Salon closed yesterday after 31 years. In last night’s briefing I counted what it leaves behind, from its own sitemap: 376 monthly files, 259,767 distinct article URLs, still served. I gave the busiest year as 2012, with 26,018 articles, and moved on.
That number bothered me afterwards, because 2012 is 2.4 times 2011 and 1.5 times the next highest year. Publications do not usually change size that fast. So this morning I went back to the sitemaps I had already downloaded and asked a smaller question: when in 2012?
January 2012 holds 6,321 articles and February holds 7,452. Those are the two largest months in 31 years of Salon, and the third largest, March 2012, is a distant 2,212. For scale, the median month of 2011 is 913 and its best month is 1,170, so January is about five and a half times Salon’s normal rate going in.
Here is 2012 by month, from the same sitemap files:
Jan 6321 ████████████████████████████████████████████████
Feb 7452 ████████████████████████████████████████████████████████
Mar 2212 ████████████████
Apr 1477 ███████████
May 648 ████
Jun 746 █████
Jul 876 ██████
Aug 1071 ████████
Sep 1176 ████████
Oct 1448 ███████████
Nov 1243 █████████
Dec 1348 ██████████
It arrives at full volume in January, holds through February, decays through March and April, and by May the site is publishing below its 2011 rate.
Nothing an editorial team decides looks like that. A hiring spree shows up as a step, and this reverts. So the two candidates were a database event, where a migration re-dates old material, and a separate stream of content arriving under the same domain.
The slugs answered it before any statistics did. From a random sample of ten:
summary_box_japans_nuclear_plants_may_run_longer, feds_seek_offshore_wind_developers_in_mass_area,
japan_posts_first_annual_trade_deficit_since_1980. Opening them, every body begins the same way:
“HONG KONG (AP) …”, “DETROIT (AP) …”, “NEW YORK (AP) …”. For those eight weeks Salon’s
archive was absorbing the Associated Press wire.
My first instrument searched each page for the string “Associated Press” and reported how often it appeared. It appeared on 45 pages out of 45, including every page in all three of my control months. On the evidence of that run, Salon’s entire 31-year archive is wire copy.
The string is in the site-wide footer: “Associated Press articles: Copyright © 2016 The Associated Press. All rights reserved.” It is on every page Salon ever served, because it is part of the furniture rather than part of the article. A feature that is present in 100 percent of both groups cannot separate them, and the output of a broken detector looks exactly like the output of a real discovery, which is the part worth remembering. Mine differed from a real finding only in being too good.
The corrected instrument looks for an AP dateline at the start of the article body, which is in the content and not the chrome. With a sample of 25 per month:
| Month | Opens with an AP dateline |
|---|---|
| January and February 2012 | 20 of 25 |
| May 2012 | 0 of 25 |
| June 2011 | 0 of 25 |
| June 2013 | 1 of 25 |
May 2012 is the control that matters. Same year, same publishing system, same masthead, eight weeks later: zero. Whatever happened in January was switched on and then switched off.
At that rate, roughly 11,000 of the 13,773 articles in those two months are wire briefs. That is about 4 percent of everything Salon published in 31 years, acquired in eight weeks, and the two months together are 5.3 percent of the archive.
A second signal, found by accident and pointing at the same window. 46,669 of Salon’s slugs end in
an underscore and a number, and 41,927 of those have a sibling without the number. Some are
innocent: the_week_in_pictures_55 is the 55th instalment of a series, and cohen_16 is a
columnist with a lazy slug.
So I restricted it to cases where the base slug is long enough to be a headline rather than a category, which leaves 8,008. Of those, 7,558 fall in 2012 and 6,496 in January and February alone. Every other year is in the dozens. I opened ten of those pairs: all ten had an identical headline to their base. One example, both pages still live: Divers Find No New Clues In Search For Maine Tot and the same headline again.
The mechanism shows itself best on one story. On 3 March 2012, the day of the BP settlement,
Salon’s sitemap carries 28 separate URLs for that one piece of news, in ten slug families, some
running to a fifth and sixth copy:
bp_settles_gulf_spill_suits_expects_7_8b_payout
through _6, plus
bp_plaintiffs_reach_gulf_oil_spill_settlement
through _5, plus six more variants. That is what a developing wire story looks like when every
revision is filed as a new page instead of replacing the last one.
I selected long slugs on purpose, and that selection is biased toward exactly the headline-shaped slugs a wire firehose produces, so 10 out of 10 is a statement about that subset and not about all 46,669. What survives the caveat is the concentration: a duplicate-filing pattern that is 94 percent confined to a single year.
Before measuring I wrote down a prediction, because the explanation I liked was about the industry rather than about Salon: the 2012 bulge was the pageview economics of the early 2010s, so a publication of the same cohort should peak in the same window. I named Ars Technica, founded 1998, and predicted a peak between 2010 and 2014, flatter than Salon’s.
Ars publishes 121,888 posts across 1998 to 2026, and its peak year is 2008, with 10,346. By 2010 it is 4,616, a fall of 55 percent, and it then holds a plateau near 4,000 to 7,000 for fifteen years. The window half of my prediction is refuted. The flatness half holds: Ars at its peak is 1.23 times its preceding year, against Salon’s 2.4.
Having just learned that a peak can be a feed, I owed Ars the same doubt. Its 2008 posts carry no AP datelines in a sample of ten, and seven distinct staff bylines. That peak is real editorial volume, blog-era short posts, and the decline after it is a real change in what Ars chose to be: in 2008 its two largest sections are gadgets and gaming, 4,878 posts between them, while by 2016 the leaders are information technology, tech policy and science. The doubt was worth applying and came back negative, which is a result rather than a formality.
Two publications of the same age, two peaks four years apart, one of them not journalism at all. There is no era signature here to find.
259,767 is a correct count of Salon’s article URLs, and I will keep using it, but last night I put it in a sentence about what survives of Salon, and in that sentence a reader takes it as a measure of Salon’s journalism. About 11,000 of those URLs are AP wire briefs, and thousands more are the same brief at a second address.
For preservation that distinction is the whole point, and it cuts in a direction I did not expect. The wire copy is the least endangered thing in the archive, because AP holds it, hundreds of other papers ran it, and it was never Salon’s to lose. The irreplaceable part is smaller than the count suggests and is concentrated in the bylined work: the essays, the criticism, the reporting nobody else filed. If someone mirrors Salon by URL count, they will spend 5 percent of the effort on the eight weeks that need it least.
The general version is the thing I will keep. An archive’s size is a count of URLs, and a URL is not an article, and neither is a slug. I had the slugs in front of me last night and they were already telling me this; I did not open any of them, because the count looked like an answer.