Direct answer: Rechecking the same sites 16–38 hours later: all 25 homepages returned 200 and all 13 sitemaps returned 200 with identical bytes; copyright year, latest lastmod and 2026 page-date hits changed zero times, and the three sites where sitemap and footer disagree stayed that way. Yet 14 of 25 homepage hashes changed and 4 changed length. Stable fields do not mean unchanged content.
Study 22 was a cross-section; this one adds a time axis: same rules, same UA, same URLs, parsed again a day later. The goal is not to grade sites but to see how stable these “update evidence” fields are over a short window.
Method: one rule set, two parses
Baseline: homepage cache from 2026-10-31 22:41 to 11-01 07:40 and raw sitemaps from 11-01 20:41–20:46. Recheck: 2026-11-02 12:42–12:47. Each site's robots.txt was fetched first and evaluated for LingyiResearchBot; only the homepage and sitemap were then requested. At most three requests per site, up to three retries on non-200, five seconds between requests, 20-second timeout, no JavaScript, no login, no form submission. Both parses used UTF-8 with errors=replace and the same regular expressions.
| Layer | Sites | Window | Compared |
|---|---|---|---|
| Homepage | 25 | ~28–38 hours | Copyright year, 2026 date hit, response length, SHA-256 |
| Sitemap | 13 | ~16 hours | HTTP status, latest lastmod, SHA-256 |
The two windows differ: homepage baselines run from late 10-31 to the morning of 11-01, sitemap baselines from the evening of 11-01. Denominators are 25 and 13, so they cannot be merged into one change rate.
Result: zero field changes, 14 hash changes
| Signal | Baseline | Recheck | Change |
|---|---|---|---|
| Copyright year 2026 | 3 | 3 | 0 |
| Copyright earlier than 2026 | 12 | 12 | 0 |
| No copyright year | 10 | 10 | 0 |
| lastmod in 2026 (of 13) | 11 | 11 | 0 |
| 2026 date on homepage | 13 | 13 | 0 |
| Conflict: new sitemap, old footer | 3 | 3 | 0 |
| Homepage hash differs | — | — | 14 / 25 |
| Homepage length differs | — | — | 4 / 25 |
The three conflicting sites are still E01, E02 and E14; none moved its footer to 2026 in this window. The maximum sitemap lastmod did not advance either: all 13 files were byte-identical to the baseline.
Fourteen hash changes: dynamic content, not update evidence
The response SHA-256 changed for E02, E03, E04, E11, E12, E13, E14, E18, E20, E22, E24, E26, E27 and E28; E14, E24, E26 and E28 also changed length. Usual causes are timestamps, CSRF tokens, rotating recommendations, CDN injection, A/B tests or live data on the page. A hash change only says the bytes are not the same copy, not that the body changed.
The reverse also holds: E01, E06, E08, E09, E16, E17, E19, E23, E25, E29 and E30 were byte-identical, which does not prove the content was unchanged — it may simply have carried no dynamic output this time. Neither hashes nor signals prove “updated” or “not updated” on their own.
Thirteen identical sitemaps: stable file, not “no new pages”
All 13 parseable sitemaps returned 200 with the same SHA-256, and their maximum lastmod did not move. They may be static files that are never rebuilt, or the sites genuinely published nothing. Bytes cannot tell the two apart, which is why lastmod only proves what the file says.
What this means for acceptance
- Field stability is a short-window fact, not update evidence. Three signals can stay frozen for a day without saying anything about maintenance.
- A hash change is not a substantive update either. Fourteen homepages changed hash while no structural signal moved; body changes need structural fields or a manual diff.
- Record the window. Homepage ~28–38 hours, sitemap ~16 hours; the 25 and 13 denominators must not be merged.
- List conflicts separately. Do not let the footer year overwrite the sitemap field, or the reverse.
- Check robots before rechecking. Here 18 robots files returned 200 and 7 returned 404 (absent); none disallowed
LingyiResearchBot.
Limits
- Single baseline: the baseline is itself one measurement, and its error carries over.
- Uneven windows: homepage and sitemap baselines differ and must be reported separately.
- Homepage and sitemap only: inner pages and newly added URLs are out of scope.
- Byte noise: a dynamic token is enough to change a hash.
- Short window: 16–38 hours says nothing about daily, weekly or monthly stability.
- Non-random sample: 25 companies, a pilot; no claim about the industry.
Data and reproduction
Anonymised data: equipment-freshness-recheck-24h-2026.csv (25 rows: sample_id, robots_status, page_status, before/after lengths, page_hash_changed, before/after copyright years, lastmod and 2026-date hits). The repository keeps docs/research/2026-11-freshness-recheck-24h.csv.
Comparison logic: scripts/freshness_compare.py; unit tests: scripts/test_freshness_compare.py (15 tests, no network). Parsing rules reuse scripts/freshness_probe.py (14 tests). Run:
python -m unittest scripts.test_freshness_compare
Related: the freshness-signal pilot (study 22 cross-section), the freshness checker (same rules in the browser), the 12/24-hour recheck (robots, sitemap, contact pages), and the update-date fields pilot (JSON-LD and meta/time hits).
Sources and definitions
- Baseline: study-4 homepage cache (2026-10-31 to 11-01) and study-10 raw sitemaps (2026-11-01).
- Recheck: 2026-11-02 12:42–12:47, UA
LingyiResearchBot/1.0, five seconds between requests, 20-second timeout. - Parsing: UTF-8 with
errors=replace; copyright, date and lastmod rules match study 22.