1. Research Question｜研究问题
Among eligible modern-English lexical families derived from the same frozen population, is the Direct-Target proportion higher for orthographic W initials than for preregistered pooled eligible non-W consonant controls?
在同一冻结现代英语词汇总体中，以正字法 W 开头的合格独立词汇家族，其 Direct Target 比例是否高于预注册的合并非 W 辅音对照？
2. Scope｜检验范围
This is a statistical semantic-concentration test. It does not test whether W means water, whether its shape originated in a wave, or whether relevant words are historically related.
本实验检验统计性语义集中，不检验 W 本身是否表示水、W 字形是否源自水波，或相关词是否具有共同历史来源。
3. Direct-Target Domains｜直接目标语义域
- A — WATER: water, a body of water, or water's presence/state.
- B — WETNESS / WATER-CONTACT: wetness, soaking, washing, or direct water/liquid contact.
- C — WAVE / FLOW: wave, flow, oscillation, or undulating movement.
A 为水、水体或水的存在／状态；B 为湿润、浸泡、冲洗或直接水／液体接触；C 为波、流动、振荡或起伏运动。
4. Coding｜编码
1 = Direct Target; 2 = Associated/Borderline; 0 = Not Target; U = Unclear. Only code 1 enters the primary endpoint.
1＝直接目标；2＝关联／边界；0＝非目标；U＝不明确。主终点只使用 1。
5. Inclusion and Exclusion｜纳入与排除
A conventional lexical sense must directly encode the target. Habitat, typical use, material composition, cultural association, incidental knowledge and merely being a liquid are insufficient. Pure metaphorical extensions do not create independent target evidence. Annotators may not expand the domain after seeing items.
常规词义必须直接编码目标概念。栖息环境、典型用途、物质组成、文化联想、附带现实知识或“碰巧是液体”均不足以纳入；纯隐喻延伸不产生新的独立目标证据；查看材料后不得扩大语义域。
6. Polysemy｜多义词
Sampling is lemma-level and annotation is sense-level. All eligible conventional senses are presented under a standardized procedure. Researchers may not select a favorable sense after outcomes are known. Presence in the primary/common sense is recorded separately as robustness information.
抽样单位为 lemma，标注单位为 sense；按标准程序呈现全部合格常规义项，不得事后挑选有利义项。是否出现在主要／常用义中另作稳健性记录。
7. Source and Population｜数据源与总体
The source is a frozen local copy of public SUBTLEX-US frequency data. Filename, retrieval record, checksum, raw rows, selected sheet/delimiter, column mapping and processed rows must be archived before sampling. W and controls use the same population and rules. No researcher-added vocabulary is permitted.
数据源为本地冻结的公开 SUBTLEX-US 词频数据。抽样前必须归档文件名、获取记录、校验和、原始行数、工作表／分隔符、列映射与处理后行数。W 和对照使用同一总体与规则，不得人工加词。
8. Deterministic Preprocessing｜确定性预处理
Normalization, eligibility handling, lemmatization algorithm, exception table, software/runtime versions and output hashes are implementation artifacts frozen before sampling. Purely inflectional forms do not create independent lemma observations.
规范化、资格处理、词元化算法、例外表、软件／运行时版本和输出哈希均属于实施记录，须在抽样前冻结；纯屈折变体不得形成独立 lemma 观察。
9. Sampling｜抽样
Frequency-stratified random sampling is used. Known examples such as water, wave, wet, wash and wavelength enter only through the formal procedure.
采用词频分层随机抽样。已讨论的 water、wave、wet、wash、wavelength 只能通过正式程序进入。
10. Sample Size and Shortfall｜样本量与不足规则
Target: 120 W families and 240 Control families (1:2; N=360). Reference planning scenario: pC=.05, pW=.15, RR=3, RD=.10, alpha=.05, two-sided, power target ≥.80. The 5% baseline is a design parameter, not an empirical claim. If fewer than 120 eligible W families exist, retain all and sample controls at 2:1 where feasible; do not expand the population to reach N.
目标为 120 个 W 家族与 240 个对照家族。参考情景 pC=.05、pW=.15；若合格 W 家族不足 120，则全部保留并尽可能按 2:1 抽取对照，不得扩大总体凑数。
11. Controls｜对照
Primary: pooled eligible non-W consonant initials. Robustness A: frequency-matched controls. Robustness B: repeated random consonant-control sets. V and other phonological/historical comparisons are exploratory only. The exact primary-initial set is an implementation item frozen before annotation.
主对照为合并合格非 W 辅音首字母；稳健性 A 为词频匹配对照；稳健性 B 为重复随机辅音对照；V 等音系／历史比较仅属探索性。主对照的精确字母集合须在标注前作为实施项目冻结。
12. Family Deduplication｜词族去重
Transparent derivatives, compounds inheriting target semantics from the same constituent, and clearly documented near-level stem families contribute at most one observation. Deep historical relatedness alone does not force merging. Family assignment is semantic-label blind and complete before semantic annotation.
透明派生词、继承同一构词成分目标语义的复合词及有明确近层级词干依据的家族最多贡献一次；仅凭深历史关系不强制合并。词族分配须在语义标注前且看不到语义标签时完成。
13. Hypothesis-Blind Annotation｜假说盲标
Two independent annotators see anonymous ID, lemma, standardized conventional sense and necessary part of speech, in separately randomized orders. They do not see the hypothesis, group, stratum, aggregate counts, family outcome or each other's judgment, and may not consult etymology.
两名独立标注者按不同随机顺序看到匿名 ID、lemma、标准化常规义项和必要词性；看不到假说、组别、频率层、总体计数、词族结果或对方判断，也不得查词源作语义判定。
14. Disagreement and Agreement｜分歧与一致性
Original A/B judgments are preserved. Any 1/0, 1/2, 0/2 or U-involving mismatch enters adjudication by a third hypothesis-blind adjudicator. Report raw agreement and an appropriate chance-corrected statistic.
A/B 原始判断永久保留；所有不一致进入第三名假说盲裁决者处理，并报告原始一致率与适当的机会校正一致性指标。
15. Primary Analysis｜主分析
The primary 2×2 family table is W versus Controls by Direct Target versus all other eligible codes. Report raw RR, RD, preregistered 95% intervals and a two-sided Fisher exact p-value at alpha=.05. Zero cells do not replace Fisher; raw zero/infinite/undefined RR is retained and a continuity-corrected RR appears only as labeled sensitivity.
主 2×2 词族表比较 W／对照与 Direct Target／其他合格编码；报告原始 RR、RD、预注册 95% 区间和双侧 Fisher 精确检验（α=.05）。零单元格不改变 Fisher；原始零／无穷／未定义 RR 保留，连续性校正 RR 只作明确标记的敏感性结果。
16. Multiplicity and Secondary Work｜多重性与次级分析
There is one preregistered primary test, with no multiplicity adjustment for that single test. Lemma-level, frequency-stratified, Borderline-inclusive, random-control, V-control and other tests are secondary, robustness, sensitivity or exploratory and cannot upgrade the primary status.
只有一个预注册主检验，无需对该单一检验作多重校正。lemma 层、频率层、纳入 Borderline、随机对照、V 对照等均属次级／稳健性／敏感性／探索性分析，不能升级主结果。
17. Decision Rule and SESOI｜结论规则与最小有意义效应
SESOI is RR=2.0. SUPPORTED requires RR>1, the preregistered 95% RR interval to exclude 1, and two-sided Fisher p<.05. If not Supported and the interval permits both RR=1 and RR≥2, status is INCONCLUSIVE. If meaningful enrichment is excluded, or clear evidence is at/below RR=1, status is NOT SUPPORTED. A fatal protocol breach is INVALID.
SESOI 为 RR=2.0。SUPPORTED 必须同时满足 RR>1、预注册 95% RR 区间排除 1、双侧 Fisher p<.05；未支持且区间同时允许 RR=1 与 RR≥2 时为 INCONCLUSIVE；排除有意义富集或明确位于／低于 RR=1 时为 NOT SUPPORTED；致命协议违规为 INVALID。
18. No Rescue Rule｜禁止挽救结果
A failed primary result cannot be rescued by expanding the domain, adding Borderline cases, deleting counterexamples, adding W words, changing family boundaries, selecting favorable strata or controls, or replacing the statistic. Later work is labeled post-hoc exploratory.
不得通过扩大语义域、加入边界项、删除反例、补充 W 词、改变词族边界、挑选有利频率层／对照或更换统计量来挽救主结果；后续研究必须标记为事后探索。
19. Negative-Result Publication｜负结果公开
The experiment remains publicly reported whether Supported, Inconclusive, Not Supported or Invalid.
无论结果为支持、不确定、不支持或实验无效，均保留并公开。
20. Freeze Boundary｜冻结边界
Design Freeze is complete. Implementation Freeze remains pending until source checksum, provenance, preprocessing, lemmatization, exact-power verification, seeds, sampling manifest and family manifest are complete and validated. Formal annotation and primary analysis must not begin before that gate passes.
设计冻结已完成。实施冻结须等待来源校验和与 provenance、预处理、词元化、精确功效验证、随机种子、抽样清单和词族清单全部完成并验证；在闸门通过前不得开始正式盲标或主分析。