Unilanguage

Reusable Research Method · 可复用研究方法

Unilanguage Hypothesis Evaluation Framework

Unilanguage 假说评测框架

Framework overview · First formally implemented in UNI-EXP-001A-R2

Purpose

This framework turns proposed sound–meaning, cross-language and cognitive mappings into preregistered, testable and revisable research claims.

目的

该框架把语音—语义、跨语言及认知映射提案转化为预注册、可检验、可修正的研究主张。

01

Core Method · 核心方法

Predefinition · 预先定义

Freeze the question, scope, categories and decision rules before outcome inspection.

查看结果前冻结问题、范围、类别与裁决规则。

Sampling & controls · 抽样与对照

Use random or frequency-based sampling, predefined controls and comparable frequency bands.

采用随机或频率抽样、预定义对照与可比频率区间。

Blind annotation · 盲标

Keep annotators unaware of the target group where practical, and record disagreement.

在可行情况下使标注者不知道目标组，并记录分歧。

Counterexamples · 反例

Retain non-supportive cases; do not rescue the claim by changing categories after inspection.

保留不支持案例；不得在查看结果后改变类别来挽救假说。

Evidence-family deduplication · 证据家族去重

Separate related forms so one lexical family does not masquerade as many independent observations.

分离相关词形，避免一个词汇家族伪装成多项独立观察。

Effect size + 95% CI · 效应量与95%置信区间

Report magnitude and uncertainty, not only a binary significance label.

报告效应大小与不确定性，而不只给出二元显著性标签。

Transparent decisions · 透明裁决

Apply published decision rules and show which conditions were or were not met.

执行已公开的裁决规则，并显示哪些条件达到或未达到。

Negative results · 公开负结果

Publish hypotheses the experiments fail to support, together with limitations and revision paths.

公开实验未能支持的假说，并说明限制与修正路径。

02

Blind Evaluation Pipeline v1.0 · 盲评标准流程 v1.0

Five gates are non-bypassable. One program may launch the workflow, but it must stop at every gate until the required artifacts are frozen, checks pass and the named authority approves continuation. No job, script or agent may cross a gate automatically.
五道 Gate 均不可绕过。流程可以由一个程序统一启动，但必须在每道 Gate 停止，直到必需材料冻结、检查通过并由指定授权者批准继续；任何任务、脚本或智能体都不得自动跨越 Gate。

Preregistration Freeze → Annotation Freeze → Blinded Quality Review → Authorized Unblinding → Final Classification

Reusable infrastructure · 可复用基础设施

Repository-ready templates

Copy the metadata, preregistration, OPS, manifest, blind annotation, review, unblinding, analysis and classification templates for a future registered experiment.

Open template package · 打开模板包 →

03

Reference Case and Versioning · 参考案例与版本

Experiment 001A is the First Reference Implementation | 首个参考实例. Its preregistration is not a replaceable summary: it is the complete 31-section historical record frozen before outcome inspection. The reusable Pipeline abstracts the controls that worked in 001A; it does not rewrite 001A or create a competing method.
Experiment 001A 是 First Reference Implementation｜首个参考实例。其预注册不是可替换的摘要，而是在查看结果前冻结的完整31项历史记录。通用流程只抽象 001A 中有效的控制机制，不重写 001A，也不另造冲突方法。

Frozen record

UNI-EXP-001A-R2 · v1.0

Author: Jinkai Liu · Date: 2026-08-19 · Status: Frozen / Preregistered

Version rule

Append; never overwrite · 新增版本，不覆盖

Pipeline v1.1, v2.0 or later may refine the reusable method, but they must never overwrite this Pipeline release or any frozen experiment record.

未来 Pipeline v1.1、v2.0 可改进通用方法，但不得覆盖本版本或任何实验的冻结记录。

04

Evidence Chain · 证据链

Source → Observation → Mapping → Hypothesis → Experiment → Evaluation → Protocol

Established external evidence remains distinct from Unilanguage Analysis / Proposed Mapping. Read the public citation convention for the required labels, inline citations and reference-list structure.

外部已确立证据必须与 Unilanguage Analysis / Proposed Mapping（Unilanguage 分析／拟议映射）明确区分。公开引用规范规定所需标签、正文编号引文与参考文献结构。

Read Evidence & References · 阅读证据与引文规范 →