Direct answer: Reusing the 41 article-like pages cached in study 29, a zero-request scan of visible text finds 32 pages with at least one date string (163 in total) and 17 pages with a “published / date / time” label, but not one page says “last updated”. Nine pages contain no date at all. Dates are visible; they mostly label publication, not revision.
Studies 28 and 29 reported “visible update 0”, meaning no “updated / last updated” label. This study loosens the definition: any date string counts, to see whether “no visible date” and “no visible update label” are the same thing.
Method: zero requests, cached pages
Data comes from the 41 successful pages crawled in study 29 (the valid part of 46 URLs across 12 sites), with raw HTML stored off-repository. This study makes no network request: after removing scripts, styles and comments, visible text is matched for dates such as 20xx-10-30, 20xx/10/30 and 20xx年10月30日. A date counts as labelled when a label such as “发布时间 / 发布日期 / 发布 / 时间 / 更新于 / 最后更新 / published / updated” appears within 24 characters before it.
| Field | Rule |
|---|---|
| Date string | Target year + separator + month, optional day; duplicates counted |
| Labelled date | A label within the 24 characters before the date |
| Update label | Only “更新于 / 最后更新 / 最近更新 / updated” |
| No date | No date string in visible text |
Boundary: a date may come from a news list, sidebar, copyright or body; position is not classified. Labels are read before the date only. The study does not interpret the date or infer that a page was updated.
Result: dates visible, labels say “published”
| Item | Pages (41) | Note |
|---|---|---|
| At least one date string | 32 | 163 strings; one page has 18 |
| Publish/date label | 17 | 24 labelled dates |
| Update label | 0 | No “updated / last updated” |
| No date at all | 9 | No date string in visible text |
| Label | Occurrences | Typical context |
|---|---|---|
| 时间 (time) | 12 | “时间：…” in a list or body |
| 发布时间 (published) | 10 | “发布时间：…” in the article header |
| 发布 (publish) | 1 | “发布 2026-…” |
| 发布日期 (publish date) | 1 | “发布日期：…” |
| updated / last updated | 0 | — |
Dates per page: from 0 to 18
| Dates per page | Pages | Note |
|---|---|---|
| 0 | 9 | No visible date |
| 1–2 | 15 | One article date or a few list items |
| 3–5 | 6 | Body plus sidebar or related posts |
| 6 or more | 11 | News lists or many recommendations |
A high count usually means a list or recommendation block, not several revisions of one article. The 163 strings are not 163 pieces of update evidence.
Difference from the study 28/29 definition
The earlier studies defined “visible update” as an “updated / last updated” label and scored zero; this one defines “visible date” as any date string and scores 32/41. The two do not conflict: dates are visible, but the pages do not label them as updates. Searching only for “updated” misses publication dates; treating any date as an update date misreads publication as revision.
The 8 dateModified pages: JSON-LD and visible labels do not always pair
- E09 (5 pages): JSON-LD has
dateModified, and the page also carries a “发布时间” label with dates; four pages have 2–3 strings each. - E22 (3 pages): JSON-LD has
dateModified, but no label appears before the visible dates; dates may exist without a “published” prefix. - The other 33 pages: no JSON-LD update date; some still show visible dates (publication, lists, copyright), others show none.
Structured dates, visible labels and visible dates are three different things; no two can substitute for each other.
What this means for acceptance
- Define “visible date” first. Any date string and an update label differ by an order of magnitude.
- Separate publication from revision. None of the 24 labelled dates here says “updated”.
- Count is not evidence. Lists and recommendation blocks inflate the number.
- Record JSON-LD and visible labels separately. E22 has structured dates without visible labels.
- Keep the no-date cases. Nine pages have no date string; acceptance should write “no field”.
Limits
- Cached, zero-request: pages are frozen at the study-29 crawl.
- Regex boundary: only numeric dates starting with 20xx; worded dates are invisible.
- No position classification: body, lists, sidebars and copyright all count, which may overstate body dates.
- Label window is 24 characters: a line break or greater distance hides the label.
- 41 pages across 12 sites: a pilot, not an industry claim.
Data and reproduction
Anonymised data: equipment-visible-dates-pilot-2026.csv (41 rows: sample_id, URL, date count, first date, labelled count, first labelled date and label, update-label flag, JSON-LD dateModified). The repository keeps docs/research/2026-11-visible-dates-pilot.csv.
Parser: scripts/visible_date_probe.py; unit tests: scripts/test_visible_date_probe.py (18 tests, no network). Run:
python -m unittest scripts.test_visible_date_probe
Related: the expanded article-page dates study (46 URLs, 8 dateModified), the homepage update-date fields pilot, the five-source date matrix, and the visible date checker (classify publish vs update locally).
Sources and definitions
- Page cache: 41 successful pages from study 29; raw HTML kept off-repository.
- Date formats: ISO and common slash/Chinese forms; relative times are not parsed.
- Scan: 2026-11-02, zero network requests.