Evidence level: D
导言
五池 ④ 在过去 3 轮(第 17/19/20 轮)尝试用自动 agent + SearXNG 联邦检索 Schumann 山区站 vs 平原站文献基线(课题 #18/#20),累计 ≥13 次 web_search 全部命中 ASCII 词条噪声:英文人名 Schwarzenegger、北美原住民 “Two-Spirit”、维州强奸新闻报道等,与环境-生命频率接口主题完全无关。
这种”关键词命中但语义错配”的污染,根因不在 SearXNG 本身,而在自动 agent 跨语种检索时缺少 (a) 字符集白名单、(b) 专业术语罗马化对照表、(c) 三轨检索流程、(d) 失败升级路径。
本 SOP 是 #18/#20 三轮失败后沉淀的方法学交付物。它不直接评估 Schumann 信号或环境 ELF 生物效应——证据边界限定在”检索流程层”。任何后续依赖自动检索的五池课题(不论池 ①—⑤)都可以复用本 SOP。
一、问题定义:关键词 ASCII 噪声污染
1.1 三类污染源
| 类别 | 例子(来自 #18 失败实录) | 触发关键词 | 污染机制 |
|---|---|---|---|
| 人名/品牌同形异义 | Arnold Schwarzenegger 个人健身站 | Schumann → 子串匹配 Schwarzenegger |
拉丁字符子串匹配 |
| 文化概念同形异义 | “Two-Spirit”(北美原住民性别概念) | mountain + spirit 在英文诗歌/性别研究文本 |
词组跨语义场漂移 |
| 新闻热点污染 | 维基百科强奸案条目 | mountain + plain + 英文长尾词 |
热点事件占位新闻 |
1.2 触发条件矩阵
自动 agent 跨语种检索命中率 ≤30% 时即进入”噪声主导态”。本轮阈值基于 #18 三轮数据反推:
– 关键词含 mountain/altitude/station/observatory 时 ASCII 噪声概率 ≈ 60%
– 关键词含中文站名(”长白山” / “峨眉山” / “纳木错”)+ 英文专业词时,命中率反而高
– domain_filter=pubmed,mdpi 限制过严反而压制有效命中(loss > gain)
二、词条分类矩阵(前置过滤器)
每次检索前,先把目标关键词按下列矩阵分类:
| 维度 | 类型 | 过滤动作 |
|---|---|---|
| 字符集 | 纯中文 / 纯英文 / 中英混排 / 罗马化希腊字母 | 纯中文关键词禁止英文回退命中 |
| 语义场 | 专业术语 / 通用名词 / 文化概念 / 人名/品牌 | 通用名词 + 人名/品牌加入黑名单 |
| 多义性 | 单一专业义 / 多义(同名异地/同形异义) | 多义关键词必须加领域限定符 |
| 站名/观测点名 | 已知 Schumann 站 vs 未列入白名单 | 仅白名单站名(Portishead/Nagycenk/Shillong/Modra/Arrival Heights)允许直检 |
2.1 Schumann 研究已知观测站白名单
仅以下 7 个观测站允许直检:
1. Portishead (UK) — 海拔 ~sea level
2. Nagycenk (HU) — 海拔 ~150m
3. Shillong (IN) — 海拔 ~1500m
4. Modra (SK) — 海拔 ~530m
5. Arrival Heights (NZ-operated, Antarctica) — 海拔 ~sea level
6. Sondre Stromfjord (GL) — 海拔 ~sea level
7. L’Aquila (IT) — 海拔 ~700m
未列入白名单的”高海拔站”候选需人工 PubMed §1.1.6 二次确认。
三、三轨检索流程
每轮自动 agent 检索上限 = 6 次 web_search。超出即登记 blocked,转人工。
3.1 轨 A:中文专业术语优先(强制首检)
检索式: "<中文术语>" + "观测站" OR "台站" OR "测量"
例: "舒曼共振 山区 平原 对比"
期望命中: CNKI/万方/中文期刊文摘/equsci.org.cn
中文轨的语义场干净度远高于英文轨(无 Schwarzenegger 同形问题),应作为基线。
3.2 轨 B:英文术语 + 站名强制锚定
检索式: "Schumann resonance" + "<白名单站名 1>" + "<白名单站名 2>"
例: "Schumann resonance" + "Shillong" + "Portishead"
期望命中: AGU/Wiley/MDPI/Frontiers
站名锚定的作用:把多义关键词 mountain/observatory 从噪声中隔离,仅保留与白名单站共现的命中。
3.3 轨 C:DOI 直查 + Semantic Scholar API
轨 A/B 失败后启动:
– 对已知 Schumann 综述(如 MDPI 2072-4292/15/14/3557、AGU 10.1029/2022JD036582)的 forward citation 做 Semantic Scholar 抓取
– 直接查 Sentman/Nickolaenko/Price/Sátori 等核心作者主页
轨 C 失败 → 立即登记 blocked,转人工 PubMed §1.1.1–1.1.6。
四、失败判定与升级路径
4.1 失败判定阈值
| 指标 | 阈值 | 触发动作 |
|---|---|---|
| 命中率 | < 30%(4 次命中中 ≥ 3 次噪声) | 切换轨 |
| 字符集命中率 | 关键词中含中文字符但命中 100% ASCII 页 | 立即停止,切换轨 A |
| 站名命中率 | 关键词含白名单站名但命中 ≥ 50% 不相关 | 锁定到 DOI 直查(轨 C) |
| 总检索次数 | > 6 次/轮 | 登记 blocked,转人工 |
4.2 升级路径
自动 agent 6 次失败 → 写 blocked 登记到 backlog(如 #21 本轮)
↓ 7 天内
人工 PubMed §1.1.1–1.1.6 完成 → 回填 PMID 到 backlog
↓ 重启自动 agent
采用新 PMID 链 → 重新走轨 A → 轨 B → 轨 C
五、证据与局限
5.1 证据
- 本 SOP 是 #18/#20 三轮失败实录反推,未经过对照实验验证。
- 字符集白名单、站名白名单、三轨流程均为本轮首次系统化呈现。
- 阈值(30% 命中率、6 次/轮上限)来源于单点观测,不构成统计推断。
5.2 局限
- 不评估 Schumann 信号本身——本 SOP 是元方法学层,不触及环境频率对生命系统的接口效应。
- 白名单仅 7 站——可能漏掉高海拔站(如西藏/秘鲁/玻利维亚高原)的关键文献。
- 失败阈值未跨平台验证——SearXNG 不同节点命中率差异 ≥ 30%,单一节点失败阈值不能推广到所有节点。
- 未与 PRESS 2016 / PRISMA 2009 等金标准做形式化对齐——本 SOP 是工程实务层,与学术系统综述方法学有差距。
5.3 已知反证与降级条件
- 若轨 C 的 Semantic Scholar API 也返回 ASCII 噪声(即噪声污染已上溯到学术索引层),本 SOP 降级为”已知无效”,转人工 Cochrane-style 检索。
- 若 7 天内人工检索回填 < 1 篇相关 PMID,课题 #18/#20 整体降级为”假设级(E5)”,不进入 science-yixue/ 主索引。
六、相关阅读
- 五池宪章:
FIVE_POOLS_RESEARCH_CHARTER.md - 课题 #17(地形对 Schumann 谐振腔模态节点影响):
science-yixue/schumann-resonance-cavity-mode-topography.md - 课题 #20 文献检索骨架:
08-ops/experiments/pool4-schumann-mountain-vs-plain-literature-baseline.md - 课题 #14 山体 ELF 屏蔽系数:
science-yixue/mountain-elf-shielding-coefficient.md - 五池宪章 §3(每轮最低验收)、§6(慎重原则)
七、参考文献
- PMID 26678017 — McGowan J, Sampson M, Salzwedel DM, et al. PRESS Peer Review of Electronic Search Strategies: 2015 Guideline Statement. J Clin Epidemiol. 2016;75:40-46. DOI: 10.1016/j.jclinepi.2016.01.021
- PMID 19622511 — Moher D, Liberati A, Tetzlaff J, Altman DG; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 2009;6(7):e1000097. DOI: 10.1371/journal.pmed.1000097
八、版本
- v0.1(2026-10-12)— 第 21 轮首版;基于 #18/#20 三轮失败反推;待 #18 重启时实测验证
免责说明:本文属传统知识与科学文献的整理、对照与研究设计讨论,仅供理解与参考,不构成医疗建议,不能替代专业诊断与治疗。文中涉及的效应量、频率区间与暴露结论多为研究报告层面的推断,个体情况差异很大;如有身体不适或健康疑虑,请就医并由专业医师判断。
SearXNG Cross-Lingual Retrieval Term Noise Filtering SOP: Reproducible Retrieval Specification for Environment-Life Frequency Interface Research
Evidence level: D
Introduction
Five Pools ④ attempted automatic agent + SearXNG federal retrieval of Schumann mountain vs. plain station literature baselines (Project #18/#20) over the past 3 rounds (rounds 17/19/20), accumulating ≥13 web_search hits entirely on ASCII keyword noise: English names (Schwarzenegger), North American Indigenous “Two-Spirit” (a gender concept), Victorian rape news reports, etc.—all unrelated to the environment-life frequency interface theme.
This “keyword hit but semantic mismatch” contamination stems not from SearXNG itself, but from the automatic agent’s cross-lingual retrieval lacking (a) character set whitelists, (b) professional terminology romanization tables, (c) three-track retrieval workflows, and (d) failure escalation paths.
This SOP is the methodological deliverable distilled after the three failed rounds of #18/#20. It does not directly assess Schumann signals or environmental ELF biological effects—evidence boundaries are confined to the “retrieval process layer.” Any subsequent Five Pool projects relying on automatic retrieval (regardless of Pool ①–⑤) can reuse this SOP.
I. Problem Definition: ASCII Keyword Noise Contamination
1.1 Three Types of Noise Sources
| Category | Example (from #18 failure log) | Trigger Keyword | Contamination Mechanism |
|---|---|---|---|
| Person/Brand homonym | Arnold Schwarzenegger personal fitness station | Schumann → substring match Schwarzenegger |
Latin character substring matching |
| Cultural concept homonym | “Two-Spirit” (North American Indigenous gender concept) | mountain + spirit in English poetry/gender studies texts |
Phrase drift across semantic fields |
| News hotspot contamination | Wikipedia rape case entry | mountain + plain + English long-tail terms |
Hotspot event placeholder news |
1.2 Trigger Condition Matrix
Automatic agent cross-lingual retrieval hit rate ≤30% triggers “noise-dominated state.” This round’s threshold is reverse-engineered from #18 three-round data:
– Keywords containing mountain/altitude/station/observatory have ASCII noise probability ≈ 60%
– Keywords containing Chinese station names (“Changbai Mountain” / “Mount Emei” / “Namuco”) + English professional terms actually have higher hit rates
– domain_filter=pubmed,mdpi is too restrictive, suppressing valid hits (loss > gain)
II. Keyword Classification Matrix (Pre-Filter)
Before each retrieval, classify target keywords using the following matrix:
| Dimension | Type | Filter Action |
|---|---|---|
| Character Set | Pure Chinese / Pure English / Mixed Chinese-English / Romanized Greek letters | Pure Chinese keywords prohibit English fallback hits |
| Semantic Field | Professional term / Common noun / Cultural concept / Person/Brand | Add blacklist for common nouns + person/brand names |
| Polysemy | Single professional meaning / Multi-meaning (same name different place/homonym) | Multi-meaning keywords must add domain qualifiers |
| Station/Observatory Name | Known Schumann station vs. not on whitelist | Only whitelist stations (Portishead/Nagycenk/Shillong/Modra/Arrival Heights) allow direct search |
2.1 Schumann Research Known Observatory Whitelist
Only the following 7 observatories allow direct search:
1. Portishead (UK) — altitude ~sea level
2. Nagycenk (HU) — altitude ~150m
3. Shillong (IN) — altitude ~1500m
4. Modra (SK) — altitude ~530m
5. Arrival Heights (NZ-operated, Antarctica) — altitude ~sea level
6. Sondre Stromfjord (GL) — altitude ~sea level
7. L’Aquila (IT) — altitude ~700m
Unlisted “high-altitude station” candidates require manual PubMed §1.1.6 secondary confirmation.
III. Three-Track Retrieval Workflow
Each round’s automatic agent retrieval limit = 6 web_searches. Exceeding this registers blocked and transfers to manual.
3.1 Track A: Chinese Professional Terms Priority (Mandatory First Check)
Search query: "<Chinese term>" + "observatory" OR "station" OR "measurement"
Example: "Schumann resonance mountain plain comparison"
Expected hits: CNKI/Wanfang/Chinese journal abstracts/equsci.org.cn
The semantic field cleanliness of the Chinese track far exceeds that of the English track (no Schwarzenegger homonym issue), so it should serve as the baseline.
3.2 Track B: English Terms + Station Name Mandatory Anchoring
Search query: "Schumann resonance" + "<whitelist station name 1>" + "<whitelist station name 2>"
Example: "Schumann resonance" + "Shillong" + "Portishead"
Expected hits: AGU/Wiley/MDPI/Frontiers
Station name anchoring serves to isolate polysemous keywords mountain/observatory from noise, retaining only hits co-occurring with whitelist stations.
3.3 Track C: DOI Direct Lookup + Semantic Scholar API
Initiated after Track A/B failure:
– Forward citation retrieval of known Schumann reviews (e.g., MDPI 2072-4292/15/14/3557, AGU 10.1029/2022JD036582) via Semantic Scholar
– Direct lookup of core author pages (Sentman/Nickolaenko/Price/Sátori, etc.)
Track C failure → immediately register blocked, transfer to manual PubMed §1.1.1–1.1.6.
IV. Failure Determination and Escalation Path
4.1 Failure Thresholds
| Metric | Threshold | Trigger Action |
|---|---|---|
| Hit rate | < 30% (≥3 of 4 hits are noise) | Switch track |
| Character set hit rate | Keywords contain Chinese characters but 100% ASCII page hits | Stop immediately, switch to Track A |
| Station name hit rate | Keywords contain whitelist station names but ≥50% irrelevant hits | Lock to DOI direct lookup (Track C) |
| Total retrieval count | > 6 per round | Register blocked, transfer to manual |
4.2 Escalation Path
Automatic agent 6 failures → write blocked entry to backlog (e.g., #21 this round)
↓ within 7 days
Manual PubMed §1.1.1–1.1.6 completion →回填 PMID to backlog
↓ restart automatic agent
Adopt new PMID chain → re-run Track A → Track B → Track C
V. Evidence and Limitations
5.1 Evidence
- This SOP is reverse-engineered from the three failed rounds of #18/#20, not validated by controlled experiments.
- Character set whitelist, station name whitelist, and three-track workflow are presented systematically for the first time in this round.
- Thresholds (30% hit rate, 6 per round limit) derive from single-point observation and do not constitute statistical inference.
5.2 Limitations
- Does not assess Schumann signals themselves—this SOP is at the meta-methodological layer, not touching environmental frequency interface effects on living systems.
- Whitelist includes only 7 stations—may miss key literature for high-altitude stations (e.g., Tibet, Peru, Bolivian plateau).
- Failure thresholds not cross-platform validated—SearXNG node hit rate differences ≥30%; single-node failure thresholds cannot be generalized to all nodes.
- Not formally aligned with PRESS 2016 / PRISMA 2009 gold standards—this SOP is at the engineering practice layer, with gaps compared to academic systematic review methodology.
5.3 Known Counter-Evidence and Degradation Conditions
- If Track C’s Semantic Scholar API also returns ASCII noise (i.e., noise contamination has propagated to the academic indexing layer), this SOP degrades to “known invalid” and transfers to manual Cochrane-style retrieval.
- If manual retrieval within 7 days yields <1 relevant PMID, Project #18/#20 overall degrades to “hypothesis level (E5)” and does not enter the science-yixue/ main index.
VI. Related Reading
- Five Pools Charter:
FIVE_POOLS_RESEARCH_CHARTER.md - Project #17 (terrain impact on Schumann cavity mode nodes):
science-yixue/schumann-resonance-cavity-mode-topography.md - Project #20 literature retrieval skeleton:
08-ops/experiments/pool4-schumann-mountain-vs-plain-literature-baseline.md - Project #14 mountain ELF shielding coefficient:
science-yixue/mountain-elf-shielding-coefficient.md - Five Pools Charter §3 (minimum acceptance per round), §6 (caution principle)
VII. References
- PMID 26678017 — McGowan J, Sampson M, Salzwedel DM, et al. PRESS Peer Review of Electronic Search Strategies: 2015 Guideline Statement. J Clin Epidemiol. 2016;75:40-46. DOI: 10.1016/j.jclinepi.2016.01.021
- PMID 19622511 — Moher D, Liberati A, Tetzlaff J, Altman DG; PRISMA Group. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. 2009;6(7):e1000097. DOI: 10.1371/journal.pmed.1000097
VIII. Version
- v0.1 (2026-10-12) — first edition of round 21; reverse-engineered from #18/#20 three-round failures; awaits real-world validation when #18 restarts.
Disclaimer: This document constitutes an organized comparison and research design discussion of traditional knowledge and scientific literature, intended solely for understanding and reference. It does not constitute medical advice and cannot replace professional diagnosis and treatment. Effect sizes, frequency ranges, and exposure conclusions mentioned herein are largely inferences at the research report level; individual variation is significant. If you experience discomfort or health concerns, please consult a physician for professional judgment.
Leave a Reply