Less obvious, but more exciting, is the availability of ranked retrieval, which permits one to enter terms that may not exactly match those in any particular index and still retrieve highly relevant materials. This is only available at the "UCB Physical Sciences" site and the "UC Berkeley Digital Library" site, which use the Cheshire II server and search engine.
Even less obvious is the recent addition of various servers accessed via the Digital Library InfoBus (a CORBA-based digital information sharing architecture) through a Z39.50 proxy. These are indicated in the information services menu as "InfoBus access to xxx" for a variety of different services. The InfoBus Z39.50 proxy at Stanford sends back the results of searches in a simplified MARC format.
Least obvious of all is the fact that the bibliographic records housed on the Cheshire II server are in SGML format. While you will see no difference between these and the standard MARC format records provided by the other hosts on the menu, Cheshire II's use of SGML means that in the future, Cheshire searches will be able locate and deliver all sorts of documents, and not just bibliographic records.
The list currently includes the UC Berkeley NSF/NASA/ARPA Digital Library of full-text environmental documents. The UC Berkeley Physical Sciences Libraries, including Astronomy, Mathematics and Statistics, Chemistry, Physics, Geology, etc., three Melvyl (UC-wide) catalogs, the libraries at Penn State University, Duke University, Carnegie Mellon University, UNC Chapel Hill, the Hong Kong University of Science and Technology, and ATT Bell Labs, the CIA World Factbook, the Government Information Locator Service, and four periodical indexes -- ABI Inform, the IAC Computer Database, the Expanded Academic Index, Inspec, and the National Newspaper Index -- which can only be accessed from Berkeley campus computers. (The preceding periodical index links lead to short descriptions of the databases in question and, indirectly, to detailed instructions for accessing them through MELVYL. These contain potentially useful information for the Web Cheshire user, even though specific command sequences and the like will not apply.)
When you retrieve records from one of these hosts, you have the option of clicking an author name or subject heading to get 10 more items by the same person or on the same topic. When you do, these will be retrieved from the same host as the original record set, even if you have changed hosts on the main search screen in the meantime. (If you want or need more than 10 items, click the button at the bottom of the record display to get 10 more; or return to the main screen, type in the author or subject, and ask for as many as you wish.)
Cheshire's non-WWW interface does not have this problem, since it adjusts index lists whenever the user chooses a new host; but HTML forms do not currently support this sort of interaction. It is possible that incorporating Java into Web Cheshire will eliminate this limitation, but it has not yet been tried. In the meantime, hosts accessible through Web Cheshire can only be searched by author, title, and subject or topic. (See below for more information on these last two.)
While most of the index selections available on the WWW version of Cheshire II will be familiar to users of online catalogs, it is worth remembering that these are not treated in exactly the same way everywhere. For example, some treat title or subject search terms as keywords and will return any item in which those terms occur in the appropriate fields. Duke University's catalog treats them as a phrase, however. That is, a title search for endangered wetlands will only return items in which those to words occur adjacent to one another in the title. Wetlands: Endangered Habitats won't be found.
As a further example, the CIA World Fact Book will get what you want if you submit the "Algeria" as a topic, author, LC subject heading, or title; but in all but the last instance you also get data on the Cape Verde Islands, the Central African Republic, and other countries.
Topic will be the one unfamiliar index. When you submit a topic search to the UCB Astr-Math-Stat (that is, the Cheshire server itself), the search engine looks in several places for terms similar to those entered, computes the degree of match, and returns a relevance-ranked list of items. Where this kind of sophisticated searching is not supported (that is, everywhere else), but multi-index exact-match searches are, the latter is used. In the remaining cases a standard search is performed against the Library of Congress (LC) subject heading index.
Needless to say, Cheshire's judgment of relevance will not always coincide with its user's, but you should generally find some useful items among the top ranked retrievals; so topic searching "UCB Astr-Math-Stat" is a good way to begin looking for materials in an unfamiliar field of inquiry. (To get a sense of this do a topic search for "local compactness." The top ranked retrievals are clearly relevant, despite the fact that the terms local and compactness don't appear. Title and subject searches for "local compactness" in Melvyl, for example, find nothing.) Avoid topic searches that include common terms -- like "astronomy" or "algebra" -- or terms that have multiple meanings, however, as these frequently lead to disappointing results.
Most servers allow one to narrow or broaden your search by choosing to "and" or "or" search terms (by means of the buttons directly above the search entry windows). There are exceptions, however. The Hong Kong University of Science and Technology catalog permits only single-index searches.
If you just want to know how many items are available at a site, you can set the number to retrieve to 0 or, equivalently, put nothing in the window at all. (Putting nothing in the starting record window, however, is considered to be a request to begin the display with record number 1.)
If you are searching one of periodical indexes, however, you may well want to use the long display format from the start, since it will include article abstracts if they are available, while the short display format will suppress them.
Cheshire tries to distinguish among these and displays diagnostic messages when it can, some of which may be cryptic. Efforts are underway to clarify the diagnostic messages, to provide diagnostics where none exist, and to sort out all the error codes returned by the various servers Cheshire communicates with.