Privacy and measurements
How first-party catalog interaction reports work, what they measure, and how to turn them off. No advertising identifiers or third-party analytics services are used by this feature.
This browser's preference
This switch stores only a local on/off preference, not an identifier. It applies to future reports in this browser on this site and does not remove earlier aggregated counts. Do Not Track (DNT) and Global Privacy Control (GPC) also disable reports, regardless of this switch. Clearing browser storage resets this preference. Core browsing and copying still work.
Reading this browser's preference...
What the collector receives
A report identifies a published server or skill by type and slug, and either a visible detail-page view or a successful Clipboard API copy. Copy reports include the selected client and, when relevant, platform or install scope. The server assigns the time. Copied text, commands, credentials, environment values, account IDs, URLs and referrers are not included. The reporting request omits cookies and referrers, does not set analytics cookies, and is not retried. Server rendering and prefetching alone do not send view reports.
Limited network-based deduplication
The server derives temporary pseudonymous fingerprints from a network-caller key using a daily random HMAC salt. The analytics file does not store raw IP addresses or user-agent strings. Identical reports are counted at most once per 15 minutes for the same network caller, entry and client. Changing scope or platform does not count as a new copy in that interval. Shared networks can undercount different people; changing networks or sending forged requests can affect counts. Without explicitly configured trusted proxy forwarding, all callers share a conservative bucket. Capacity limits can also drop reports. This is not a unique-person identifier or a fraud-proof popularity measure.
Storage and retention
Daily aggregate counters cover the last 30 UTC calendar days, including today, in a separate private file. The public projection excludes older rows immediately. Expired deduplication fingerprints, one-minute rate-limit buckets and old daily rows are removed on a subsequent accepted write; the random salt rotates on the first accepted write of a new UTC day. Still-live fingerprints retain the previous salt for the remainder of their 15-minute window. An idle file can therefore retain expired records until traffic resumes. Operational backups can retain earlier snapshots under their separate retention policy. Published pages expose only aggregate counts, measurement windows and coverage, never the salt or request fingerprints.
What these numbers do not prove
A copy report means the browser reported successful copying, not that a command was run or a server installed. A view is a hydrated, visible detail-page report, not a unique visitor. Browser privacy settings, disabled JavaScript, blocked requests and shared-network deduplication reduce coverage. Missing reports remain unknown, not measured zero. Site reports are kept separate from GitHub evidence, historical verified snapshots and package downloads. Catalog replacements do not inherit the previous entry's counters.
Other site data
This page describes the interaction collector, not every hosting or security log. Infrastructure may process request metadata separately. Sign-in uses separate session and OAuth cookies; submissions and saved stacks are explicit account features with separate private storage. Disabling interaction reports does not disable authentication or those features. Never submit private API keys or other secrets as public catalog metadata.