Historical Link Audit of 650,000 Web Addresses Reveals 77 Percent Decayed Over a Decade

An empirical audit of historical web data has revealed that nearly 77 percent of web links created between 2009 and 2014 no longer point to functional web pages. As first reported by Hacker News, the findings stem from an automated analysis of 657,607 pre-2015 short links preserved in a database backup from Macedonian URL shortener service 0.mk. The dataset provides a rare, large-scale snapshot of online sharing behavior during the early 2010s tech ecosystem, documenting the widespread decay of small-scale web infrastructure over the past decade.

Originally launched in 2009 by a team of three developers, 0.mk operated as the first dedicated link-shortening service in Macedonia. The founders ran the platform as a part-time side project for several years before shutting down operations in 2014. At the time, hosting fees and the engineering burden of filtering constant spam and reviewing abuse reports outstripped the platform's revenue. Seventeen years after building the original system, the team recovered a historical backup and restored its records to evaluate how much of the early web survived.

In August 2026, the team deployed an automated crawler to test the destination targets across all 657,607 restored records. After filtering out 2,429 entries that pointed to internal, malformed, credentialed, or policy-restricted destinations, the audit evaluated 655,178 safe links. The results showed that 76.7 percent of those addresses failed to load a working web page. Furthermore, the remaining 23.3 percent that did respond included login portals, parked domains filled with display ads, and error notifications, indicating that the true retention of original content is significantly lower.

To account for duplicate links pointing to identical destinations, the researchers deduplicated the dataset into 494,781 unique URLs, of which 492,620 were crawlable. Among unique destinations, the failure rate increased to 78.7 percent, with only 21.3 percent successfully loading. Connectivity failures at the network layer accounted for 55.0 percent of total unique URL errors following retries from a secondary network. HTTP-level errors made up another 23.7 percent of failures, led by 76,403 instances of 404 Not Found status codes and 29,663 instances of 403 or 429 status codes that may reflect automated request blocking.

The decay pattern was equally pronounced at the domain level, where 98,778 out of 133,605 crawlable hostnames—nearly 74 percent—failed to yield a single functioning destination. The researchers also identified structural anomalies within the raw link volume; for instance, a single user account in 2011 generated 83,398 distinct links targeting pelaphptutorials.com, where 92.5 percent of URLs eventually failed. Excluding that automated cluster reduced the 2011 total from 103,053 records to 19,655, bringing its underlying decay rate into line with surrounding years at roughly 61.7 percent.

The audit highlights a growing divide between centralized technology platforms and independent web publishers. While major infrastructure platforms such as YouTube, Wikipedia, and Google services generally maintained active status or redirected traffic, smaller personal blogs, online discussion forums, and niche media sites vanished in large numbers. For example, none of the 789 distinct Facebook content delivery network URLs in the dataset loaded. Conversely, legacy domain endpoints for shuttered services like Google Code and PureVolume continue to respond for the majority of their historical paths despite the original platforms being inactive.

Because 0.mk catered primarily to users in Macedonia, the dataset serves as a digital archive of a regional web ecosystem that has largely dissolved. Numerous short links targeted national news outlets that subsequently collapsed, including A1 Television, which went off the air in 2011, and newspapers such as Utrinski Vesnik, Dnevnik, and Vest, which ceased operations in 2017. Additionally, the audit tracked 4,478 links pointing to secondary shorteners like bit.ly, TinyURL, and Google's goo.gl. Following Google's complete retirement of the goo.gl service in 2025, those redirection chains broke permanently.

The recovered database also recorded several notable milestones from the platform's history. The very first link created on July 14, 2009, targeted a WordPress CSS stylesheet and logged just two clicks, while a link generated the following day pointed to local host address 127.0.0.1. The most visited link on the service, designated 0.mk/7, was established on July 15, 2009, targeting google.com; it remains operational today with over 95,999 recorded clicks. The longest entry was a 38,753-character CodePen URL created in 2012 to test URL length limits, featuring a query string repeating the phrase TRYING_THE_MAXIMUM_URL.

The revival of 0.mk was made feasible by recent advances in artificial intelligence, which automated many of the labor-intensive tasks that previously forced the service to close. Modern AI tools now manage continuous development, spam identification, moderation requests, and infrastructure monitoring that the small founding team could not sustain manually. The restored service currently operates across a distributed edge network spanning more than 300 cities worldwide, providing low-latency redirection while preserving the historic dataset.