Known Limitations#

Disclosed honestly rather than left for you to discover.

PvP-only combat#

combat.reach.a/attack_rate.a/multitarget.a/bad_order.a only evaluate player-versus-player attacks. Reach/rewind reuse the target's own tracked movement history, which a mob does not have — a documented, safe-direction scope cut (under-detect PvE abuse rather than risk a false accusation from an imprecise mob-position estimate).

CANCEL is not yet live through the normal policy pipeline#

A policy or correlation rule's response-ceiling/response: CANCEL (see the Tuning Guide) is structurally implemented in ActionDispatcher but has no live trigger yet: no current input adapter evaluates a check synchronously early enough in a cancellable Bukkit event to supply a cancel hook. Today, every CANCEL reached through the normal policy pipeline is honestly recorded as a SUPPRESSED action in ff_actions//ff status — never a silent no-op, and never a fake success.

packet.payload.a is the sole check exception, and it doesn't go through this pipeline at all: a non-finite (NaN/infinite) movement coordinate is rejected by a hardcoded event.setCancelled(true) safety guard inside MovementListener itself, unconditionally and regardless of observe-only/enforcement-enabled — see Enforcement & First-Start and the Check Parameter Reference.

protection.yml's crash guard, which packet.payload.b reports on, works the same way outside the pipeline: it drops packets no vanilla client sends on the network thread itself, before the server handles them — see Configuration § Crash guard.

PostgreSQL/MariaDB paths are implemented but not as heavily exercised as SQLite#

If you run one of these, keep an eye on /ff status after upgrades.

No dedicated main-thread-scheduling-delay or TPS/MSPT health-monitor subsystem#

Live TPS is already read and used to scale finding confidence, and per-check count/timing/ circuit-breaker state is exposed via /ff checks, but there is no single consolidated "health monitor" object tracking scheduling delay.

movement.phase.a samples points, not the whole hitbox#

It tests points along the movement path at three heights inside the player against each block's real collision shape, not a full swept-AABB sweep of the player's hitbox — so it can miss a clip through a thin edge between sample points. Adequate for an experimental, alert-only check, not precise geometry.

Unused configuration keys#

Six shipped keys are parsed but currently have no effect on behavior — disclosed here so you don't spend time tuning any of them expecting a result:

  • checks.yml's defaults.observe-only is never read by the config loader. See Check Parameter Reference § Shared defaults for which switch actually controls enforcement.
  • alerts.yml's discord.privacy.include-exact-coordinates is parsed and stored, but the Discord embed builder only ever reads discord.include-coarse-location — toggling include-exact-coordinates currently changes nothing about what's sent to Discord.
  • alerts.yml's ingame.alert-cooldown and ingame.aggregate-window are both parsed into InGameAlertConfig but never read anywhere: NotificationDispatcher only consults ingame.enabled and ingame.minimum-confidence before dispatching, and InGameAlertDispatcher sends immediately to every eligible online player with no cooldown or aggregation logic of its own. Unlike Discord delivery (whose aggregate-window is live), an in-game alert has no built-in throttling beyond the upstream per-check alert-cooldown in checks.yml (which gates whether a finding is alert-eligible at all, not specifically in-game delivery).
  • storage.yml's retention.aggregate-statistics-days and retention.ledger-entries-days are both parsed into StorageConfig but never consulted by any purge logic. Only ff_findings (governed by retention.detailed-evidence-days, which is live — read directly at startup) is ever purged by the scheduled task. ff_actions (the action ledger) has no retention-expiry column at all and grows unbounded — see Data Model § ff_actions. There is also no dedicated aggregate-statistics table in the current schema for aggregate-statistics-days to apply to.