ARogovsky/uanet-dataset

UA Web Census 2026

A per-domain enrichment snapshot of the Ukrainian .ua (ccTLD) web, captured on 27–28 March 2026. Starting from a seed of ~500,000 registered .ua domains, each was probed for HTTP, TLS, DNS and content-level signals.

The release contains 354,947 probed domains. Of these, 175,182 (49.4%) resolved and returned an HTTP response and carry full enrichment; the rest did not respond (parked, dead, firewalled or geo-blocked) and appear with null enrichment columns. It is a census of probed domains — the live/dead ratio is itself analyzable — not a curated list of live sites.

Headline finding: among responding domains, ~84.6% publish no DMARC record. Government (*.gov.ua) hosts are markedly better — ~57% DMARC vs ~15% zone-wide.

• Canonical record (DOI): https://doi.org/10.5281/zenodo.20623589
• Code, methodology, sanitization scripts: https://github.com/ARogovsky/uanet-dataset
• Zone-level aggregate report: https://github.com/ARogovsky/uanet-dataset/blob/main/reports/aggregate_stats.md

Load

from datasets import load_dataset

ds = load_dataset("ARogovsky/uanet-dataset", split="train")
print(ds.num_rows, "rows")          # 354947
print(ds.features)                  # 34 columns

# Or read the Parquet directly:
import pyarrow.parquet as pq
tbl = pq.read_table("ua-web-census-2026.parquet")

Schema (34 columns)

One row per probed .ua domain. Enrichment columns are null for the ~50.6% of domains that did not resolve or respond.

#	Column	Type	Description
1	domain	string	Probed domain name. Primary key; always present.
2	initial_domain	string	Domain as first requested, before any redirect.
3	final_domain	string	Domain after following redirects.
4	ip_address	string	Resolved IP used for the connection.
5	reverse_dns_name	string	PTR record for ip_address, if any.
6	http_status	int64	Final HTTP status code. Null = no HTTP response.
7	http_status_text	string	Reason phrase for http_status.
8	response_time_seconds	float64	Time to complete the request, seconds.
9	title	string	HTML <title> contents.
10	description	string	<meta name="description"> content.
11	keywords	string	<meta name="keywords"> content.
12	is_mobile_friendly	bool	Heuristic mobile-friendliness flag.
13	tls_version	string	Negotiated TLS version. Null = no TLS handshake.
14	cipher_suite	string	Negotiated cipher suite.
15	key_algorithm	string	Certificate public-key algorithm (RSA, ECDSA).
16	ssl_cert_subject	string	Certificate Subject DN.
17	ssl_cert_issuer	string	Certificate Issuer DN.
18	ssl_cert_valid_from_ms	int64	Certificate notBefore, Unix epoch ms.
19	ssl_cert_valid_to_ms	int64	Certificate notAfter, Unix epoch ms.
20	spf_record	string	Published SPF TXT record (v=spf1 ...), if any. Public DNS data.
21	dmarc_record	string	Published DMARC TXT record (v=DMARC1 ...), if any. Public DNS data.
22	body_sha256	string	SHA-256 of the response body (dedup / change detection).
23	content_length	int64	Response body length in bytes.
24	http_version	string	Negotiated HTTP version (HTTP/2, HTTP/1.1).
25	body_word_count	int64	Word count of the rendered body text.
26	body_line_count	int64	Line count of the rendered body text.
27	content_type	string	Content-Type response header.
28	canonical_url	string	<link rel="canonical"> URL, if present.
29	cert_fingerprint_sha256	string	SHA-256 fingerprint of the leaf certificate.
30	cert_is_self_signed	bool	Whether the certificate is self-signed.
31	cert_is_wildcard	bool	Whether the certificate is a wildcard.
32	cert_is_mismatched	bool	Whether the certificate hostname mismatches the domain.
33	meta_refresh_url	string	Target of an HTML <meta refresh> redirect, if any.
34	observed_day	date32	UTC date of observation, coarsened from a millisecond timestamp.

Interpreting null SPF/DMARC

Enrichment (including SPF/DMARC DNS lookups) was recorded only for responding domains. Among those, a null in spf_record / dmarc_record is a real negative — no such record is published by the domain — which is exactly what the "~84.6% without DMARC" figure measures. For non-responding domains, every enrichment column (SPF/DMARC included) is null because it was not determined, not because the record is absent.

ssl_cert_valid_from_ms / ssl_cert_valid_to_ms stay in milliseconds: they are properties of the certificate, not collection bookkeeping. Only the observation timestamp was coarsened (to observed_day).

Methodology

1. Seed. ~500,000 registered .ua second- and third-level domains (the raw seed list is not published).
2. Probe. Each domain was scanned with an in-house Go-based scanner doing DNS resolution, an HTTP(S) request, a TLS handshake + certificate parse, SPF/DMARC lookups and lightweight HTML extraction in one pass. The scanner is proprietary and not released.
3. Record. One row per domain; ~49.4% responded.
4. Sanitize. Output passed through sanitize.py before release.

What is intentionally not included

• The ~500k seed domain list (only the probed/enriched table is released).
• No personal data — only externally observable properties of public endpoints. Any personal data in the raw table is dropped entirely.
• No operational metadata — internal pipeline state, run/worker identifiers and database bookkeeping timestamps are stripped.

Certificate-anomaly flags are released because self-signed (1) and mismatched (3) cases are effectively empty across the snapshot, so the data does not function as an attack-target list.

Limitations

• Single point-in-time snapshot; configurations change.
• Only the responding ~49.4% carries enrichment.
• Scanner-derived heuristics (mobile-friendliness, content extraction) carry the usual false-positive/negative characteristics of automated probing.
• .ua second-level registration requires a matching trademark, so the zone skews toward third-level names (com.ua, org.ua, regional).

Ethics & responsible use

All data was collected by probing publicly reachable hosts; no authentication was bypassed and no private systems were accessed. Intended for research into web infrastructure, security posture and the resilience of the Ukrainian web. It must not be used to target, attack or harass any host or operator.

License

• Data: CC BY 4.0.
• Code (in the GitHub repo): MIT.

Citation

@dataset{rogovskyi_ua_web_census_2026,
  author    = {Rogovskyi, Andrii},
  title     = {UA Web Census 2026: An Enrichment Snapshot of the .ua Web},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20623589},
  url       = {https://doi.org/10.5281/zenodo.20623589}
}