inst/extdata/README.md

In pagerankr: Modular Toolkit for PageRank Calculation

Case-study fixture: a 67-page reviews microsite, before and after

Two Screaming Frog crawls of the same small site, taken either side of a deliberate internal-linking intervention. They exist so the case-study vignette and the placement/boilerplate analyses run from package data instead of an external repository.

reviews-microsite-before/    reviews-microsite-after/
  all_inlinks.csv              all_inlinks.csv
  internal_all.csv             internal_all.csv

What the two crawls differ by

Between them, sitewide in-content links were restructured: methodology and disclosure bylines that had appeared in every page's body were moved into a semantic container, and body-content links were added to the most-orphaned pages. The point of shipping both is that the same site is measured twice, so a change in PageRank is attributable to the intervention rather than to a difference between sites.

notes/edge-weighting-model.md and notes/pagerank-behavior-field-notes.md analyze this pair; the vignette reproduces the headline before/after table.

Loading

crawl <- function(phase) {
  d <- system.file("extdata", paste0("reviews-microsite-", phase), package = "pagerankr")
  screaming_frog_bundle(
    internal = file.path(d, "internal_all.csv"),
    links    = file.path(d, "all_inlinks.csv")
  )
}

before <- crawl("before")
after  <- crawl("after")

pagerank_screaming_frog(before)

Shape

| | before | after | | --- | ---: | ---: | | all_inlinks rows | 4,440 | 4,131 | | internal_all rows | 92 | 85 | | graph-eligible (Hyperlink) edges | 3,820 | 3,599 | | nodes | 92 | 85 | | ranked pages | 74 | 69 |

Both exports are trimmed to the columns declared by sf_contract() — 18 for all_inlinks, 24 for internal_all, down from the 103 columns Screaming Frog writes by default. Link Path is retained deliberately: it is the entire signal the boilerplate detector reads, and dropping it would make the fixture unable to answer the questions it exists for. The whole fixture is ~70 KB gzipped.

The site is pseudonymous, and this is not reversible

The underlying crawls are of a real site, and everything identifying has been replaced rather than redacted:

• Hosts map to reviews-microsite.example.net (the crawled host) and reviews-microsite.example.com (the canonical target, which was never crawled). Third-party hosts become external-NN.example.org.
• Paths are relabeled to /s2/p07/-style tokens that preserve depth and sibling grouping and discard meaning. No real path segment survives.
• Anchors and alt text are regenerated from the destination's label. Only the empty / non-empty distinction is preserved, because an empty anchor is a real signal for image links.
• Titles, meta descriptions, headings and author/category fields are blanked. They carry identity as directly as URLs do.

Retained verbatim, because they are structural rather than identifying: status codes, indexability, canonical and redirect relationships, crawl depth, word counts, link counts, response times, Link Position, and Link Path — whose attribute vocabulary is ten generic component names (nav-item, table-wrap, faq-list and similar).

The mapping is deterministic — the same source reproduces this fixture exactly — but it is not stored anywhere in this repository, and it is not invertible from the fixture alone. data-raw/build-fixture.R regenerates it from the private source crawls, which are not distributed. A scrub gate in that script fails the build if any real host token or path segment reaches the output, and tests/testthat/test-fixture-scrub.R re-checks the shipped files independently.

pagerankr documentation built on Oct. 1, 2026, 5:09 p.m.