Senrigan

» see senrigan.ascii.co.jp as it lived, on the wayback machine

~ the obit ~

Senrigan's FAQ answered the question about system configuration in a single line. A plain PC with 128 megabytes of memory and a plain SCSI disk, running Solaris 2.5.1 on a 200 MHz Pentium Pro. The word for plain there is ふつー, and Kent Tamura used it twice in the one sentence. That was the machine holding the index of the Japanese web.

Tamura built it as a student in the Muraoka laboratory at Waseda University, and by its own account it spent its first ten months with no name on it whatsoever. Later write-ups call that period Searcher in Waseda. From around September 1995 he asked the public to suggest something better. Several dozen entries came in, and the one he took was sent by a reader named Tatsuya Kogo: 千里眼, senrigan, the ordinary Japanese word for clairvoyance.

The crawler ran from a Waseda machine named aniki and identified itself to web servers as Senrigan, and what it produced was the first robot-built search engine in Japan. On 28 March 1997 the whole thing moved onto servers at ASCII Corporation, the computer publisher, leaving a one-line stub on the old Waseda page saying it had moved. Tamura's university mail address stayed in the footer right through to the end. ASCII closed it in March 1999. The masthead had read "since Dec 1994" the whole way through, a month later than the contemporary accounts place the opening.

Senrigan held .jp pages and nothing else, and its results came back in ISO-2022-JP, an encoding the English-language version of the page warned you about in advance. That is about as locked to one country as a web service gets. It reads clearly from here anyway, because nearly everything worth knowing about Senrigan is a number or a line of configuration, and those travel.

~ the rap sheet ~

~ the whole machine ~

The front page stated the size of the index in a single sentence: 1.27 million URLs extracted from 510,000 pages in the .jp domain. The search form capped results at 32 by default and let you raise that or lift the cap entirely. One switch dropped other URLs sitting in the same directory as a result you had already seen, and a protocol selector was there because in 1996 a Japanese user hunting for software was as likely to want an FTP path as a web page.

One option on that form is worth stopping on. "Display in order of most inbound links" was there as a radio button, with the words "yes (very slow)" printed beside it. Senrigan counted who linked to what and let you sort on it, while telling you up front that asking would cost you a wait. Beside the search box sat a second tool called Trace links: you typed in a URL and it told you which pages pointed at it. A public backlink checker for a national domain, on one box, years before the word backlink meant anything.

What it did not attempt is as telling. Japanese does not put spaces between words, so a search engine has to decide where words end, and Senrigan declined to decide. The manual says so outright: there is no dictionary, so matching is done on raw character strings. Searching for かぶ would also hit かぶりもの and 知ったかぶり, while 組み立て and 組立 would not find each other at all, because to the system those were simply different strings. Tamura's advice was to type the pieces separately and let the AND do the work. It indexed titles, headings, keyword tags and the text of links, never the body of the page.

~ no way to submit a url ~

The FAQ is unusually direct about what Senrigan would not do. No manual additions, no manual edits, ever. If your page was in .jp the robot would find it eventually; if it was elsewhere, it existed in the database only as a link target. Tamura wrote that mail asking him to register a URL would not get it registered.

Another entry was aimed at Japanese webmasters of 1996, who were in the middle of a long argument about whether linking to a site without asking permission was rude. Senrigan's answer was that you could link from any page you liked without permission. Please do not write in and check first, it added, because being asked every time was the actual nuisance. Magazine screenshots and CD-ROM collections got the same shrug, with no compensation wanted.

~ the index that stopped ~

It had been growing. A Japanese search-engine news log put Senrigan at about 220,000 records in July 1996, so the jump to 1.27 million URLs was real work done over the following year. Then it stopped. That number was on the front page when the Internet Archive first captured the site in May 1997, it was on the page in February 1998, it was still there in December 1998, and it was still there in the last capture before the end, taken in February 1999. Across twenty-one months of archived evidence the index did not grow by a single page.

Somewhere in that stretch a line appeared above the search box, and it is in every capture from February 1998 onward: 最近忙しくてデータの更新ができない状況です。申しわけありません。 I have been busy lately and have not been able to update the data. I apologize. Inside the FAQ, a question about how often the index refreshed had been commented out rather than deleted, so the old answer still sits in the HTML where anyone can read it. About a month and a bit.

For the last stretch of its life the first robot search engine in Japan was answering queries out of a snapshot nobody was refreshing, under a standing apology from the one person responsible for it. By then Tamura had finished at Waseda and gone to IBM's Tokyo research laboratory, where he worked on the XML parser that became Xerces and on XML digital signatures. He is at Google now, and has spent years on the rendering engine inside Chrome. The crawler stayed where it was.

~ what the japanese web was linking to ~

Senrigan published a file called "The link ranking 1000 from .jp domain", a list of 1,006 URLs ranked by how many .jp pages pointed at them. It is raw counting rather than anything recursive, and the gzipped copy still carries its build date of August 1996. The order is not what you would guess. Top of the list is Yahoo in America with 1,984 inbound links, ahead of the Asahi newspaper on 1,130. Third, on 1,059, is the search page of Waseda University's own information science department, which is where Senrigan had started.

NTT's corporate pages take six separate entries inside the top thirty, which says something about how much of the early Japanese web was signposted by the phone company. Project Gutenberg, hosted at a university in New Zealand, sits at five. Yahoo! Japan is at thirteen on 668 links, roughly a third of what the American Yahoo pulled out of Japanese pages. ODiN, the rival crawler running at the University of Tokyo, is at fifteen. AltaVista scrapes in at thirty-two, one place above a Sony search page.

~ the goodbye in two languages ~

The page that replaced it is 682 bytes. The logo stays at the top, 444 pixels wide, a blue eye drawn in outline next to three black characters. Under it sit two sentences. The Japanese one says Senrigan finished its service as of March 1999 and thanks you for having used it. The English one was put there for whoever had been arriving from outside: "Senrigan" was closed in March, 1999. Thank you so much for visiting our Web site. Then the company name, ASCII Corporation, aligned to the right.

No day is named in either sentence, and ASCII does not say why. The only cause anyone put in writing came from outside the company: a Japanese search-engine news log, posting the closure that April, said Senrigan had given up the leading role when goo appeared. That is one observer's reading rather than ASCII's. The timing behind it is real enough: goo opened in March 1997, the same month Senrigan was moved into ASCII. The log signed off by thanking Tamura for the long run.

ASCII had already closed ASCII Net, its own dial-up service, in August 1997, eighteen months before this. The 682-byte goodbye then sat on senrigan.ascii.co.jp without a single edit until at least August 2000, the eye still at the top of it, with nothing underneath it left to search.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »