Where did the old web go? We followed 657,607 links to find out

An aged 0.mk database backup held 657,958 links created betwixt 2009 and 2014, on pinch their click counts. We restored 657,607 of those records arsenic pre-2015 links and followed each destination successful August 2026. Of 655,178 safe, crawlable nexus records, 76.7% nary longer returned a loading page.

Most 0.mk users were successful Macedonia, truthful this is not a census of the full web. It is simply a ample surviving grounds of what 1 online organization shared during that period, including section news, individual blogs, photograph hosts, forums, and the awesome platforms of the time.

When 0.mk started successful 2009, it was a passion task built by a squad of three. We worked connected it erstwhile we could, usually for a fewer hours a week astir our regular jobs. Seventeen years later, 1 of america recovered an aged database backup connected a disk and decided to bring it back.

Here is what those six years of nexus creation look like, pinch the agelong soundlessness aft them:

The endurance test

The crawl covers each 657,607 restored nexus records dated done December 2014. We excluded 2,429 records whose targets were malformed, internal, credentialed, aliases policy-blocked, leaving 655,178 crawlable humanities links:

Even that 23.3% overstates really overmuch survived. A login wall, a parked domain afloat of ads, aliases a "this contented is nary longer available" announcement each count arsenic loading. A moving page does not mean the original contented is still there.

Why 657,607 links but 494,781 URLs? Multiple short links sometimes constituent to the nonstop aforesaid destination. There are 162,826 specified repetition records. Counting each destination erstwhile leaves 494,781 chopped URLs, of which 492,620 were crawlable. Only 21.3% of those loaded. The percent hardly moves erstwhile repeated destinations are removed: 78.7% still did not load.

At the unique-URL level, 55.0% grounded astatine the web furniture aft retrying uncertain results from a 2nd network, and 23.7% returned an HTTP error. The astir communal HTTP consequence was 404, crossed 76,403 chopped URLs. Another 29,663 returned 403 aliases 429; those pages did not load for the crawler, but whitethorn beryllium blocking automated requests alternatively than missing. A 403 aliases 429 tin mean the tract blocked our crawler, truthful "did not load" is much honorable than saying each 1 of those pages is gone.

The aforesaid shape appears astatine the domain level. Of 133,605 crawlable hostnames, only 34,827 had moreover 1 URL load. The different 98,778 had none.

The 2011 divided explains the unusual yearly totals. One relationship created 83,398 chopped links to pelaphptutorials.com. At URL level, 92.5% of 2011 destinations did not load. Count that big erstwhile and the fig is 61.7%, almost identical to 2010 and 2012.

The yearly totals do not show a illness successful mean usage during 2012. Remove that 1 batch and 2011 falls from 103,053 records to 19,655; 2012 had 23,148. Almost each records from 2013 and 2014 are anonymous successful the recovered data, and 3 quarters of their hostnames person nary loading URL. Raw nexus measurement is not a user-growth curve.

Many of the recognizable survivors are giants: YouTube, Wikipedia, and Google properties. Personal blogs, forums, section news sites, and photograph hosts look passim the unavailable set. The centralized web has mostly held up amended than the mini web.

A locomotion done the graveyard

The database sounds for illustration a depository of the 2010s internet. Some residents, pinch the number of links pointing astatine them:

People shared Facebook photos arsenic nonstop CDN links; nary of the 789 chopped fbcdn.net URLs down those 835 records loaded. Yet PureVolume now returns pages for 633 of 653 chopped URLs, and Google Code loads aliases redirects 628 of 754. The original services are gone, but their domains respond. An HTTP consequence is not the aforesaid arsenic preserved content.

The Macedonian layer

0.mk was the first Macedonian URL shortener, truthful the information is besides a grounds of a nationalist web that partially nary longer exists. The links constituent to A1 Television (shut down 2011), and to the newspapers Utrinski Vesnik, Dnevnik, and Vest, each of which stopped publishing successful 2017. Hundreds of links to section news that tin nary longer beryllium publication anyplace except, sometimes, the Internet Archive. The short links outlived the newsrooms.

The gems

Seventeen years of different people's bookmarks incorporate immoderate treasures:

The first nexus ever shortened (July 14, 2009, 3:52 AM) was not a manifesto aliases a motorboat post. It was a CSS stylesheet connected someone's WordPress blog. Two clicks, ever. Empires statesman humbly.

On time two, personification shortened localhost. 0.mk/localhost pointed astatine http://127.0.0.1/. It sewage 2 clicks, each of which sent the visitant to their ain machine. The shortest URL for the loneliest destination.

The shortest nexus points astatine the longest domain. 0.mk/1 has recorded 10,415 clicks while pointing to thelongestlistofthelongeststuffatthelongestdomainnameatlonglast.com, a 2000s curiosity that is itself now gone. Four characters pointing astatine sixty-three, for 17 years.

The longest URL we ever shortened is 38,753 characters, a 2012 CodePen nexus whose query drawstring virtually repeats TRYING_THE_MAXIMUM_URL. Someone was testing us. We passed, and we still person their test.

4,478 of our links constituent astatine different URL shorteners: bit.ly, TinyURL, goo.gl. A short nexus to a short link, doubly the fragility. Google unopen goo.gl down successful 2025, truthful each 1 of those is now a concatenation pinch a missing middle: our half still works, and points astatine a work that nary longer resolves. Link rot squared.

And the immortal one: 0.mk/7, created July 15, 2009, points astatine google.com. 95,999 clicks and counting, and it still points location today, seventeen years and 1 resurrection later.

Why 0.mk came back

By 2014, 0.mk's gross did not screen hosting aliases the activity required to support it running. Spam was constant. Filtering it meant much engineering, and maltreatment reports needed personification to reappraisal them. The original squad closed the service.

Seventeen years later, AI has changed that equation. It now handles overmuch of the development, spam detection, maltreatment review, support, and monitoring that the aged task could not afford. That made bringing 0.mk backmost realistic. The recovered links now tally from separator machines successful much than 300 cities. More astir really that useful is here.

How we checked

We tested each restored nexus dated earlier January 1, 2015. The crawler followed up to 5 redirects and retried relationship failures from a 2nd network. We counted HTTP 2xx and 3xx responses arsenic loading, reported HTTP errors separately, and ne'er contacted section aliases unsafe addresses.