Privacy & Retention#

For the exact tables and columns this page describes, see the Data Model reference.

What's stored#

FlagForge stores player UUIDs and last-known names, never IP addresses, chat, signs, books, or inventory contents unrelated to a finding — only the numeric/contextual evidence needed to explain that specific finding.

An unexplained item gain's finding may also record the label of the last command the player ran in the two minutes before (/kit), never its arguments, so a password typed into /login is never stored. It's kept with that finding's evidence and purged with it.

What's sent#

The only request FlagForge makes on its own is the update check: an HTTPS GET of https://flagforge.szalik.dev/latest.json every 12 hours, carrying FlagForge's version in its User-Agent and nothing about players or the server. The site's host sees the server's IP address, as with any web request. update-check.enabled: false in config.yml turns it off. Discord alerts, if configured, go only to the webhook you set.

Retention#

Configured in storage.yml. Only detailed evidence (ff_findings, default 14 days) is actually enforced today, by a scheduled async purge task (interval configurable via retention.check-interval) that never blocks the main thread. retention.aggregate-statistics-days (default 90) and retention.ledger-entries-days (default 14) are parsed but not currently backed by any purge logic — there is no dedicated aggregate-statistics table, and the action ledger (ff_actions) has no expiry column and grows unbounded until manually addressed. See Known Limitations.

Manual purge#

/ff purge player <name> confirm deletes every stored finding for one player; /ff purge before <ISO-8601-timestamp> confirm deletes everything older than a date. Both require the exact literal confirm argument as a deliberate two-step safeguard, and both are audit-logged.

Audit log#

Every mutating staff action (reload, exempt, enforcement toggle, purge, check toggle) is recorded in the append-only ff_audit_log table with who/what/when/target/outcome — see Data Model § ff_audit_log. It is not subject to the retention windows above; it's the accountability trail, kept independently.

Location precision#

Coarse location (block coordinates, not exact) is what staff-facing surfaces (Discord embeds, /ff finding) display when a finding carries a position — see QualityContext.location, and alerts.yml's discord.privacy.include-exact-coordinates (default false) for the Discord-embed opt-in to full precision. Note that the underlying stored value is already block-granularity in every case (ff_findings.location_x/y/z), never sub-block — "coarse" describes what's displayed, not an additional rounding step on top of storage.