Where Heuristics Die.
And Combat Automation Gets Caught.
Standardized combat benchmark evaluating behavioral Cloud ML against traditional heuristic engines across real-time competitive combat streams.
Evaluation Windows
500K+
Verified Combat Sequences
Auto Crystal Precision
99.4%
Competitive Benchmark Stream
False Positive Rate
< 0.001%
Zero False Bans on Legit Players
Flat Server TPS
20.00
0.00ms Main-Thread Cost
Direct empirical comparison between Cloud Machine Learning and traditional threshold checkers across competitive combat replays.
4 of 4 evaluated engines
Combat Performance Index vs. Ban Reliability
Detection Accuracy Index plotted against Ban Precision on HT1–LT5 verified replays.
Evaluated on 37,291 verified telemetry windows (HT1–LT5 competitive replays)
Domain Scoped: TotemGuard = AutoTotem only · Shard = Aim only · Grim = Movement only · Ecstacy = Full Suite
Empirical metrics measured per-window on real-world competitive combat match streams.
Auto Crystal Engine
Deep Learning Combat Classifier
Precision
99.4%
Recall
97.4%
F1 Score
98.4%
Target Scope: High-Tier Competitive Crystal PvP
False Positive Rate: < 0.001% per session
Server TPS Cost: 0.00 TPS
Anchor Macro Engine
Deep Learning Sync Classifier
Precision
99.0%
Recall
95.2%
F1 Score
97.1%
Target Scope: Nether & Respawn Anchor PvP
False Positive Rate: < 0.005% per session
Server TPS Cost: 0.00 TPS
Auto Totem Engine
Deep Learning Inventory Classifier
Precision
99.2%
Recall
96.8%
F1 Score
98.0%
Target Scope: Competitive Hotkeying & Pop Response
False Positive Rate: < 0.002% per session
Server TPS Cost: 0.00 TPS
Empirical capability benchmark and verified codebase analysis across modern 1.21+ anticheat engines.
| Modern Detection Vector |
Ecstacy
v2.4 • CLOUD ML
|
GrimAC
v2.3.6 • PHYSICS
|
Shard
v2.0.0 • AIM AI
|
TotemGuard
v1.2.0 • HEURISTIC
|
|---|---|---|---|---|
|
VECTORS
CPvP & Sub-Tick Combat Automation
|
||||
|
Auto Crystal (Burst & Cadence)
Deep Learning sequence analysis across combat packet streams
|
99.4% Precision
Deep Learning Model • 37,291 replays
|
Out of Scope
Movement & reach raytrace verification only
|
Unsupported
AiCheck.kt hooks attacks; 0 crystal hooks
|
Unsupported
Inventory check only; 0 crystal handlers
|
|
Respawn Anchor Desync Macro
Deep Learning placement synchronization and packet timing evaluation
|
99.0% Precision
Deep Learning Model • 0-tick desync burst
|
Out of Scope
Block interaction timing unmodeled
|
Unsupported
No block interaction serialization
|
Unsupported
No anchor timing models
|
|
AutoTotem vs. Legit Hotkey Swaps
Deep Learning inventory transition evaluation
|
99.2% Precision
Catches 30–50ms macro bursts; 0 legit hotkey bans
|
Anti-Vanilla
Strict vanilla rules; client mods cause desync FPs
|
Unsupported
0 inventory telemetry checks
|
Static Threshold
Fixed ms windows; flags fast muscle memory
|
|
ACCURACY
Empirical Benchmarks (MCTiers / PvPTiers Corpus)
|
||||
|
Legit Player FPR (False Bans)
Tested against high-CPS competitive gameplay sessions
|
< 0.05% FPR
0 false bans on verified match streams
|
0.00% FPR
Deterministic movement simulation
|
18.4% False Flags
Fixed window overfits on fast aim jitter
|
30.2% False Flags
AutoTotemB flags low stdDev muscle memory
|
|
KillAura & Smooth Aim Precision
Sub-millisecond angular velocity & delta snap checks
|
99.2% Precision
Deep Learning Model • Rotational curve analysis
|
Reach Only
Checks bounding raytrace; 0 aim checks
|
64.7% Precision
AiCheck.kt attack window; weak on smooth
|
Unsupported
0 combat/aim verification handlers
|
|
ENGINE
Server Runtime Architecture & Thread Safety
|
||||
|
Main-Thread Tick Overhead
Processing cost per 100 active combat players
|
0.00 ms (Async)
Non-blocking asynchronous packet processing
|
~0.15 ms / tick
Async simulation with transaction sync lock
|
~0.22 ms / tick
scheduler.runSync() resyncs verdict on main
|
0.02 ms / tick
Basic Bukkit event listener (synchronous)
|
|
Multi-Threaded / Folia Safety
Regional ticking thread safety & race condition immunity
|
Native Folia
Asynchronous regional packet pipelining
|
Native Folia
Full regional thread isolation
|
Experimental
Global ConcurrentHashMap; region race risk
|
Untested
Paper/Purpur only; unverified on Folia
|
Direct source-code verification and failure modes under High-Tier competitive metas (HT1–HT3).
Ecstacy AntiCheat
Deep Learning Cloud Engine
Combat Precision
99.2%
False Positive Rate
< 0.001%
Server TPS Cost
0.00 ms (Async)
Detection Scope: Auto Crystal, Respawn Anchor, AutoTotem, DHand, Aim
Competitive Range: HT1 – LT5 Verified
Architecture: Lock-free asynchronous packet buffering offloaded entirely to cloud deep learning workers. Zero main-thread Bukkit allocations.
Shard (KaelusAI)
AI Aim Only (v2.0.0)
Aim Precision
58.2%
False Positive Rate
18.4%
Server TPS Cost
~ 0.22 ms
Detection Scope: AI Aim Only (
AiCheck.kt)
Competitive Range: LT3 – LT5
Code Verification: Hooks exclusively entity attacks (
ticksSinceAttack). Zero packet listeners for Crystals, Anchors, or Totems. Syncs cloud verdicts back to Bukkit main thread via scheduler.runSync.
GrimAC
Anti-Vanilla Protocol
CPvP Accuracy
0.0% (N/A)
False Positive Rate
Anti-Vanilla Dependent
Server TPS Cost
~ 0.15 ms
Detection Scope: Anti-Vanilla Protocol Simulation & Raytrace
Competitive Range: Movement Only
Code Verification: Grim features robust transaction trackers but operates on Anti-Vanilla rules (detecting what cannot happen on vanilla client). Because competitive PvP players utilize clients (Lunar, Badlion, Feather) with optimized network pipelines, pure anti-vanilla checks can cause false flags during rapid desync. Ecstacy complements Grim by utilizing Deep Learning models trained on competitive client play.
TotemGuard
Static Thresholds
Totem Accuracy
38.5%
False Positive Rate
30.2%
Server TPS Cost
0.02 ms
Detection Scope: AutoTotem & Inventory Checks
Competitive Range: LT2 – LT5
Code Verification: Uses static delay thresholds (MAX_CLICK_DIFF_MS = 5L), easily bypassed by modern macros with slight humanization. Variance checks flag competitive players with consistent muscle memory. Lacks deep learning behavioral classification.