Library Research Guides Catalogue Sources: 71 Calculation Tools in 35,640 Outbound Links

A survey conducted on 18 August 2026 across 368 English-language university library installations on the LibGuides platform extracted 35,640 outbound links from 2,193 research guides covering calculation-heavy subjects. Of those links, 71 pointed to a calculator, converter, or spreadsheet tool. Five of the 71 were dead, and all five were assignment planners rather than calculators. The result places a concrete number on an assumption that circulates without evidence: that academic library guides are a productive place to find where computational tools are cited and recommended.

What the survey set out to measure

Library research guides occupy a specific position in the web graph. They are maintained by subject specialists, updated on institutional schedules, and linked from university domains that carry substantial authority. They are frequently proposed as a discovery surface for tools, on the reasoning that a subject librarian curating a page on structural engineering or clinical nutrition would naturally include the calculators practitioners use.

The survey tested that reasoning directly. Two questions were posed. First, at what rate do these guides link to calculation tools as opposed to information sources. Second, among the links they do carry, what proportion have decayed, and what kind of resource fails most often.

Seven subject areas were selected because each has an established body of routine computation: statistics, construction, HVAC, finance, logistics, nutrition, and electrical work. A guide entered the sample if its title or subject classification placed it in one of those areas.

The funnel

The compression between rows carries most of the information. The 35,640 raw links collapse to 10,747 unique URLs, a deduplication ratio of 3.3 to 1, which means the average distinct resource appears in slightly more than three places. Those 10,747 URLs resolve to 3,600 domains, so the average domain is cited three times across the corpus.

The final row is the finding. Tool links constitute 0.199 percent of extracted links, approximately two in every thousand. Distributed across 2,193 guides, that is one tool link for every 31 guides in subjects defined by their computational content.

Method

Guides were enumerated from each installation's sitemap.xml. No search engine was used at any stage. This choice was deliberate and it shapes what the numbers mean: search indexes reflect crawl priority and popularity, while a sitemap reflects what the institution itself publishes. Enumeration by sitemap produces a census of a platform rather than a sample of what ranks well.

Links were extracted from guide bodies, with navigation chrome and platform boilerplate excluded. Classification as a tool link required the anchor text or the URL path to contain a tool indicator: calculator, converter, spreadsheet, and their common variants. This is a lexical test and it is conservative in one direction and permissive in another, a point taken up under limitations.

Status testing used HEAD requests, falling back to GET where the server returned a method-not-allowed response or otherwise failed to implement HEAD. Only 404 and 410 were counted as death. Status codes 403, 406, and 429 were treated as robot blocking and excluded from the dead count, since they describe the crawler's reception rather than the resource's existence.

Every candidate failure was then rechecked with a browser user-agent string. That second pass discarded 24 false positives. The clearest illustration is app.dimensions.ai, which returns 404 to an automated client and 202 to a browser. Without the recheck, the reported decay rate would have been inflated by roughly a third of its candidate volume.

Separately, 67 domains had no DNS record at all. These are unambiguous failures and require no HTTP negotiation to confirm.

Finding one: the corpus catalogues sources

The dominant content of these guides is bibliographic. Outbound links resolve overwhelmingly to databases, journal platforms, publisher sites, government statistical agencies, standards bodies, and professional associations. A guide on HVAC points to ASHRAE and to engineering databases. A guide on nutrition points to federal dietary data and to indexed literature. The librarian's function in these pages is to route a reader to authoritative material, and the pages perform that function consistently.

Calculation tools sit outside that remit. A load calculator, a unit converter, or an amortisation spreadsheet is an instrument of practice rather than a source to be cited, and the guides reflect the distinction. The 71 tool links across 35,640 are the quantitative expression of a professional boundary that librarians observe in practice.

The practical consequence is directional. Any project attempting to map where calculation tools are recommended on the web should treat LibGuides installations as low-yield and look to other page types: course pages, professional association resource lists, trade publication roundups, and vendor documentation.

Finding two: broken links point inward

Among the guides carrying at least one dead link, 172 in total, half of the broken links pointed to infrastructure belonging to the institution that published the guide. These were proxy servers, catalogue endpoints, internal request forms, and authentication paths. The recurring example is proxy.lib.unc.edu, an EZproxy-style prefix that becomes invalid when access configuration changes.

This inverts the usual framing of link rot. The standard concern is that the external web moves beneath a curated page. The measurement here shows that the institution's own systems move at a comparable rate, and that a guide can break without any external publisher doing anything. Proxy migrations, catalogue replacements, and form retirements all invalidate links inside the library's own control.

The remediation implication is favourable. Internal breakage is fixable by the organisation that caused it, usually in bulk, since a proxy prefix change affects every link constructed from that prefix.

Finding three: template links inflate naive counts

A single dead URL can appear many times across a campus because guides are cloned. The OWASP page at owasp.org/index.php/Cross-Site_Request_Forgery_(CSRF) is dead because the project migrated off MediaWiki, and it appeared in six guides at one institution. Counted per link instance, this registers as six failures. Counted per unique URL, it is one failure requiring one correction.

Any decay statistic that omits deduplication overstates both the problem and the work. The 3.3 to 1 ratio between extracted links and unique URLs in this corpus sets the approximate scale of that overstatement. Remediation planning should be organised around unique targets, since a single edit to a shared template or a single find-and-replace resolves the whole cluster.

Practical implications for guide maintenance

Three operational points follow from the data.

Dead links in research guides are ordinary. They should be treated as a maintenance category with a scheduled cadence rather than as an incident.

Audits should begin with institutional infrastructure. Given that half of the broken links in affected guides pointed to the library's own systems, a review of proxy prefixes, catalogue URL patterns, and internal form endpoints will clear a large share of failures before any external checking begins.

Link checkers that resolve only the root domain will miss the most common form of external decay. The instructive case is energy.gov, which returns 200 at the home page and 404 at /energysaver/. A checker testing domain reachability reports the link as healthy while the cited section has been removed. Full-path testing is required, along with browser user-agent verification to suppress the false positives that cost this survey 24 candidate failures, and treatment of 403, 406, and 429 as inconclusive rather than as death.

Limitations

The sample is English-language only. Of 469 institutions in the public directory, 368 were analysed, and the 101 excluded installations may distribute tool links differently. Practices in other language communities are unmeasured here.

The LibGuides resultCountEstimate field saturates, which is why sitemap enumeration was used instead of the platform's own search interface. Where a search count would have been convenient, it caps rather than reports, and no total derived from it can be trusted at scale. The sitemap route avoids this but inherits whatever omissions exist in each institution's sitemap generation.

Tool classification is lexical. A calculator linked with anchor text that names the publisher rather than the function will be missed, so 71 is a lower bound on the true count. The direction of error does not threaten the conclusion, since the observed rate would have to be understated by more than an order of magnitude to change the interpretation.

The survey covers LibGuides installations exclusively. University pages built on other platforms, departmental sites, course management systems, and personal faculty pages are outside its scope, and nothing measured here describes them.

Finally, the survey is a single snapshot taken on 18 August 2026. It measures state, not rate of change, and no longitudinal claim about decay speed can be drawn from it.