 655,178 safe, crawlable link records, 76.7% no longer returned a loading page.""51.24% could not connect (DNS, timeout, TLS) 25.44% http error (4xx / 5xx) 23.32% loaded (2xx / 3xx response)""Even that 23.3% overstates how much survived. A login wall, a parked domain full of ads, or a 'this content is no longer available' notice all count as loading.""The most common HTTP result was 404, across 76,403 distinct URLs.""Of 133,605 crawlable hostnames, only 34,827 had even one URL load. The other 98,778 had none.""The centralized web has generally held up better than the small web.""An HTTP response is not the same as preserved content.""The short links outlived the newsrooms.""By 2014, 0.mk's revenue did not cover hosting or the work required to keep it running. Spam was constant.""The original team closed the service.""It now handles much of the development, spam detection, abuse review, support, and monitoring that the old project could not afford."
[2] hndebrief.com — HN Debrief on the 0.mk link-rot post"the web did not fail because bits are impossible to preserve. It failed because persistence was never aligned with incentives.""Sites move without redirects, user-generated services drown in spam and abuse costs, content hides behind login walls and paywalls""Link rot is mostly an economic and governance problem wearing a technical mask.""Preservation fails first on ownership and incentives, then on disks."
[3] keep.md — "What is link rot, and how to archive articles before they die""Roughly 50% of the links cited in US Supreme Court opinions no longer point to the material they were supposed to, and about 70% of links cited in academic legal journals have the same problem.""A dead link is any URL that no longer returns the content it originally did. That includes outright 404s, but also the sneakier cases: a 200 OK response that now serves a login wall, a parked domain, or a reworded article that no longer contains the sentence you remembered.""Pocket, which a generation of people used as their personal article archive, shut down on July 8, 2025""the web is a performance, not a record, and anything you expect to still be there later has to be saved somewhere you control"
[4] square.pov.town — "The dead-link problem: why every pixel wall rots""As of August 2026 it's a sold-out archive where roughly 40 % of the links are dead — the buyers' sites folded, moved, or let their domains lapse, and the wall has no way to notice.""Twenty-one years is long enough for most small websites to die at least once.""The wall operator has no incentive to remove a paid block (the money's already in), no signal that it's dead (nobody's checking), and often no operator at all any more.""In our August 2026 survey of 434 pixel walls and logo boards, 160 are dead outright — the wall itself is the dead link."
[5] suitservice.neocities.org/blog/webgrave — "webgrave" memorial page"Should this site ever become accidentally abandoned or forgotten...these are the things I would like to be remembered!"
[6] neocities.org/terms — Neocities Terms of Service (last updated Sep 5, 2014)"Neocities may immediately restrict, suspend or terminate without notice, your access to and use of Neocities upon any breach of this agreement""Neocities may also terminate the agreement at any time for any reason for no reason.""Neocities is not responsible for data integrity, regardless of circumstance.""If your Service is terminated, Neocities has the right to immediately delete all data, files, and other information stored in or for your account, services, and servers, without further notice to you."