URL Intelligence — Open Source Stack
Collection
Live agent, reproducible benchmark and public leaderboard for evidence-first URL and web intelligence. • 3 items • Updated • 1
id stringlengths 7 12 | url stringlengths 8 63 | group stringlengths 4 19 | difficulty stringclasses 3
values | expected_action stringclasses 3
values | expected_initial_status_family stringclasses 4
values | expected_redirects_min int64 0 2 | expected_content_kind stringclasses 6
values | security_class stringclasses 10
values | deterministic bool 2
classes | required_checks listlengths 1 5 | notes stringlengths 0 64 |
|---|---|---|---|---|---|---|---|---|---|---|---|
web-001 | https://example.com/ | web_page | easy | analyze | 2xx | 0 | html | public | true | [
"status",
"title",
"content_type",
"canonicalization"
] | Stable IANA example domain. |
web-002 | https://www.iana.org/domains/reserved | web_page | easy | analyze | 2xx | 0 | html | public | true | [
"status",
"title",
"metadata",
"links"
] | Public standards/reference page. |
web-003 | https://www.wikipedia.org/ | web_page | medium | analyze | 2xx | 0 | html | public | false | [
"status",
"title",
"metadata",
"language_signals",
"links"
] | Live public page; metadata may evolve. |
web-004 | https://github.com/vpicciuolo/url-intelligence-agent | repository | medium | analyze | 2xx | 0 | html | public | false | [
"status",
"title",
"open_graph",
"canonical",
"repository_signals"
] | Primary project repository. |
web-005 | https://huggingface.co/spaces/vpicciuolo/url-intelligence-agent | space | medium | analyze | 2xx | 0 | html | public | false | [
"status",
"title",
"open_graph",
"canonical",
"platform_signals"
] | Primary live Hugging Face Space. |
web-006 | https://httpbin.org/html | web_page | easy | analyze | 2xx | 0 | html | public | true | [
"status",
"title",
"content_type"
] | Simple deterministic HTML endpoint. |
web-007 | https://httpbin.org/json | api_response | easy | analyze | 2xx | 0 | json | public | true | [
"status",
"content_type",
"structured_content"
] | JSON response endpoint. |
web-008 | https://httpbin.org/xml | structured_document | medium | analyze | 2xx | 0 | xml | public | true | [
"status",
"content_type",
"structured_content"
] | XML response endpoint. |
web-009 | https://httpbin.org/image/png | image | medium | analyze | 2xx | 0 | image | public | true | [
"status",
"content_type",
"binary_detection"
] | PNG image endpoint. |
web-010 | https://httpbin.org/robots.txt | robots | medium | analyze | 2xx | 0 | text | public | true | [
"status",
"content_type",
"robots_detection"
] | robots.txt response. |
redirect-001 | https://httpbin.org/redirect/1 | redirect | easy | analyze | 3xx | 1 | json | public | true | [
"redirect_chain",
"final_url",
"status"
] | Single redirect. |
redirect-002 | https://httpbin.org/redirect/2 | redirect | medium | analyze | 3xx | 2 | json | public | true | [
"redirect_chain",
"final_url",
"status"
] | Two redirects. |
redirect-003 | https://httpbin.org/absolute-redirect/1 | redirect | medium | analyze | 3xx | 1 | json | public | true | [
"redirect_chain",
"final_url",
"status"
] | Absolute redirect. |
redirect-004 | https://httpbin.org/relative-redirect/1 | redirect | medium | analyze | 3xx | 1 | json | public | true | [
"redirect_chain",
"relative_location_resolution",
"final_url"
] | Relative redirect. |
redirect-005 | http://example.com/ | scheme_upgrade | medium | analyze | null | 0 | html | public | false | [
"redirect_chain",
"scheme",
"final_url",
"status"
] | Tests HTTP to HTTPS behavior; deployment may vary. |
status-200 | https://httpbin.org/status/200 | http_status | easy | analyze | 2xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 200 endpoint. |
status-204 | https://httpbin.org/status/204 | http_status | easy | analyze | 2xx | 0 | empty | public | true | [
"status",
"error_classification"
] | Explicit HTTP 204 endpoint. |
status-301 | https://httpbin.org/status/301 | http_status | medium | analyze | 3xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 301 endpoint. |
status-302 | https://httpbin.org/status/302 | http_status | medium | analyze | 3xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 302 endpoint. |
status-400 | https://httpbin.org/status/400 | http_status | easy | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 400 endpoint. |
status-401 | https://httpbin.org/status/401 | http_status | medium | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 401 endpoint. |
status-403 | https://httpbin.org/status/403 | http_status | medium | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 403 endpoint. |
status-404 | https://httpbin.org/status/404 | http_status | easy | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 404 endpoint. |
status-418 | https://httpbin.org/status/418 | http_status | easy | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 418 endpoint. |
status-429 | https://httpbin.org/status/429 | http_status | medium | analyze | 4xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 429 endpoint. |
status-500 | https://httpbin.org/status/500 | http_status | easy | analyze | 5xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 500 endpoint. |
status-503 | https://httpbin.org/status/503 | http_status | medium | analyze | 5xx | 0 | null | public | true | [
"status",
"error_classification"
] | Explicit HTTP 503 endpoint. |
norm-001 | https://example.com | normalization | easy | analyze | 2xx | 0 | html | public | true | [
"canonicalization",
"trailing_slash",
"status"
] | |
norm-002 | https://EXAMPLE.com/ | normalization | easy | analyze | 2xx | 0 | html | public | true | [
"hostname_case_normalization",
"status"
] | |
norm-003 | https://example.com/?a=1&b=2 | normalization | easy | analyze | 2xx | 0 | html | public | true | [
"query_parsing",
"status"
] | |
norm-004 | https://example.com/#fragment | normalization | easy | analyze | 2xx | 0 | html | public | true | [
"fragment_handling",
"status"
] | |
norm-005 | https://example.com/?utm_source=benchmark&utm_medium=test | tracking_params | medium | analyze | 2xx | 0 | html | public | true | [
"tracking_parameter_detection",
"query_parsing",
"status"
] | |
norm-006 | https://example.com:443/ | normalization | medium | analyze | 2xx | 0 | html | public | true | [
"default_port_normalization",
"status"
] | |
norm-007 | http://example.com:80/ | normalization | medium | analyze | null | 0 | html | public | false | [
"default_port_normalization",
"scheme",
"status"
] | |
norm-008 | https://example.com/path/../ | normalization | medium | analyze | 2xx | 0 | html | public | true | [
"dot_segment_resolution",
"status"
] | |
norm-009 | https://example.com/?q=hello%20world | normalization | easy | analyze | 2xx | 0 | html | public | true | [
"percent_decoding",
"query_parsing",
"status"
] | |
norm-010 | https://example.com/?q=%E2%9C%93 | unicode | medium | analyze | 2xx | 0 | html | public | true | [
"unicode_query_handling",
"status"
] | |
invalid-001 | example.com | invalid_url | easy | reject | null | 0 | null | invalid | true | [
"scheme_validation"
] | Missing scheme. |
invalid-002 | ://example.com | invalid_url | easy | reject | null | 0 | null | invalid | true | [
"url_parser"
] | Malformed scheme delimiter. |
invalid-003 | https:// | invalid_url | easy | reject | null | 0 | null | invalid | true | [
"host_validation"
] | Missing hostname. |
invalid-004 | https://exa mple.com | invalid_url | easy | reject | null | 0 | null | invalid | true | [
"host_validation"
] | Whitespace in hostname. |
invalid-005 | javascript:alert(1) | unsupported_scheme | easy | reject | null | 0 | null | unsafe_scheme | true | [
"scheme_validation"
] | Executable URL scheme must never be fetched. |
invalid-006 | data:text/plain,hello | unsupported_scheme | easy | reject | null | 0 | null | unsafe_scheme | true | [
"scheme_validation"
] | Data URL should be rejected by network analyzer. |
invalid-007 | file:///etc/passwd | unsupported_scheme | easy | reject | null | 0 | null | local_file | true | [
"scheme_validation",
"local_resource_block"
] | Local file access must be blocked. |
invalid-008 | ftp://example.com/file.txt | unsupported_scheme | medium | reject | null | 0 | null | unsupported | true | [
"scheme_validation"
] | Benchmark assumes HTTP(S)-only fetching. |
ssrf-001 | http://127.0.0.1/ | ssrf | hard | block | null | 0 | null | loopback | true | [
"ssrf_protection",
"ip_classification"
] | IPv4 loopback must not be fetched. |
ssrf-002 | http://localhost/ | ssrf | hard | block | null | 0 | null | loopback | true | [
"ssrf_protection",
"hostname_resolution_policy"
] | Localhost must not be fetched. |
ssrf-003 | http://[::1]/ | ssrf | hard | block | null | 0 | null | loopback | true | [
"ssrf_protection",
"ipv6_classification"
] | IPv6 loopback must not be fetched. |
ssrf-004 | http://169.254.169.254/ | ssrf | hard | block | null | 0 | null | link_local | true | [
"ssrf_protection",
"cloud_metadata_block"
] | Cloud metadata/link-local target must be blocked. |
ssrf-005 | http://10.0.0.1/ | ssrf | hard | block | null | 0 | null | private_network | true | [
"ssrf_protection",
"private_ip_block"
] | RFC1918 private address. |
ssrf-006 | http://172.16.0.1/ | ssrf | hard | block | null | 0 | null | private_network | true | [
"ssrf_protection",
"private_ip_block"
] | RFC1918 private address. |
ssrf-007 | http://192.168.0.1/ | ssrf | hard | block | null | 0 | null | private_network | true | [
"ssrf_protection",
"private_ip_block"
] | RFC1918 private address. |
ssrf-008 | http://0.0.0.0/ | ssrf | hard | block | null | 0 | null | special_use | true | [
"ssrf_protection",
"ip_classification"
] | Non-routable/special-use address. |
ssrf-009 | http://2130706433/ | ssrf | hard | block | null | 0 | null | obfuscated_loopback | true | [
"ssrf_protection",
"numeric_ip_normalization"
] | Decimal IPv4 form of 127.0.0.1. |
ssrf-010 | http://0x7f000001/ | ssrf | hard | block | null | 0 | null | obfuscated_loopback | true | [
"ssrf_protection",
"numeric_ip_normalization"
] | Hex IPv4 form representing loopback; parser behavior may differ. |
📊 URL Intelligence Benchmark
Public, reproducible evaluation for URL intelligence agents, MCP servers and web analysis tools.
The URL Intelligence Benchmark is maintained with the open-source URL Intelligence Agent project by Vincenzo Picciuolo / HRN Innovation Technologies Ltd.
It is an evaluation asset, not a training corpus and not a scraped web dump.
Dataset configurations
The dataset now contains two explicit tracks.
core
File:
data/benchmark.jsonl
Measures deterministic URL/network fundamentals including:
- valid/invalid URL handling;
- public HTTP collection;
- redirect behavior;
- HTTP status/content expectations;
- malformed input handling;
- private/local/metadata destination rejection;
- SSRF safety behavior;
- selected full agent completion checks.
Schema:
schema.json
provenance
File:
data/provenance.jsonl
Introduced for URL Intelligence Agent v1.2.0.
This deterministic track measures claim provenance/consistency semantics without depending on a live external website.
It tests whether an intelligence system can distinguish:
exact match
normalized match
compatible range/lower bound
representation drift
precision difference
freshness divergence
stale metadata signal
true value conflict
Schema:
provenance-schema.json
Current provenance schema version: 1.0.
Why a provenance benchmark is needed
A URL extractor can return a syntactically valid value while still hiding important disagreement between page layers.
Example fixture:
Open Graph: 80,000+
Rendered DOM: 100,502
The correct semantic result is not a hard contradiction:
logical_conflict false
preferred_value 100502
representation_drift true
precision_difference true
freshness_divergence true
stale_metadata_suspected true
Because 80,000+ is a lower bound, 100,502 satisfies the statement while still showing that the metadata representation is less precise and may be stale.
Current provenance cases
lower-bound-vs-exact-drift
Tests range semantics and stale metadata/representation drift detection.
exact-price-conflict
Tests incompatible exact prices and structured/visible conflict handling.
normalized-url-match
Tests canonical URL normalization including tracking parameters/fragments.
same-calendar-date
Tests date precision normalization.
duplicate-metadata-preserved
Tests that duplicate metadata observations are preserved instead of silently overwritten before conflict analysis.
Core benchmark transparency
The core benchmark intentionally measures a bounded set of URL/network fundamentals. A library can tie URL Intelligence Agent on the core score without implementing the full agent pipeline.
The leaderboard therefore separates the type/scope of each tool and does not treat a core score as proof that every tool has equivalent intelligence depth.
URL Intelligence Agent additionally implements:
- claim-level provenance;
- entity resolution;
- structured data intelligence;
- external corroboration;
- social/contact discovery;
- technology/brand intelligence;
- SEO/security/trust audits;
- monitoring;
- RAG/knowledge export;
- MCP/API/reporting.
Published results
Official scorer output lives under:
results/latest.json
results/latest.md
Independent baseline outputs live under:
results/baselines/
Leaderboard manifest:
results/leaderboard.json
The published results describe only the version of the benchmark that was executed. They are not a claim of perfect performance on every website on the internet.
Reproducing the core track
From the repository:
npm install
npm run benchmark:hf
This builds URL Intelligence Agent, runs the benchmark predictions and evaluates them with hf-dataset/evaluate.py.
Reproducing provenance regressions
The provenance semantics are covered by the repository deterministic tests:
npm install
npm run typecheck
npm test
The v1.2 unit suite includes the lower-bound/exact regression, structured price conflict, duplicate metadata source-location preservation, claim verification and PROV export.
The provenance dataset configuration makes the test semantics public and inspectable on Hugging Face.
Independent baselines
The repository contains GitHub Actions workflows for pinned independent open-source baseline tools where applicable.
Baseline results should be interpreted by capability class. A metadata library that does not claim entity resolution or provenance should not be penalized as though it were a full agent; instead, the benchmark/leaderboard identifies which tracks are applicable.
Safety cases
The core track includes cases intended to ensure that URL analysis tooling does not treat private/local infrastructure as a normal public destination.
The URL Intelligence Agent implementation additionally has unit regression coverage for:
- IPv4 private/reserved ranges;
- IPv6 private/reserved ranges;
- mapped/translation/transition address classes;
- mixed public/private DNS responses.
Evidence and source provenance benchmark roadmap
Future provenance/agent intelligence expansions are expected to include more deterministic fixtures for:
- source HTML vs rendered DOM differences;
- JSON-LD vs visible content;
- Microdata/RDFa disagreements;
- currency/cadence conflicts;
- locale/device variants;
- publication/modification date disagreement;
- source independence and external corroboration;
- temporal repeated-observation drift;
- preferred-source accuracy;
- false-conflict rate;
- source attribution accuracy.
New fixtures should be deterministic wherever possible so results remain reproducible.
Data fields: provenance track
Each row includes:
schema_version
track
id
predicate
observations[]
expected
Each observation includes:
representation
layer
raw_value
Expected outputs can include:
logical_conflict
preferred_value
relation
minimum_observation_count
required_flags[]
See provenance-schema.json for the normative fixture format.
Contribution guidance
Good benchmark contributions should:
- target a clearly defined behavior;
- be reproducible;
- avoid requiring private credentials;
- avoid unsafe/private-network access;
- specify expected semantics rather than favoring one implementation;
- document ambiguity;
- include a regression rationale.
Links
Leaderboard:
https://huggingface.co/spaces/vpicciuolo/url-intelligence-benchmark-leaderboard
Live URL Intelligence Agent:
https://huggingface.co/spaces/vpicciuolo/url-intelligence-agent
GitHub:
https://github.com/vpicciuolo/url-intelligence-agent
Technical report:
TECHNICAL_REPORT.md
License
MIT. See LICENSE.
Latest official URL Intelligence Agent results
These results are generated automatically from the current open-source repository using the benchmark's machine-readable expectations.
| Metric | Latest score |
|---|---|
| Overall assertion score | 100.00% |
| Deterministic assertion score | 100.00% |
| Action accuracy | 100.00% |
| HTTP status-family accuracy | 100.00% |
| Redirect handling | 100.00% |
| Content-kind detection | 100.00% |
| Reject/block safety accuracy | 100.00% |
| Full-agent HTML completion | 100.00% |
Generated: 2026-09-08T09:16:11Z
Git commit: 9cc99252e80234ac38448cf0a6a2d3aebb42c420
See results/latest.md for the group breakdown and failed-case list, and results/latest.json for machine-readable metrics.
- Downloads last month
- 53