Privacy and retention
Anonymous package discovery and report viewing require no email or account. Package reports, artifact identities and recorded observations are public. Do not put secrets or personal data in public package contents or output.
Request handling
The ingress uses your network address to limit abuse. Admission stores a keyed identifier that rotates daily, groups IPv6 addresses by /64, and is removed after seven days. Raw addresses are not stored in the catalog. The default deployment disables access logs and filters request details from proxy errors. There is no advertising or third-party browser analytics integration.
For successful package-version lookups, we retain the public package name, exact version, UTC day and whether current, earlier or no eligible evidence was available. Repeated lookups of the same package, version and availability count at most once per ten-minute window across all visitors. These totals include automated clients and are not visitor or unique-user counts. We store no address, cookie, user agent, referrer or search text with them. Collection is best effort, capped at 1,000 package/version/availability rows per day, and expires after 30 UTC days. Viewing evidence never submits a scan; an operator reviews missing evidence before adding packages to the separate coverage queue.
Retention
With JavaScript enabled, the website sends first-party signals for visible page views, searches, version selections, opened evidence and successful share-link copies. Only fixed action and page categories, a coarse document-referrer group, a voluntary research flag and an automation flag are sent. No search text, package name, report ID, full URL, clipboard content, cookie or visitor/session identifier is sent with these signals. We honor browser Global Privacy Control and Do Not Track signals.
These categories count at most once per shared ten-minute window across visitors and replicas, expire after 30 UTC days and are limited to 288 possible rows per day. Collection is best effort; blocked scripts, network errors and rate limits can lose signals. Research and automation flags can be incomplete or forged. An unflagged signal is not proof of a human, a conversion or a useful visit. The receiver briefly uses a rotating keyed network identifier for an in-memory abuse limit; it is discarded when the minute changes and is never stored with the activity.
We also keep aggregate ten-minute activity windows for package pages, report pages, public API reads and MCP evidence tools. We record the UTC day, the read surface, whether evidence was available, an immediate referrer group (such as search, GitHub or internal), and a voluntarily supplied research marker. We discard the full referrer URL and user agent after choosing those fixed categories. These records contain no visitor identifier, package name, report ID or browsing history, and expire after 30 UTC days. They include bots and repeat reads; they do not measure people, conversions or usefulness.
- Reports, exact locks, identities and provenance remain available indefinitely. Operators can invalidate reports and remove retained package logs after reviewing a request.
- Sanitized package logs expire after 30 days. The report remains available after log expiry.
- Sealed workspaces and installer caches are retained for at most seven days within the worker storage cap. Reproduction availability is shown separately.
- Service logs are retained for 14 days; operator and security audit events for 180 days.
- Encrypted database backups are retained for seven days. Deleted identifiers or logs can remain in those backups until they expire.
Maintainer accounts
Maintainer tools are coming soon. GitHub sign-in, repository linking, release monitoring and email alerts are not available on the public website. We will explain their data handling and account controls before they become available.
Service providers
Discovery requests public npm registry metadata. The control service may send fixed operational error names to an operator-configured Sentry endpoint. It does not send package contents, request bodies, credentials, network addresses or user profiles to error tracking.
Corrections and removal
For an incorrect report or public content removal request, open a repository issue with the report URL and reason. Send sensitive security details through the private disclosure process. Removal does not rewrite the original compatibility classification. Log removal retains the report's structured observations and provenance; invalidation marks the report as unsuitable for reuse. The operator audits the reason. Do not include sensitive content in a public issue.
Retained behavioral evidence
A report may include historical named assertions from public GitHub commits. Their source, fixture hashes, approved capabilities and results remain part of the public evidence. Registering new maintainer assertions through the website is not available yet.