QuantiX connects deterministic detection, behavioral intelligence, and model-assisted review into one explainable protection system-without handing enforcement to a black box.
The model is a second set of eyes for statistical evidence-the kind where latency, desync, skill, and manipulation can resemble one another. It reads the whole shape of an incident while hard system boundaries keep the final authority with evidence and people.
The reviewer receives measured signals, a player’s recent detector history, staff resolutions, account context, and-when available-24-hour aim statistics. It never sees secret thresholds or the action the pipeline already proposed, protecting both security and independent judgment.
One mode asks whether a particular Tier 2 flag looks real. Another looks across a player’s recent activity to find patterns no individual detector can see-especially when separate signal families begin telling the same story.
The response is schema-locked to three verdicts: likely cheating, inconclusive, or likely legitimate, with bounded reasoning and indicators. Player-controlled text is neutralized before review. Even a compromised response cannot invent a command.
Deterministic Tier 1 signals never go to a probabilistic reviewer.
A “legitimate” opinion below the minimum cannot suppress escalation.
A “cheating” opinion needs the higher bar before adding suspicion.
The model cannot ban, refuse a join, or resolve a case-at any confidence.
Communities can evaluate verdicts for a week without changing outcomes.
If review fails or times out, the original case continues unchanged.
Every detector states what it measures, which evidence tier it belongs to, and whether the observation came from a client process or a server authority.
Your recorded travel speed exceeded what was possible for the vehicle or character state at the time.
Your character moved a distance in one instant that cannot be travelled in that time.
Your character moved through terrain or remained airborne in a way normal movement does not allow.
A vehicle you were driving behaved outside the limits of its handling.
Your viewpoint was separated from your character in a way normal play does not produce.
Your character covered ground while the game reported it as almost stationary, which happens when a position is set directly rather than moved to.
A weapon appeared in your inventory without being obtained through the server.
The damage you dealt over a short period exceeded the maximum the weapon can produce.
A weapon was fired faster than its mechanical rate of fire allows.
Damage was registered at a range the weapon used cannot reach.
Your health or armour increased without a corresponding in-game source.
A weapon that does not exist in the game was used during this session.
The weapon reported by your game client did not match the weapon the server observed you holding.
Damage was applied to your character and your health did not change.
Your character absorbed a sustained run of damage without any loss of health or armour.
The damage recorded for a shot did not match what the weapon used can produce.
You damaged an unusually large number of different players in a very short time.
You killed an unusually large number of different players in a very short time.
Explosions were attributed to you that do not match the equipment you were carrying.
You damaged other players while your character was invisible to the server.
Damage dealt to your character had far less lasting effect than it has on other players: it was reduced, undone straight away, or absorbed by health above what the server provides.
Your weapons removed far more health per hit than the same weapons do for other players on this server.
A weapon kept landing shots for longer than its magazine holds, without pausing long enough to reload.
Your aim swept onto a target and then stayed fixed to it as it moved, with none of the correction normal aiming shows, immediately before a shot landed.
Your aim followed a moving target with a steadiness outside the range normal aiming produces.
Your headshot rate against other players, over enough hits to be meaningful, was far above the range manual aiming produces.
Your aim settled onto players who were not visible from your position at the time.
Your aiming statistics across many sessions sat outside the range produced by manual aiming.
Damage was registered on a player your weapon was not pointing at.
The reported path of a shot did not run from your weapon towards where it was aimed.
Damage was registered on a player with solid cover between you and them.
Your shots landed faster after players moved into your aim than a person reacts, repeatedly, while your aim was held still.
Software matching a known cheat tool was detected running alongside the game client.
A known cheat menu was identified on the game client during this session.
The set of scripts running on your game client changed unexpectedly during the session.
The game client triggered a server action that the game cannot produce through normal play.
Network data from the game client was malformed in a way the unmodified game does not produce.
A game action was repeated far faster than the game itself allows it to be repeated.
Your game client sent synchronised state updates far faster than the game produces them.
A server action was triggered with values the game itself cannot produce for it.
An inventory was opened from a position or state the server does not allow it to be opened from.
The pattern and timing of actions sent by your game client did not match ordinary play.
The game client attempted to control an object or vehicle it did not have control of.
A vehicle or object appeared under your control without the server being able to account for where it came from.
An object was attached to another player’s character from your game client.
An object created by your game client was structurally invalid for the type it claimed to be.
Objects or vehicles appeared under your control faster than the server can account for.
Objects were attached to players in a pattern used to trap or move them against their control.
Another player was accelerated by an object under your control beyond what a collision produces.
A vehicle appeared under your control that the server has no record of issuing.
The anticheat component running on your game client did not match its expected state.
The anticheat component on your game client stopped reporting while you remained connected.
Several independent server-observed behaviours associated with cheat tools occurred together during the session.
Items appeared in your inventory without a matching transaction on the server.
A reward was granted to you more times than it can be claimed.
Your balance increased without a matching credit issued by the server.
Invalid weapons, ownership contradictions, forbidden events, known cheat fingerprints, malformed contracts. The first trigger may kick. Repeated server-authoritative evidence can create a temporary ban. Client-originated evidence always stops at human review.
Speed, aim, damage, physics, integrity, patterns, and economy anomalies accumulate in a decaying suspicion score. Repeats, episode saturation, trust dampening, and corroborating families matter more than any isolated spike.
Detection is only useful when staff can understand it, act proportionally, explain the result, and preserve the right evidence without collecting everything forever.
Every flag carries structured measurements and a 5 Hz state window from five seconds before to five seconds after the trigger. Combat evidence includes the players who actually interacted; movement evidence collects only the subject. Frames compress directly to object storage and are never uploaded until something fires.
A live roster and map connect staff to a closed vocabulary of safe server actions: message, freeze, bring, return, revive, heal, strip weapons, clear entities, screenshot, watch, record, kick, and ban. Every command names its operator, records the reason, and confirms whether it reached the server.
Watch up to sixteen player screens at approximately one frame per second. Each tile shows frame age and interval so staff never mistake the feed for video. Watching requires a written reason; unrecorded frames expire after ninety seconds, while retained sessions enter the evidence lifecycle.
An optional rolling reconstruction stores server-observed state at 2 Hz for thirty minutes. Investigators can revisit an incident from any angle even when no detector fired, then promote only the relevant segment into retained evidence with a named reviewer and written reason.
A published 32-factor score combines local and network enforcement history, identifier quality, account age, tenure, playtime, cross-server presence, detector history, appeals, connection intelligence, and human adjustments. New players start neutral at 50; every factor and weight is inspectable.
Ordered observe-first rules cover peds, vehicles, objects, models, population types, and trusted-resource bypasses. Named statistical detections can be suppressed inside bounded gameplay zones for legitimate mechanics. Deterministic Tier 1 detections remain unsuppressible everywhere.
Leased review queues prevent duplicate claims and abandoned locks. Bans can be scoped to one server, a community, or the network, including players who are offline. Public appeals prove identity with single-use claim codes and hold relevant evidence until a decision is complete.
The join gate starts off, supports observe mode, and can fail open during outages. Manual grants, approved versioned applications, Discord roles, and staff access are additive. Automatic review may approve a clearly trusted applicant, but it can never automatically deny one.
When a server is full, admitted players wait on a live loading-screen card and enter as soon as capacity opens. Reconnect grace protects a dropped player’s place, while Discord roles can carry named priority levels. Staff see the live line in the console and through the bot.
Every downloadable client and server bundle is minified, source-map free, and content-seeded for a different obfuscated layout after each source change. Packages are checksum-verifiable and carry an exact build identity; outdated or unverified installs stop safely instead of pretending to protect a server.
Join history can include country, region, network, and positively identified VPN, proxy, Tor, or relay context without treating an IP address as identity. Optional region and anti-VPN policies fail open when intelligence is unavailable, and every decision remains visible to staff.
Typed rules can notify, flag, watch, kick, refuse, or temporarily ban when connection or detection conditions match. They cannot execute Lua or JavaScript and cannot issue a permanent ban. Rules launch in observe mode, and every match records exactly what happened.
Staff sign in with Discord, permissions are recomputed from mapped guild roles on every login, and the bot brings lookups, cases, live actions, bans, appeals, recordings, analytics, watchlists, allowlists, and whitelist reviews into the community’s existing workflow.
The opt-in ban network exchanges peppered identifier hashes and aggregate counts-not names, reasons, or the communities that reported them. Alias links are visible to investigators but never trigger automatic action, and IP addresses are deliberately excluded as identity.
Postgres row-level security and composite tenant keys isolate every community. Scoped credentials limit each consumer. Separate retention clocks govern transient and confirmed evidence, open appeals pause expiry, and erasure deletes clean players or anonymizes banned ones without silently lifting enforcement.
Forty-nine detectors, model-assisted intelligence, server-side evidence, and human-governed enforcement-working as one.
