Upload README.md with huggingface_hub

README.md

@@ -0,0 +1,98 @@
+---
+license: cc-by-4.0
+pretty_name: "We Fingerprinted the Top 1,000,000 Websites — aggregate technology census by Tranco rank"
+tags:
+  - web-technology
+  - tech-stack
+  - builtwith
+  - wappalyzer
+  - tranco
+  - dataset
+---
+
+# We Fingerprinted the Top 1,000,000 Websites — open data
+
+Aggregates over a homepage technology-stack fingerprint of the **Tranco top-1,000,000 domain
+list**, snapshot **2026-08-25** (`run_id=top1m-2026-08-25-full-v2`, detector version
+`2026.10.47`). One record per domain — CDN, CMS, ecommerce platform, web server, SSL/TLS
+issuer, DNS provider, analytics, and more — detected from homepage HTML and response headers,
+aggregated into the tables below, plus a stratified ~10,000-domain row-level sample.
+
+Companion to the study **[We Fingerprinted the Top 1,000,000 Websites. The Top 1,000 Run a
+Completely Different Stack.](https://crawlora.net/blog/top-1-million-websites-techstack-2026)**.
+The same underlying census is browsable and filterable as a
+[queryable Tech Stack dataset](https://crawlora.net/datasets/techstack).
+
+## Headline findings
+
+- **The stack flips as you go down the rank list.** Enterprise bot management runs **26.3%** of
+  the top 1,000 sites by Tranco rank, falling to **9.8%** of the 100K–1M tail — a 2.7×
+  gradient. React meta-frameworks (Next.js, Nuxt.js, Astro, Gatsby, SvelteKit, Remix) fall the
+  same way, 16.4% → 6.0%. WordPress and free-tier Cloudflare CDN adoption run the *opposite*
+  direction — 10.6% → 19.8% and 20.8% → 42.2% respectively.
+- **84.7%** of the 1,000,000 domains resolve a working homepage (847,491 reachable). A
+  `reachable=false` record means the fetch itself failed — it is never counted as "zero
+  technology detected." See `METHODOLOGY.md`.
+- **Chinese-TLD (`.cn`) sites are measurably harder for a Western-signature-trained detector to
+  see.** Reachable `.cn` sites average **3.56** detected technologies vs. **7.84** for the rest
+  of the reachable universe — a 2.2× gap — and show zero detected technology at **5.09%** vs.
+  **0.54%** elsewhere (9.4×). This is a detector-coverage limitation, not a claim about the
+  underlying sites — see the "What this dataset is NOT" section of `METHODOLOGY.md` before
+  citing it.
+- **Let's Encrypt and Google Trust Services combined secure 72.4%** of every site carrying an
+  SSL/TLS issuer signature — the free-CA era, measured directly.
+- **This exact top-1K/10K/100K/1M rank segmentation is BuiltWith's paid "Market Analysis"
+  product**, $295–995/month. This release computes the same kind of cut from the open Tranco
+  census, free.
+
+## Files
+
+| File | Contents |
+| --- | --- |
+| `data/summary.json` | Run identity: `run_id`, detector version, snapshot date, total/reachable/unreachable counts |
+| `data/category-prevalence.csv` | Domains carrying a detected technology in each of 40 categories (CDN, CMS, Analytics, …) |
+| `data/cdn-vendors.csv` | Vendor share within the CDN category |
+| `data/cms-platforms.csv` | Platform share within the CMS category |
+| `data/ecommerce-platforms.csv` | Platform share within the Ecommerce category |
+| `data/web-servers.csv` | Web server share within the Web server category |
+| `data/ssl-issuers.csv` | Certificate-issuer share within the SSL/TLS category |
+| `data/dns-providers.csv` | DNS-provider share within the DNS provider category |
+| `data/rank-band-stack-inversion.csv` | The headline table — bot management / React frameworks / WordPress / Cloudflare CDN / avg tech count, split across 4 Tranco rank bands |
+| `data/tld-distribution.csv` | Domain count by TLD, top 40 + a long-tail "other" bucket |
+| `data/cn-visibility-gap.csv` | `.cn` vs. rest-of-universe reachability and detection-coverage comparison |
+| `data/analytics-tracking-load.csv` | Analytics-tag adoption, privacy-friendly-analytics share, zero-analytics rate |
+| `data/captcha-crosscheck.csv` | CAPTCHA-technology adoption, cross-validated against the separately-run Anti-Bot Adoption Index |
+| `data/sample-domains-10k.csv` | ~9,850-row stratified sample (systematic draw by Tranco rank within each of the 4 bands above) — see `data/sample-schema.md` for the exact method and field list |
+
+## Methodology (condensed — see `METHODOLOGY.md` for the full version)
+
+Homepage HTML and response headers for the Tranco top-1,000,000 domains were fingerprinted
+against a technology-signature database (CDN, CMS, ecommerce, analytics, web server, and more)
+— the same category of detection BuiltWith and Wappalyzer use. HTTP-only (no browser
+rendering); Tranco-seeded only. Collected via the [Crawlora](https://crawlora.net) web-data API.
+
+### Caveats
+
+- **`reachable=false` does not mean "no technology detected."** It means the homepage fetch
+  itself failed. Every adoption-rate figure in this release excludes unreachable domains from
+  its denominator.
+- **The rank-band table is a pinned snapshot, not a live query** — the public
+  `/datasets/techstack/*` API sorts by rank but has no rank-*range* filter, so this specific cut
+  is reproducible only from a full-census pass, not the REST API.
+- **`.cn` figures describe detector coverage, not the underlying Chinese web** — see
+  `METHODOLOGY.md`'s "What this dataset is NOT" section before citing the visibility-gap numbers.
+- **This is a frozen dated snapshot, not a live feed.** It will not reflect stack changes made
+  after 2026-08-25. For current data, query the [live API](https://crawlora.net/datasets/techstack).
+- **The row-level sample is ~1% of the full census** (9,850 of 1,000,000 domains), stratified by
+  rank band, not the full dataset — see `data/sample-schema.md` for exactly which fields are and
+  aren't included and why.
+
+## License & citation
+
+Released under **[CC BY 4.0](LICENSE)** — use it for anything, just attribute.
+
+> Crawlora (2026). *We Fingerprinted the Top 1,000,000 Websites — aggregate technology census by
+> Tranco rank.* https://github.com/Crawlora-org/techstack-census-data (CC BY 4.0)
+
+See [`CITATION.cff`](CITATION.cff) for machine-readable citation metadata, and
+[`CHANGELOG.md`](CHANGELOG.md) for the refresh history and cadence.