Article-Page Dates Expanded: 8 of 41 Pages Declare dateModified, All on 2 Sites

Direct answer: Expanding the sample to five pages per site: 12 sites and 46 URLs, 41 returned 200 and 5 returned 404 (all on E11). Among the 41 valid pages, only 8 declare dateModified and datePublished, all from 2 sites: E09 5/5 and E22 3/5; the other nine sites contributed 31 pages with zero. meta, <time datetime> and visible “last updated” remain zero; all 8 dated pages match their sitemap lastmod day.

The first version took 1–2 pages per site (21 URLs) and found 3/19. This version takes five equidistant pages to test whether “article pages rarely carry update dates” holds, and whether dates cluster on a few sites.

Method: five equidistant pages per site

For the 8 sites with a usable sitemap, URLs whose path contains news / news_xq / newsinfo / news_list / article / blog and ends in a numeric .html are sorted by lastmod newest-first, and five equidistant pages are taken (positions 0, 25%, 50%, 75%, 100%). For the 4 sites without a sitemap, the first link matching the same rule on the cached homepage is used. The result is 12 sites and 46 URLs.

Before fetching, each site's robots.txt is evaluated for LingyiResearchBot; one request per URL, up to three retries on non-200, five seconds apart, 20-second timeout, no JavaScript, no login, no form submission.

Result: 41 valid pages, 8 with dates

The dates sit on two sites

This is not “19.5% of pages generally carry a date” but “two site templates emit the date and the other nine emit none”. A page-level rate averages away the site difference; acceptance should look per site.

Eight dates: spread across years, same day as the sitemap

All 8 dateModified values match their sitemap lastmod on the same day; datePublished ranges from 2015 to 2026. When the fields exist, the two layers align — but this is still a page claim, not a record of body changes.

Against the first 21-page version: the same conclusion

The rates are close and the site count is identical: E09 and E22 both times. Expanding did not change the conclusion; it made “two templates write dates” clearer.

What this means for acceptance

1. Count per site, not just per page. 8/41 looks like a fifth, but it is 2/11 sites and zero everywhere else.
2. Record 404s separately. All five are E11; a URL in the sitemap does not mean the page still exists.
3. Check type and source when a date exists. All 8 are Article and match the sitemap day — the strongest alignment seen so far.
4. Do not rely on meta / time / visible updates. None of the 41 pages had them.
5. Expanding the sample reveals structure. The same two sites appear under both selections, which points to template-level differences.

Limits

• Equidistant is still not random: five pages by lastmod order cannot represent a whole site.
• List pages may be included: URL shape cannot guarantee a body page.
• Uneven sites: E09 and E22 contribute 5 each while nine sites share 31 pages; the rate is driven by the two.
• Five 404s: all from E11, which may understate other sites' validity.
• 12-site pilot: no industry claim.

Data and reproduction

Expanded data: equipment-article-date-expanded-2026.csv (46 rows: selection mode, URL, status, JSON-LD blocks and validity, dateModified / datePublished, meta, time, visible update, has-any-date). The repository keeps docs/research/2026-11-article-date-expanded.csv. The first 21-row sample remains in 2026-11-article-date-pilot.csv.

Summary: scripts/article_date_summary.py; unit tests: scripts/test_article_date_summary.py (6 tests). Parsing reuses scripts/date_fields_probe.py (20 tests). Run:

python -m unittest scripts.test_article_date_summary

Related: the homepage update-date fields pilot (1/25), the five-source date matrix (9 sites in conflict), and the freshness-signal pilot.

Sources and definitions

1. schema.org dateModified / datePublished: schema.org.
2. sitemap lastmod: sitemaps.org protocol (W3C Datetime).
3. Sampling: 2026-11-02 17:2x Asia/Shanghai, UA LingyiResearchBot/1.0, five seconds between requests.