Dataset License & Access Change Monitor

Monitor the public policy metadata of one Hugging Face dataset. Each successful check compares license declarations, custom license name/link, gating and disabled status against a persistent snapshot. Use this in recurring dataset inventory reviews and data ingestion alerts, without downloading dataset files.

What makes this different

A general dataset scraper returns metadata to compare yourself. This Actor maintains history per Apify user, Actor, dataset ID and monitor name, ignores download counters and timestamps, and returns typed before/after changes. Optional exact-label license preferences and an ungated preference produce structured warnings. This is a narrow metadata workflow; Hugging Face also offers native webhooks for users building their own integrations.

Input example

{"datasetId":"stanfordnlp/sst2","monitorName":"default","acceptedLicenses":["apache-2.0"],"requireUngated":true,"includeUnchanged":true}

Use an exact namespace/dataset ID, not a URL. No account token is accepted. Reuse the monitor name for history; a new name starts a baseline. Preference changes do not reset history. Schedule with Apify Tasks or Schedules for periodic checks.

Output and history

BASELINE establishes the first snapshot; CHANGED lists field names and previous/current values; UNCHANGED confirms the same policy metadata. Each emitted dataset row includes previous/current snapshots, source URL, timestamp and policyWarnings. Default key-value record OUTPUT summarizes every check, even if unchanged rows are suppressed.

Warnings: DATASET_GATED, NO_ACCEPTED_LICENSE_REPORTED, and CARD_AND_TAG_LICENSES_DIFFER. Missing card licenses or disabled flags remain unknown, not an assumed permissive state. License labels are publisher declarations. At least one matching declared label satisfies the optional label check, including a tag; this is not verification of license terms. A license link is recorded but never visited.

Failed source checks, private/incomplete metadata, invalid inputs or delivery/charging failures do not replace the last committed snapshot. A bounded best-effort lease rejects overlapping checks; avoid scheduling the same monitor concurrently because key-value storage does not provide an atomic lock. Named history storage persists until you delete it or the platform changes its retention. Changing/deleting it starts a new baseline. Output may exist before a failed charge; trust SUCCESS and committed history rather than a partial row alone.

Pricing

$0.01 per successful dataset policy check, with platform usage included. Baseline and unchanged checks are charged, including checks that suppress unchanged rows. Failed source checks are not charged. One dataset per run, one successful-check event, no separate per-row or startup charge. Set a maximum charged amount of at least $0.01 for one check.

Scope and limits

Public Hugging Face dataset metadata only. No dataset files, crawling, login, gated terms acceptance, private repositories, proxy service, bulk search, model metadata or legal compliance determination. Gating metadata does not prove your ability to access files, and an ungated declaration does not grant rights to use data. Automatic/manual gating and disabled flags are tracked when reported. One fixed API endpoint, 20-second network deadline, 2 MB response cap and same-origin redirects. A large metadata response or unsupported schema fails safely. Upstream rate limits or schema changes may require maintenance.

Built on Node.js and Apify SDK 3.7.2; native HTTPS source requests use no cookies or response cache. A known transitive http-cache-semantics advisory GHSA-ch52-4w7c-c8xp has no published patch as checked October 5, 2026. The vulnerable cross-user cache scenario is not used by this source client; this is not a claim that the whole SDK dependency tree is vulnerability-free. Dependency follow-up remains tracked.

Unaffiliated with Hugging Face. Source documentation: https://huggingface.co/docs/hub/api , https://huggingface.co/docs/hub/datasets-gated and https://huggingface.co/docs/hub/datasets-cards .