Direct answer: from the 2024 Su–Xi–Chang first-set equipment list, 16 Changzhou companies had an identifiable official website. After excluding two failed or undersized pages, 14 homepages remained. Median HTML is 54,666 bytes; median visible-text share is 6.8% (0.6%–20.0%); inline SVG appears on only 4 of 14 pages, with a median of 0% and just 1.3% of total bytes. The median script share is 10.7%, and one page is 91.6% inline style.
This differs sharply from the 36 article pages in study 89: there, median visible text was 4.0% and median inline SVG 30.8%. Within the same company frame, different page types put their HTML bytes in different places — article pages lean on SVG, homepages on style and script.
Method: a new homepage sample, same byte splitter
The frame is the 30 Changzhou companies on the 2024 Su–Xi–Chang first-set equipment list (the same parent frame as the article-page studies, but this round takes the 16 companies not already in the 36-page sample). On 2026-11-07 each homepage was fetched once with a browser UA, 1.5 s between sites and a 25 s timeout; HTTP non-200 and pages under 5,000 bytes were excluded, leaving 14.
The byte buckets follow study 89: remove comments, script, style and inline SVG in order, then measure the remaining tags and text nodes; UTF-8 bytes, percentages over that page's HTML bytes.
| Item | Rule | Boundary |
|---|---|---|
| Homepage | The site root or the homepage recorded for the company | Inner pages excluded |
| Text | Visible text after removing script/style/SVG/comments/tags | Includes navigation and footer, not body copy |
| Inline SVG | The whole <svg>…</svg> block | Icons and illustrations not separated |
| Excluded | Non-200 or HTML <5,000 bytes | E21 502, E20 1,329 bytes |
Result: text 6.8%, median SVG 0%
| Bucket | Median share | Min–max |
|---|---|---|
| Visible text | 6.8% | 0.6%–20.0% |
| Inline SVG | 0.0% | 0%–7.2% |
| Markup | 47.3% | 2.2%–62.7% |
| Whitespace/other | 15.0% | 0%–56.1% |
| Script | 10.7% | 0.6%–94.2% |
| Style | 0.3% | 0%–91.6% |
| Comment | 1.1% | 0.1%–10.9% |
Five of 14 pages are below 5% text and nine are below 10%. Median HTML is 54,666 bytes and the largest is 294,026 bytes; the 14 pages total 1.31 MB, of which text is 69 KB and inline SVG only 17 KB.
Compared with article pages: a different centre of gravity
| Metric (median) | 14 homepages | 36 article pages (study 89) |
|---|---|---|
| HTML bytes | 54,666 | 106,260 |
| Visible-text share | 6.8% | 4.0% |
| Inline-SVG share | 0.0% | 30.8% |
| Script share | 10.7% | 4.1% |
| Style share | 0.3% | 0.0% |
| Markup share | 47.3% | 39.2% |
Two gaps stand out: article pages have a median 30.8% inline SVG while homepages have 0%; homepages have a median 10.7% script share while article pages have 4.1%. This is not a ranking of quality; page types differ. Article pages reuse many icon/decorative SVGs, while homepages are more likely to write styles and scripts directly into the HTML.
A few homepages are style/script shells
| Site | HTML bytes | Text | Script | Style |
|---|---|---|---|---|
| E25 | 294,026 | 0.6% | 0.6% | 91.6% |
| E18 | 16,124 | 2.3% | 94.2% | 0.0% |
| E24 | 158,886 | 2.1% | 25.4% | 36.0% |
| E17 | 126,998 | 7.0% | 5.5% | 35.6% |
E25 is 294 KB, 91.6% of it an inline style block; E18 is 16 KB, 94.2% of it a script block. These pages may inline template CSS/JS on every page. This round records bytes only and does not judge whether either should be externalised or cached.
What this means for acceptance
- Measure by page type. Homepage and article-page composition differ widely; a homepage audit cannot stand in for article pages, or vice versa.
- On homepages, look at inline CSS/JS. The outliers here are dominated by style or script; article pages are dominated by SVG.
- Do not extrapolate the SVG finding. The SVG results from studies 89/91 belong to article pages; only 4 of 14 homepages contain SVG.
- Text share is not a quality score. Homepage text includes navigation, product names and footer; it is not body copy.
- Check compressed size too. This round splits raw HTML; gzip/brotli transfer was not measured.
Limits
- 14-page pilot: two of 16 identifiable homepages were excluded (E21 502, E20 1,329 bytes); this is not the full 30-company sample.
- Selection bias: only companies with an identifiable official site; four without one are absent.
- Single snapshot: one fetch on 2026-11-07; no redesign recheck.
- HTML only: images, external CSS/JS, fonts and HTTP headers are excluded; compressed size is not measured.
- No quality judgement: inline style/script may be a template choice; it is not automatically poor performance.
- Pilot: 14 pages are not an industry estimate; sites are anonymised.
Data and reproduction
Anonymised data: equipment-homepage-html-weight-2026.csv (14 rows × 24 columns: status, HTML bytes, text, script/style/SVG/comment/markup/whitespace bytes and shares, tag counts, class-attribute bytes, error). In-repo copy: docs/research/2026-11-homepage-html-weight.csv.
Fetch and parse: scripts/homepage_html_weight_probe.py (reuses the byte splitter in article_html_weight_probe.py); tests: scripts/test_homepage_html_weight_probe.py (no network, 5 tests: SID=URL parsing, text/SVG, script/style separation, empty page). Run:
python -m unittest scripts.test_homepage_html_weight_probe
python scripts/homepage_html_weight_probe.py --site E01=https://example.com/ --out homepage.csv
Related: the article HTML weight composition pilot, the article HTML compression pilot, and the article inline-SVG duplication pilot.
Sources and definitions
- Frame: 2024 Su–Xi–Chang first-set equipment list, 30 Changzhou companies; this round uses the 16 with an identifiable site and not already in the 36-page article sample.
- Fetch: 2026-11-07, browser UA, 1.5 s between sites, 25 s timeout, one homepage request each.
- Split: study-89 method; UTF-8 bytes; non-200 and <5,000-byte pages excluded.