No Anticheat
Lab

Defensive calibration room

Measure before you enforce.

POLICY local deterministic builder

METHOD published fixture definitions

RESULT evidence required

METHOD NOTE / NAC–M1.0 / PREREGISTERED SHAPE

Methodology

Every future finding follows one shape: control, treatment, difference, uncertainty, limitations. This page is the preregistration of that shape — published before any result so the method cannot bend to fit an outcome.

FOLIO 01 / REVISION 2.0.0

RUN ENVELOPE / REQUIRED FIELDS

The run manifest

Nothing runs without a signed manifest recording, at minimum:

  • M–01

    hardware class, OS image, JVM build and flags;

  • M–02

    Minecraft/server build, proxy and translation layers, with exact versions;

  • M–03

    plugin artifact SHA-256 and its full configuration;

  • M–04

    world fixture hash, player count, workload seed;

  • M–05

    network profile (latency, jitter, loss), warm-up window, duration;

  • M–06

    collector versions and raw capture locations.

A finding that cannot be reproduced from its manifest does not publish. Evidence records will be published alongside the corresponding methods with signed manifests and raw series, so verification does not depend on summary prose alone.

Publication requires
reproduction

BENCH SEQUENCE / PARITY FIRST

Controls and repetition

  1. The control is untreated — same stack, no enforcement plugin — and re-proven for each study, not reused from a previous month.

  2. Treatments are the plugin’s default configuration plus explicitly named variants; no anonymous “tuned” setups.

  3. Run order is randomized and repetition counts follow a sample plan fixed before the first run.

INTERPRETATION / THREE STATES

Statistics, in one honest paragraph

Results report distributions — medians, spread, and interval estimates — never a single average. A comparison resolves to exactly one of three states. Underpowered comparisons say inconclusiveout loud instead of borrowing significance they did not earn.

observed difference

no clear difference

inconclusive

RESULT LANGUAGE / HUMAN FIRST

Language rules

  • 01

    A detection event is a flag, never an accusation; the lab measures software, not players.

  • 02

    “Detection-event rate in a known-clean fixture” is the claim; a universal false-positive rate is not.

  • 03

    “We did not test this” appears wherever it is true.

Continue to the fixture definitions or the publication limits.