ADSAntiray2 v5.0 Changelog

English

---

Completely refactored, supporting versions 1.7 through 26.2, with the vast majority of issues fixed and a significantly enhanced user experience!

Detection Logic Overhaul

Effect: Drastically cuts false positives for legit miners. X-Rayers and bots now get flagged based on mathematical impossibility rather than arbitrary speed limits.

Principle: Swapped 10 basic heuristic checks for 8 statistical modules. The system no longer asks "Is this player mining fast?" It calculates "What are the odds of finding this many diamonds in this specific pattern by pure chance?" using binomial distributions, entropy, and regression.

Config:

• Removed: Legacy module configs (fvd, tas, h_norm, ta, bts, spc, slm, ape).
• Added: ore-expected-rates (MC 1.18+ world gen baselines).
• Updated: qss-weights now maps to: osi, autotool_fast, autotool_session, path, view, speed, temporal, decoy.
• Tuned: vl-decay-per-minute bumped from 8.0 to 10.0 to match new scoring precision.

OSI (OreSeekerIndex)

Effect: Catches players who beeline for ores or mine in suspicious clusters, even if their break speed looks normal. Replaces old value-density checks.

Principle: Merges three statistical tests:

1. Statistical Anomaly: Binomial survival test. Calculates p-value of finding K ores in N breaks. Tiny probability = high suspicion.
2. Directness: Measures angle between movement vector and ore location. Moving straight at hidden ores spikes the score.
3. Clustering: Checks spatial density. Tight groups of ore finds indicate targeted vein seeking, not random luck.

Config:

• detection.osi.min-blocks: Sample size threshold (default: 80). Prevents flags on short sessions.
• Weights: 0.45 * stat + 0.35 * direct + 0.20 * cluster.

Path & View Analysis

Effect: Identifies branch miners and X-Rayers who scan walls without natural head movement. Consolidates old SPC, SLM, APE, and TA modules.

Principle:

• Path: Directional entropy + linear regression. Detects robotic 90-degree turn patterns correlated with ore finds. Low entropy (only moving in 2 directions) = suspicious.
• View: View-cone correlation + angular velocity stats. High blind-break ratio (breaking blocks never looked at) = X-Ray. Unnatural snap-aiming patterns detected via IQR outliers.

Config:

• detection.view.cone-degrees: FOV angle for correlation (default: 45).
• detection.path: Sensitivity for linearity and branching patterns.

Behavioral Consistency (Speed / Temporal / AutoTool)

Effect: Separates humans from bots via natural imperfection. Humans have variable reaction times; bots are consistently fast or perfectly rhythmic.

Principle:

• Temporal/Speed: Coefficient of Variation (CV). Humans: CV 0.3–0.7. Bots: CV < 0.2.
• AutoTool: Two layers. Fast catches sub-150ms switches (physically impossible). Session tracks long-term optimal tool usage ratios.

Config:

• detection.autotool_fast.rapid-threshold-ms: Impossible switch threshold (default: 150ms).
• detection.speed.short-interval-ms: Defines "suspiciously fast" consecutive mining.
• detection.temporal: Consistency vs randomness sensitivity.

Persistence & Decoys

Effect: Relogging no longer resets suspicion. Decoy hits tracked separately from main VL for cleaner admin logs.

Principle: Player data persists across sessions: total blocks, total ores, per-type counts, decoy history. Tool switch records now store full slot transitions, not just timestamps.

Config:

• kick-vl-threshold: 100.0 (same value, new meaning).
• ip-ban-vl-threshold: 150.0 (same value, new meaning).
• Decoy system integrated into nested data structure. No legacy config needed.

---

中文

---

彻底重构，版本支持1.7-26.2，修复了绝大多数问题，体验大幅升级！

检测逻辑重构

效果： 正常玩家误报大幅降低。X-Ray 和机器人现在因“数学上的不可能”被封禁，而非触碰任意速度阈值。

原理： 10 个基础启发式检查替换为 8 个统计模块。系统不再问“这人挖得快不快”，而是通过二项分布、熵分析和回归计算“纯靠运气在这种模式下找到这么多钻石的概率是多少”。

配置：

• 删除： 旧模块配置（fvd, tas, h_norm, ta, bts, spc, slm, ape）。
• 新增： ore-expected-rates（MC 1.18+ 世界生成基准）。
• 更新： qss-weights 映射至：osi, autotool_fast, autotool_session, path, view, speed, temporal, decoy。
• 调整： vl-decay-per-minute 从 8.0 提至 10.0，适配新评分精度。

OSI (矿物探寻指数)

效果： 即使挖掘速度正常，也能抓出直奔矿脉或可疑聚集挖掘的玩家。替代旧版价值密度检查。

原理： 三项统计检验合一：

1. 统计异常： 二项生存检验。计算 N 次破坏中找到 K 个矿石的 p 值。概率极低 = 高度可疑。
2. 直接性： 移动方向与矿石位置夹角。直冲隐藏矿石会拉高分数。
3. 聚集性： 空间密度分析。矿石呈紧密群组而非随机分布 = 定向找矿。

配置：

• detection.osi.min-blocks: 最小样本量（默认 80），防止短会话误报。
• 权重：0.45 * 统计 + 0.35 * 直接 + 0.20 * 聚集。

路径与视角分析

效果： 识别分支矿工和不自然转头扫墙的 X-Ray。整合旧版 SPC、SLM、APE、TA 模块。

原理：

• 路径： 方向熵 + 线性回归。检测与发现矿石相关的周期性 90 度转向。低熵（只朝两个方向移动）= 可疑。
• 视角： 视锥相关性 + 角速度统计。盲挖比例高（挖了没看过的方块）= X-Ray。通过 IQR 异常值检测不自然的急停瞄准。

配置：

• detection.view.cone-degrees: 相关性检查视锥角（默认 45）。
• detection.path: 直线度和分支模式灵敏度。

行为一致性 (速度 / 时间 / 自动工具)

效果： 用“自然的不完美”区分人机。人类反应有波动；机器人持续过快或节奏完美。

原理：

• 时间/速度： 变异系数 (CV)。人类 CV 0.3–0.7；机器人 CV < 0.2。
• 自动工具： 双层。Fast 抓 <150ms 切换（人类做不到）。Session 追踪长期最优工具使用率。

配置：

• detection.autotool_fast.rapid-threshold-ms: 不可能切换阈值（默认 150ms）。
• detection.speed.short-interval-ms: 定义“可疑快速”连续挖掘。
• detection.temporal: 一致性 vs 随机性灵敏度。

持久化与诱饵

效果： 重登不再重置嫌疑。诱饵命中与主 VL 分离，管理员日志更清晰。

原理： 玩家数据跨会话持久化：总方块、总矿石、分类计数、诱饵历史。工具切换记录存完整槽位转换，非仅时间戳。

配置：

• kick-vl-threshold: 100.0（数值不变，含义更新）。
• ip-ban-vl-threshold: 150.0（数值不变，含义更新）。
• 诱饵系统集成至嵌套数据结构，无需旧版配置。