Manual/Anticheat/Anticheat — methodology
v1.0.364 notes · build 2026-11-07 12:10 UTC

Anticheat — methodology

detection framework · evidence standards · topic index
#anticheat #detection

Applied section. Each topic is one automation problem: what the tool does, which protocol fact makes it possible, which observable it leaves behind, and the method that turns the observable into evidence. Built directly on the reference notes — nothing here works without the protocol map.

The one rule

Patterns beat peaks. A human can be the fastest digger on the server for one evening. A human cannot be equally fast, with equal variance, at 03:47, seven days a week, forever. Detection targets the shape of behavior over time, never a single impressive number. Every method below is a variance/consistency test first and a speed test never.

Pipeline (same for every topic)

witness account (officer-owned, in-alliance, read-only)
│  sees exactly what any member sees: broadcasts, claims, reactions
▼
collector ── normalizes frames to timed event rows (JSONL, spooled to disk)
▼
daily export ── mailed out for evaluation
▼
analysis pass (AI or script) ── per-player distributions, outlier flags
▼
human-check ── an unautomatable interaction only a real person passes
▼
evidence pack to R4 ── suspicion is never a verdict, nothing goes public

Evidence standards

  • Every flag carries a confidence and the rows behind it. No verdict from a single signal.
  • Event-burst windows (SVS/KV ±30 min) are excluded — honest hardcore play aliases to bot behavior there.
  • One exonerated suspect per few guilty ones is the expected error rate; the human-check exists to catch them, and an apology is part of the process.
  • Witness data stays alliance-internal. It is a check on tools, not a weapon on people.

Topics

TopicStatusContents
1 · Technical — digslivewire loophole · module anatomy (redacted) · rhythm/quantization tells · confrontation protocol · rules the tools can't follow
2 · Non-technical — the short versionlivethe member-facing briefing: what auto does and can't do, the two friendly rules, the one-day policy
Stations / shop snipingin collectionrotation-instant burst shapes
Rally joinsin collectionmillisecond-percentile join spikes without interaction side-channel
Kill farmingqueuedkill surges without matching march activity

The two methods

Every topic is worked twice: detect (telemetry → patterns → astronomical probability → friendly replication on camera, never accusation) and displace (rules targeting features no tool offers — today: liking a message, sending a sticker — quick, equal for everyone, judgement-free, one-day penalty). Detection finds them; rules make the tool worthless.

Access

Collector internals stay redacted on purpose — tool vendors read public pages too. Full material goes to alliance management that can vouch for themselves. Same rule as everywhere on this site: prove you're working against automation and the redactions come off.

notes build 2026-11-07 12:10 UTC · lw-notes@9c47e1b · sources: own capture/analysis cross-checked against the public dossier