clearai-dsh 0.3.1 → 0.5.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (150) hide show
  1. package/CHANGELOG.md +172 -0
  2. package/README.md +30 -30
  3. package/README.zh-CN.md +30 -30
  4. package/bin/clearai.mjs +29 -231
  5. package/cordis.patch.yml +3 -3
  6. package/lib/client.js +862 -2555
  7. package/lib/domain-language.js +614 -48
  8. package/lib/fold.js +449 -1020
  9. package/lib/host.js +47 -557
  10. package/lib/invariant.js +9 -12
  11. package/lib/knowledge-view.js +421 -225
  12. package/lib/lang.js +81 -0
  13. package/package.json +2 -2
  14. package/presets/clearai/agent.cordis.yml +57 -82
  15. package/presets/clearai/clearai.patch.yml +57 -82
  16. package/presets/clearai/plugins/clearai-kernel.js +2103 -5457
  17. package/presets/clearai/plugins/ontology.js +76 -20
  18. package/presets/clearai/plugins/prompts.js +75 -363
  19. package/presets/clearai/skills/clearai-loop/SKILL.md +81 -59
  20. package/presets/clearai/plugins/brain.js +0 -547
  21. package/presets/clearai/plugins/commands.js +0 -199
  22. package/presets/clearai/template/knowledge/README.md +0 -25
  23. package/presets/clearai/template/memory/README.md +0 -34
  24. package/presets/clearai/template/project.md +0 -49
  25. package/presets/clearai/template/skills/README.md +0 -37
  26. package/presets/clearai/template/skills/chart-diagram-qa/SKILL.md +0 -43
  27. package/presets/clearai/template/skills/citation-management/SKILL.md +0 -73
  28. package/presets/clearai/template/skills/citation-management/references/bibtex_formatting.md +0 -908
  29. package/presets/clearai/template/skills/citation-management/references/citation_validation.md +0 -794
  30. package/presets/clearai/template/skills/citation-management/references/google_scholar_search.md +0 -725
  31. package/presets/clearai/template/skills/citation-management/references/metadata_extraction.md +0 -870
  32. package/presets/clearai/template/skills/citation-management/references/pubmed_search.md +0 -839
  33. package/presets/clearai/template/skills/citation-management/scripts/doi_to_bibtex.py +0 -204
  34. package/presets/clearai/template/skills/citation-management/scripts/extract_metadata.py +0 -569
  35. package/presets/clearai/template/skills/citation-management/scripts/format_bibtex.py +0 -349
  36. package/presets/clearai/template/skills/citation-management/scripts/generate_schematic.py +0 -139
  37. package/presets/clearai/template/skills/citation-management/scripts/generate_schematic_ai.py +0 -817
  38. package/presets/clearai/template/skills/citation-management/scripts/search_google_scholar.py +0 -282
  39. package/presets/clearai/template/skills/citation-management/scripts/search_pubmed.py +0 -398
  40. package/presets/clearai/template/skills/citation-management/scripts/validate_citations.py +0 -497
  41. package/presets/clearai/template/skills/data-analysis/SKILL.md +0 -92
  42. package/presets/clearai/template/skills/data-analysis/checklists/readiness_check.md +0 -23
  43. package/presets/clearai/template/skills/data-analysis/templates/analysis_report.md.tpl +0 -63
  44. package/presets/clearai/template/skills/data-analysis/templates/cleaning_rules_draft.yaml.tpl +0 -32
  45. package/presets/clearai/template/skills/data-analysis/templates/data_dictionary.md.tpl +0 -12
  46. package/presets/clearai/template/skills/data-analysis/templates/domain_knowledge_template.md.tpl +0 -316
  47. package/presets/clearai/template/skills/data-analysis/templates/feature_candidates.json.tpl +0 -20
  48. package/presets/clearai/template/skills/data-analysis/templates/quality_scorecard.md.tpl +0 -30
  49. package/presets/clearai/template/skills/data-analysis/workflows/01-data-profiling.md +0 -42
  50. package/presets/clearai/template/skills/data-analysis/workflows/02-quality-audit.md +0 -36
  51. package/presets/clearai/template/skills/data-analysis/workflows/03-physical-correlation.md +0 -25
  52. package/presets/clearai/template/skills/data-analysis/workflows/04-unstructured-mining.md +0 -26
  53. package/presets/clearai/template/skills/data-qa-analysis/SKILL.md +0 -102
  54. package/presets/clearai/template/skills/data-qa-analysis/checklists/readiness_check.md +0 -62
  55. package/presets/clearai/template/skills/data-qa-analysis/templates/best_in_class_report.md.tpl +0 -56
  56. package/presets/clearai/template/skills/data-qa-analysis/templates/cleaning_rules_draft.yaml.tpl +0 -56
  57. package/presets/clearai/template/skills/data-qa-analysis/templates/data_dictionary.md.tpl +0 -13
  58. package/presets/clearai/template/skills/data-qa-analysis/templates/data_source_inventory_and_lineage.md.tpl +0 -146
  59. package/presets/clearai/template/skills/data-qa-analysis/templates/data_status_report.md.tpl +0 -60
  60. package/presets/clearai/template/skills/data-qa-analysis/templates/steady_state_rules.yaml.tpl +0 -41
  61. package/presets/clearai/template/skills/data-qa-analysis/templates/subsystem_registry.md.tpl +0 -101
  62. package/presets/clearai/template/skills/data-qa-analysis/templates/unified_execution_plan.md.tpl +0 -100
  63. package/presets/clearai/template/skills/data-qa-analysis/workflows/01-data-source-inventory-and-lineage.md +0 -194
  64. package/presets/clearai/template/skills/data-qa-analysis/workflows/02-data-alignment-and-tag-semantics.md +0 -122
  65. package/presets/clearai/template/skills/data-qa-analysis/workflows/03-steady-state-identification.md +0 -126
  66. package/presets/clearai/template/skills/data-qa-analysis/workflows/04-consumption-analysis.md +0 -152
  67. package/presets/clearai/template/skills/data-qa-analysis/workflows/05-best-in-class-and-optimization-space.md +0 -78
  68. package/presets/clearai/template/skills/domain-presearch/SKILL.md +0 -131
  69. package/presets/clearai/template/skills/domain-presearch/checklists/domain_checklist.md +0 -24
  70. package/presets/clearai/template/skills/domain-presearch/references/figure_code.md +0 -78
  71. package/presets/clearai/template/skills/domain-presearch/references/strategic_frameworks.md +0 -38
  72. package/presets/clearai/template/skills/exploration-loop/SKILL.md +0 -81
  73. package/presets/clearai/template/skills/exploratory-data-analysis/SKILL.md +0 -77
  74. package/presets/clearai/template/skills/exploratory-data-analysis/references/bioinformatics_genomics_formats.md +0 -664
  75. package/presets/clearai/template/skills/exploratory-data-analysis/references/chemistry_molecular_formats.md +0 -664
  76. package/presets/clearai/template/skills/exploratory-data-analysis/references/general_scientific_formats.md +0 -518
  77. package/presets/clearai/template/skills/exploratory-data-analysis/references/microscopy_imaging_formats.md +0 -620
  78. package/presets/clearai/template/skills/exploratory-data-analysis/references/proteomics_metabolomics_formats.md +0 -517
  79. package/presets/clearai/template/skills/exploratory-data-analysis/references/spectroscopy_analytical_formats.md +0 -633
  80. package/presets/clearai/template/skills/exploratory-data-analysis/scripts/eda_analyzer.py +0 -547
  81. package/presets/clearai/template/skills/hypothesis-generation/SKILL.md +0 -73
  82. package/presets/clearai/template/skills/hypothesis-generation/references/experimental_design_patterns.md +0 -329
  83. package/presets/clearai/template/skills/hypothesis-generation/references/hypothesis_quality_criteria.md +0 -198
  84. package/presets/clearai/template/skills/hypothesis-generation/references/literature_search_strategies.md +0 -622
  85. package/presets/clearai/template/skills/hypothesis-generation/scripts/generate_schematic.py +0 -139
  86. package/presets/clearai/template/skills/hypothesis-generation/scripts/generate_schematic_ai.py +0 -817
  87. package/presets/clearai/template/skills/literature-review/SKILL.md +0 -72
  88. package/presets/clearai/template/skills/literature-review/references/citation_styles.md +0 -166
  89. package/presets/clearai/template/skills/literature-review/references/database_strategies.md +0 -455
  90. package/presets/clearai/template/skills/literature-review/scripts/generate_pdf.py +0 -176
  91. package/presets/clearai/template/skills/literature-review/scripts/generate_schematic.py +0 -139
  92. package/presets/clearai/template/skills/literature-review/scripts/generate_schematic_ai.py +0 -817
  93. package/presets/clearai/template/skills/literature-review/scripts/search_databases.py +0 -303
  94. package/presets/clearai/template/skills/literature-review/scripts/verify_citations.py +0 -221
  95. package/presets/clearai/template/skills/paper-lookup/SKILL.md +0 -59
  96. package/presets/clearai/template/skills/paper-lookup/references/arxiv.md +0 -161
  97. package/presets/clearai/template/skills/paper-lookup/references/biorxiv.md +0 -118
  98. package/presets/clearai/template/skills/paper-lookup/references/core.md +0 -150
  99. package/presets/clearai/template/skills/paper-lookup/references/crossref.md +0 -181
  100. package/presets/clearai/template/skills/paper-lookup/references/medrxiv.md +0 -104
  101. package/presets/clearai/template/skills/paper-lookup/references/openalex.md +0 -174
  102. package/presets/clearai/template/skills/paper-lookup/references/pmc.md +0 -152
  103. package/presets/clearai/template/skills/paper-lookup/references/pubmed.md +0 -124
  104. package/presets/clearai/template/skills/paper-lookup/references/semantic-scholar.md +0 -203
  105. package/presets/clearai/template/skills/paper-lookup/references/unpaywall.md +0 -127
  106. package/presets/clearai/template/skills/process-presearch/SKILL.md +0 -196
  107. package/presets/clearai/template/skills/process-presearch/checklists/process_checklist.md +0 -18
  108. package/presets/clearai/template/skills/process-presearch/references/figure_code.md +0 -107
  109. package/presets/clearai/template/skills/process-presearch/references/source_attribution_example.md +0 -22
  110. package/presets/clearai/template/skills/process-understanding-extraction/SKILL.md +0 -69
  111. package/presets/clearai/template/skills/process-understanding-extraction/checklists/readiness_check.md +0 -34
  112. package/presets/clearai/template/skills/process-understanding-extraction/templates/docx_raw_dump_extractor.py.tpl +0 -132
  113. package/presets/clearai/template/skills/process-understanding-extraction/templates/entity_map_unit_topology.json.tpl +0 -86
  114. package/presets/clearai/template/skills/process-understanding-extraction/templates/process_brief.md.tpl +0 -89
  115. package/presets/clearai/template/skills/process-understanding-extraction/templates/process_brief_builder_from_raw_dump.py.tpl +0 -203
  116. package/presets/clearai/template/skills/process-understanding-extraction/templates/process_flow_mermaid.md.tpl +0 -41
  117. package/presets/clearai/template/skills/process-understanding-extraction/templates/unified_execution_plan.md.tpl +0 -53
  118. package/presets/clearai/template/skills/process-understanding-extraction/workflows/01-process-doc-discovery.md +0 -173
  119. package/presets/clearai/template/skills/process-understanding-extraction/workflows/02-process-understanding-and-diagramming.md +0 -106
  120. package/presets/clearai/template/skills/scientific-brainstorming/SKILL.md +0 -64
  121. package/presets/clearai/template/skills/scientific-brainstorming/references/brainstorming_methods.md +0 -326
  122. package/presets/clearai/template/skills/scientific-critical-thinking/SKILL.md +0 -72
  123. package/presets/clearai/template/skills/scientific-critical-thinking/references/common_biases.md +0 -364
  124. package/presets/clearai/template/skills/scientific-critical-thinking/references/evidence_hierarchy.md +0 -485
  125. package/presets/clearai/template/skills/scientific-critical-thinking/references/experimental_design.md +0 -496
  126. package/presets/clearai/template/skills/scientific-critical-thinking/references/logical_fallacies.md +0 -478
  127. package/presets/clearai/template/skills/scientific-critical-thinking/references/scientific_method.md +0 -169
  128. package/presets/clearai/template/skills/scientific-critical-thinking/references/statistical_pitfalls.md +0 -506
  129. package/presets/clearai/template/skills/skill-creator/SKILL.md +0 -109
  130. package/presets/clearai/template/skills/skill-creator/references/authoring-guide.md +0 -89
  131. package/presets/clearai/template/skills/statistical-analysis/SKILL.md +0 -79
  132. package/presets/clearai/template/skills/statistical-analysis/references/assumptions_and_diagnostics.md +0 -369
  133. package/presets/clearai/template/skills/statistical-analysis/references/bayesian_statistics.md +0 -653
  134. package/presets/clearai/template/skills/statistical-analysis/references/effect_sizes_and_power.md +0 -578
  135. package/presets/clearai/template/skills/statistical-analysis/references/reporting_standards.md +0 -469
  136. package/presets/clearai/template/skills/statistical-analysis/references/test_selection_guide.md +0 -129
  137. package/presets/clearai/template/skills/statistical-analysis/scripts/assumption_checks.py +0 -538
  138. package/presets/clearai/template/skills/web-artifact/SKILL.md +0 -165
  139. package/presets/clearai/template/skills/web-artifact/assets/renderer/renderer.css +0 -229
  140. package/presets/clearai/template/skills/web-artifact/assets/renderer/renderer.js +0 -373
  141. package/presets/clearai/template/skills/web-artifact/assets/vendor/elkjs/LICENSE +0 -263
  142. package/presets/clearai/template/skills/web-artifact/assets/vendor/elkjs/UPSTREAM.md +0 -26
  143. package/presets/clearai/template/skills/web-artifact/assets/vendor/elkjs/elk.bundled.js +0 -6605
  144. package/presets/clearai/template/skills/web-artifact/references/when-drawing-a-topology.md +0 -150
  145. package/presets/clearai/template/skills/web-artifact/references/when-the-page-must-work-offline.md +0 -62
  146. package/presets/clearai/template/skills/web-artifact/scripts/check_artifact.py +0 -167
  147. package/presets/clearai/template/skills/web-artifact/scripts/render_topology.js +0 -272
  148. package/presets/clearai/template/skills/what-if-oracle/LICENSE.txt +0 -5
  149. package/presets/clearai/template/skills/what-if-oracle/SKILL.md +0 -72
  150. package/presets/clearai/template/skills/what-if-oracle/references/scenario-templates.md +0 -154
package/CHANGELOG.md CHANGED
@@ -2,6 +2,178 @@
2
2
 
3
3
  All notable changes to this project are recorded here. The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
4
4
 
5
+ ## [Unreleased]
6
+
7
+ ## [0.5.0] — 2026-10-08
8
+
9
+ 这一版是 0.5 升级的阶段一与阶段二:预期与未解释项、评估者查数据可信、不可逆命令拦在执行前、三档等级、经验与本体分类,宿主下界提到 DSH 0.2。
10
+
11
+ **开发集上的初步结果**(盲评,不当作有效性证据,见 [docs/optimization/sim-runs/](docs/optimization/sim-runs/)):反应器题漂移纠正 ClearAI 3/4、裸模型 1/4;同一现场 3 题序列里第 3 题全部用上了前两题的经验,第 3 题均分 1.5 对 1.17,差距来自漂移这一项;配方与探索深度两边持平。正式验证和探索深度([设计草稿](docs/optimization/0.5-plan/exploration-depth.zh-CN.md))留到之后的版本。
12
+
13
+ ### Added
14
+
15
+ - **预期与未解释项(反常)**:步骤可以带预期 `expect`(预计看到什么、从哪条关系或经验来),动手前也能用 `RevisePlan(action="expect")` 补写。结果和预期或本体对不上的地方,交付时写进 `AdvancePlan.anomalies`,或随时用新工具 `Anomaly` 登记;它们挂在卡上,只有三个去处:被解释、写明理由排除、交给人。不阻塞结案。`STATE_VERSION` 17(投影多了 `anomalies`)。
16
+ - **评估者查还有什么没解释**:任务书列出未解释项,纪律要求查原始数据里没登记的异常(参考读数、重复点漂移、装置之间的差)和排除理由站不站得住;裁决多一个 `anomalies` 字段,折成「评估者发现」的未解释项退回给模型。
17
+ - **不可逆动作拦在命令上**:`Frame` 的 `irreversible` 声明动作和命令特征,匹配的 bash 命令执行前当场问人;不放行或没人能答就拒。
18
+ - **经验(跨任务的「下次怎么做」)**:`Conclude(outcome="achieved")` 多一个可选的 `lessons`(每条一句,分坑 / 先核 / 会骗人的捷径 / 先验四类,带涉及的装置或量、凭据、不适用的条件)。结案评估者对照记录逐条判(裁决多一个 `lessons` 字段),支持的由系统写进 `clear/knowledge/lessons/<id>.json`(做的人写不了),其余在回执里列出。别的会话经工作区同步读到它们;卡上在立题、定计划、写预期前摆出最多 6 条,和眼下目标、下一步提到同一装置或量的排前面。人把文件里的 `status` 改成 `retracted` 就撤回。`STATE_VERSION` 18(投影多了 `lessons`,派生多了 `lessonRows`)。
19
+ - **本体只留能产出预期的东西**:概念分类别 / 度量 / 现象(`kind`),度量必须写 `unit`;关系分影响 / 定义 / 测量 / 表现为(`kind`),影响关系可写大致形状 `shape`(单调升、单调降、有峰、阈值、耦合),测量关系写读数怎么核(`check`);实体可带带日期的 `history`。卡上「用到的关系」把形状和核法递出来;词面命中也看下一步要做什么和它的预期。都是可选格,旧文件照旧有效,指纹不变。
20
+ - 0.5 的设计与升级计划:[meta 设计](docs/optimization/0.5-plan/meta-design.zh-CN.md)、[升级计划](docs/optimization/0.5-plan/upgrade-plan.zh-CN.md)。
21
+
22
+ ### Changed
23
+
24
+ - 本体与认识论状态在决策点递到模型眼前:运行态卡给出判断用到的概念释义和相关事实的原话与边界;目标还没立或已结案时,直接列出以前留下的事实。
25
+ - 本体的提示改为只写推理要用的东西:度量写成概念、释义写口径;因果关系写进关系;实体只在结论要指认它时才建。
26
+ - 独立评估者可以复跑(`auditRerun`,预设打开):评估者在工作区副本里跑产物自带的脚本核数字,原工作区只读;评估期间收下的产物被改动过,这份裁决不认。
27
+ - **等级只给三档**:L2 自己检验、L3 独立评估、L4 人放行。L0–L2 的细分从没改变过谁来判,合并;旧账里的 L0、L1 读作自己检验,新写的收成 L2。
28
+
29
+ - **宿主下界提到 DSH `0.2.0-rc.2`(`engines.dsh` 为 `>=0.2.0-rc.2`),更早的宿主不再支持。** ClearAI 用到的事件、服务方法、客户端插座和 `dsh plugin` 命令行在 0.2.x 上都没有破坏性变化,运行时代码不用改;改的是只在旧的 root 目录名册上才有意义的工具和文档。
30
+ - **`install.sh`(开发形态)改走产品那条路**:构建出包,再 `dsh plugin add` 装进 profile。旧做法把预设拷进 `~/.dsh/.agent-presets`,名册不读它,装得上却选不到 ClearAI。
31
+ - **`verify-deploy` 只验装出来的那个包**,预期的工具数与段数从部署的目录现算;`verify-clean-install` / `verify-lifecycle` 只认声明行;`e2e-run` 只在宿主还带着 `@deepseek-ai/dsh-invariants` 时才插那一行(`0.2.1-alpha.1` 已移除该包)。
32
+
33
+ ### Removed
34
+
35
+ - 对手判断(`rival` / `split_by`)与「检验可能太弱」的提示:两道难题的 8 组运行里对手 0 次使用,提示连续出现也没改变做法。
36
+ - 运行态卡上的本体提纲与计数:它从没改变过下一步。卡上只出现被判断引用到的本体项。
37
+ - `clearai-dsh root-yaml` / `seed` / `unseed`:它们服务的是旧名册的目录扫描,0.2.x 宿主里没有这个名册。`doctor` 改为检查组合里有没有 `preset-clearai` 声明行。
38
+
39
+ ### Fixed
40
+
41
+ - 判断几乎从不升格成事实:升格只看模型自己给步骤标的等级,而 L0–L2 允许自判,五个长程任务的 9 次运行里模型一律标 L1/L2,一条事实都没有。现在结案那次独立评估顺带逐条判还活着的判断,判为支持的升格(证据记为独立、等级按目标门槛),判为推翻的不升格并照常复核已有事实;不多派评估者。
42
+ - 别的会话留下的事实被新证据推翻时不会问人:这一问只认本目标的判断编号。现在判断可带 `retests: "<事实 id>"`,复检被推翻就当场问人撤回还是维持,结论写回事实文件。
43
+ - 复检旧事实时模型把判断写成「旧事实被推翻」,评估判支持后它升格成一条与旧事实相反的新事实,旧事实照旧「已确立」,也没人被问。`retests` 的说明和提示词现在要求 claim 照抄旧事实原来的说法。
44
+
45
+ ## [0.4.0] — 2026-10-02
46
+
47
+ 这一版包括「少即是多」改造的全部六个阶段(见下面各段)、本体改成文件树,以及中英两种语言的完整支持。
48
+
49
+ **两种语言。** 系统自己写的每一句话都跟着人说话的语言走。`STATE_VERSION` 16(投影多了 `language`);旧会话不迁移。
50
+
51
+ ### Added
52
+
53
+ - **语言跟着人走**:人用中文写就是中文,用英文写就是英文,按会话、按人写的消息判(系统注入的消息不算)。覆盖工具结果、运行态卡、问人的话(选项按两种语言都认)、三段提示词、六件工具的说明与参数说明(英文会话在装配提示词时换成英文版)、评估者任务书与裁决 schema、技能、`clear/` 下系统写的文件(目标文档、事实总览、阶段归档、过程本体货架)、本体文件的校验信息。面板跟着界面语言,命令行跟着 `--lang` 或系统语言。
54
+ - **两个真实案例**:[JEPA 世界模型](docs/cases/jepa-world-model.zh-CN.md)与 [Navier–Stokes](docs/cases/navier-stokes.zh-CN.md),原始记录在 `docs/cases/runs/`,中英截图在 `docs/shots/{zh,en}/`(在真 DSH 里回放两场真跑拍的)。
55
+
56
+ ### Changed
57
+
58
+ - **文档只写现在的设计**:删掉改造方案、诊断与过程记录,旧的三个虚构案例换成两个真实案例。
59
+
60
+ **本体改成文件树。** 设计见 [领域本体](docs/domain-ontology.zh-CN.md) 第 5、6 节。
61
+
62
+ ### Changed
63
+
64
+ - **本体是 `clear/ontology/` 下的 JSON 文件树**:`concepts/`、`relations/`、`entities/`,`X.json` 描述 X,孩子放在同级 `X/` 目录(概念嵌套 = is_a,实体嵌套 = 组成)。模型用原生文件工具写,系统写一份 `SCHEMA.json` 说明格式。
65
+ - **三道检查**:写时单文件校验(不合格拒写);读时跨文件校验(只标不拦,问题列在卡上与图下);升格时全量校验(不合格不升格)。
66
+ - **事实写成文件并跨会话累积**:每条一份 `clear/knowledge/facts/<id>.json`,带所用定义的指纹;定义改了,这条事实标「定义已变」进「待处理」。`INDEX.md` 从全部事实文件渲染。
67
+ - **图上目录嵌套成可收起的子图**:点多时默认收到第一层,+N 展开。
68
+ - 结案后提示模型把可复用的做法写成原生技能(`.agents/skills/<名字>/SKILL.md`)。
69
+
70
+ ### Fixed
71
+
72
+ - 后一次 `Conclude` 会把之前目标的假设再升格一次。
73
+ - 相对路径能绕过受保护目录;读受保护目录的文件也被拦。
74
+
75
+ ### Removed
76
+
77
+ - 4 件本体工具 `Define` / `Deprecate` / `RegisterInstance` / `Assert`(工具 10 → 6),以及 `clear/ontology/domain.md` 读面。
78
+
79
+
80
+ **「少即是多」第六阶段:呈现。** 按改造方案第六阶段的改动清单。机制、十件工具、账本事件与 `STATE_VERSION` 都不变。
81
+
82
+ ### Changed
83
+
84
+ - **本体格改成图为主。** 页眉一行问题、一行计数、「待处理」和一条进度轨(判断 → 检验 → 已验证 → 入本体);中间是本体图 / 实体图二选一(星图式,已验证的关系实线、待核验虚线);下面是结论清单,一条一行,按 已验证 / 待核验 / 验证中 / 不确定 / 已推翻 / 已替换 分组。点开一条依次是 进度、可信度怎么变的、补充。
85
+ - **判断有短名。** `Frame` 的每条判断可带 `name`(十二字以内),没给就取主张开头;工具里引用判断时 id、短名、主张原文都认。账本里的 id 不变。
86
+ - **说人话。** 运行态卡与工具结果不再出现内部编号与 `support / refute`;单次结果只说 支持 / 推翻 / 不确定(`AdvancePlan` 也收这三个中文词),等级写成 自己推了一遍 / 引用已有材料 / 可复算 / 独立核验 / 人放行。卡没变就不再附。
87
+ - **世界树**改成目录式:每步一行,点开才看检验了哪条判断、结果是什么;页眉不带计划编号。
88
+ - 英文界面里右栏页签叫 World Tree(此前误作 Worldlines),结果与状态词首字母大写。
89
+
90
+ ### Removed
91
+
92
+ - 面板上的命题货架、缺口栏、状态流转图、词汇维护区与本体编辑抽屉;`/api/clearai/gate` 整条路由(人门自第三阶段起由开门的那次调用当场问人)。
93
+ - 收件箱:换成派生的「待处理」(计划连拦停下、结论互相矛盾),只陈述、不放按钮;本体格页眉与输入框旁「待处理 N」读同一份。
94
+
95
+
96
+ **「少即是多」第五阶段:提示词收成三段。** 按改造方案第五阶段的改动清单。
97
+
98
+ ### Changed
99
+
100
+ - **提示词 21 段 → 3 段**,约 1.5 万字收到 2 千字出头:身份(`clearai/identity`)、循环(`clearai/loop`)、对人说话(`clearai/speaking`)。每个工具怎么用写在工具自己的说明里,现在是什么状态由运行态卡给。
101
+ - 人格前缀只留一句身份;`clearai-loop` 技能只讲提示词与工具说明都没写的部分:准入核什么、被拦之后怎么走、评估卡怎么读。
102
+
103
+ ### Removed
104
+
105
+ - 讲原生工具用法的段(文件读写、网页、技能、委派、Python 环境)、已删机制的残留说法(`lab/` 与 `products/` 目录约定、`PROJECT.md` 占位、`setup_cjk()`、`retract_fact` / `keep_fact`),以及只由提示词承载的「引擎级异常降权只读恢复」细则(身份段留一句:结局不明先看当前事实)。机制、工具、状态形状都不变。
106
+
107
+ **「少即是多」第四阶段:Conclude 门收紧,工具面收拢。** 按改造方案第四阶段的改动清单。
108
+
109
+ ### Changed
110
+
111
+ - **Conclude 只剩一道实体门**(`requireLandedEntities`),而且只管**将要升格**的判断(存活、零推翻、支持等级够升格):它断言里的每个主体都得是实体图上的实例节点(登记过的实例,或 Assert / 已升格事实带出来的节点)。「落图」从「要有一条边」改成「节点在」,所以 `RegisterInstance` 一次就够,不必再 `Assert` 同一句话。门在派评估者之前判,拦下不花评估。
112
+ - **缺口 7 → 3**:留下 `untouched_claims`、`prose_only_claims`、`entities_unlanded`。`entities_unlanded` 的读数列出具体的 `类型|id`,出路写明 `RegisterInstance` 或 Frame 修订。
113
+ - **工具 19 → 10**:`Frame`、`Conclude`、`CreatePlan`、`AdvancePlan`、`RevisePlan`、`ClosePlan`、`Define`、`Deprecate`、`RegisterInstance`、`Assert`。`RevisePlan` 以 `action: add | refine | void` 合并了改计划、补判据、作废一步;`Define` 给了 `range` 是谓词、否则是概念,同 id 再 Define 是修订展示字段,改语义(上位、定义域、值域、函数性)拒为 `semantics_changed`;`Deprecate` 兼管概念与谓词。账本事件名不变。
114
+ - 投影状态版本升到 14:假设带 `unlanded`,不再有 `skips` / `untouchedLevels`。
115
+
116
+ ### Removed
117
+
118
+ - 配置 `requireTypedPromotion`、`requireLevelReasons`(写了在组装时报 `unknown_config`);缺口 `no_language`、`unstructured_facts`、`levels_skipped`、`orphan_terms`。
119
+ - 整套跳级机制:`ExplainLevelSkip`、运行态卡的「未走过」、面板的等级导引。旧账本里的 `level/skipped` 折叠时静默跳过。
120
+ - `CheckPlan`(运行态卡每回合都在)与 `QueryKnowledge`(两轮六场模拟里一次没被调用;词表在 `clear/knowledge/`)。
121
+
122
+ **「少即是多」第三阶段:目标层挂到原生 goal,交付与裁决分开。** 按改造方案第三阶段的改动清单。
123
+
124
+ ### Changed
125
+
126
+ - **完成与结果分开。** 交付成立 ⇒ 这一步完成,不论结果是支持、推翻还是说不清;推翻和说不清都是合法结果,各记一条证据(带 `hypothesis`)。此前推翻的那一步收不了尾,只能作废。
127
+ - **评估者给两份判断**:交付成不成立(`holds`: yes / no / unclear),以及对每条被检验判断的结果(`results[]`)。交付不成立才不推进、计一次连拦。评估卡版本 `clearai.audit.v2`;旧形状的 `verdict` 仍读得懂。
128
+ - **一步可以检验多条判断**:`tests: { hypotheses: [...], level }`(旧的 `tests.hypothesis` 仍接受)。`AdvancePlan` 的参数是 `basis` 与 `results`(L0–L2);L3 以上由评估者给结果。
129
+ - **`SetGoal` → `Frame`、`CloseGoal` → `Conclude`。** `Frame` 在宿主原生 goal 上建(或改)一条目标,续跑由原生 goal 驱动;`Conclude` achieved ⇒ 原生 goal 完成并声明交付物,abandoned ⇒ 原生 goal 置阻塞。守卫拦住原生 `update_goal` 直接完成目标。
130
+ - **人门当场问。** L4 放行、计划连拦到阈值、推翻证据碰到已确立的事实,都由开门的那次调用经原生提问卡(`userQuestions`)当场问人;没人能答 ⇒ 原生 goal 置阻塞(`clearai-needs-human`)。面板上的人门动词只剩本体四个;旧日志里的撤回 / 维持照样折得出来。
131
+ - 挂上原生 `tool-goal`、`command-goal`、`plan-mode`。工具面 19 件,提示词 21 段(澄清协议只剩一段)。投影状态版本升到 13。
132
+
133
+ ### Removed
134
+
135
+ - ClearAI 自己的续跑窗口(`turnDemand`、布防 / 按住 / 收兵、`continuation/set`)与续跑额度(`maxAutoTurns`)。
136
+ - 运行档(`autonomy`)与两套澄清措辞。
137
+ - 计划审阅记号(`RequestPlanReview`、`plan/confirmed`、`confirmed_by`):它从来不是门;动手前给人看计划用原生 `/plan`。
138
+ - 收件箱里等人处置的条目与面板上的撤回 / 维持按钮;「连续两次无法判定就强制改」那条规则(说不清是合法结果)。
139
+
140
+ ---
141
+
142
+ **「少即是多」第二阶段:宿主已经有的,交还宿主。** 按改造方案第二阶段的清单删除;机制骨架(目标 / 计划 / 准入 / 证据 / 本体)不动。
143
+
144
+ ### Removed
145
+
146
+ - **世界线**(`ForkPlan` `AdvanceWorldline` `ConvergeFork` `WorldlineStatus` `AwaitWorldlines` `AbandonFork`、执行者子任务、横评仲裁、世界树里的车道)。并行探索改由原生 `subagent` 承担:做法迥异的路线就是竞争的假设,各由一个步骤检验。
147
+ - **旁路账本与文件恢复**(`FileHistory` `RestoreFile`、回合快照、交付点提交、`ledgerMaxFiles`)。下面「账本不再往用户的仓库里提交」那一节描述的旁路账本随之整体移除;每轮改了什么交给宿主的 `dsh-workspace-changes`。
148
+ - **侦察**(`SpawnScout` `MapScouts`、开工前侦察、资料面)、**外脑**(`brain.js`、`SaveSkill` `WriteMemory`、技能候选与扶正)、**模板技能与铺工作区**(`preset/template/`、空文件夹里的 `PROJECT.md` 与目录骨架)、**自己的 `/` 命令**(`commands.js`)。
149
+ - 界面:产物页签与 `/api/clearai/deliverables`、技能 · 记忆页签与 `/api/clearai/brain`、续跑状态行;人门里与上述机制相关的动作(`adopt_branch` `abandon_fork` `promote_skill` `confirm_provisional`)。
150
+ - 配置键:`templateDir` `scoutToolFilter` `gitWorldlines` `ledgerMaxFiles` `executorToolFilter` `precommitRecon` `mapScoutMax` `mapScoutConcurrency` `autoDispatchExecutors` `autoAdoptMinGap` `forkArbitration`。
151
+
152
+ ### Added
153
+
154
+ - **同一计划里两步不许声明同一个产物路径**(`CreatePlan` / `AmendPlan` 立约时就拒):子任务共用一个工作区,撞路径的两条路线会互相覆盖。
155
+
156
+ ### Changed
157
+
158
+ - 工具面 20 件(目标 2 · 计划 8 · 本体 10),提示词 21 段生效。
159
+ - 投影状态版本升到 12。旧会话照样读得开:被删机制的事件被安静跳过,不报错也不显示。
160
+ - 测试与长测剧本同步瘦身:世界线与侦察剧本换成「竞争路线」剧本;套件 17 份。
161
+
162
+ **账本不再往用户的仓库里提交。** 工作区本身是 git 仓库时,内核的账本(交付点、回合边界快照、恢复)与 git 世界线(分支、合并)此前**直接用那个仓库**:每个回合边界在用户**当前分支**上 `git add -A` 并以 `clearai <clearai@local>` 提交,世界线分支 `clearai/*` 也开在里面。于是用户没写完的改动、模型按技能约定写的 `lab/` `products/`、排查用的 `.tmp-*` 一起进了他的历史,下一次 push 就上了远端——本仓库自己就这样吃进过几十条「探索期快照」(独立验证员 2026-09 也记过同一件事,当时判为「不是源码缺陷」)。
163
+
164
+ > 过渡修复:后续计划把账本整体交给宿主(`dsh-workspace-changes`),届时本节描述的旁路账本会被移除。
165
+
166
+ ### Fixed
167
+
168
+ - **账本与世界线一律住在旁路账本**(`$DSH_HOME/storages/clearai/ledger/<slug>`):`gitContext` 删掉「工作区是 git 仓库就直接用它」那条分支。旁路账本认工作树里的 `.gitignore`、自动跳过用户的 `.git/`,所以账本看见的内容与用户看见的一致;`FileHistory` / `RestoreFile` / 世界线的行为不变,只是记录的位置换了。对用户仓库做的**唯一**一件事:在它本地的 `.git/info/exclude` 里加一行 `clear/worldlines/`,让嵌套的工作副本不出现在 `git status` 里(不进历史、不外传)。
169
+ - **分叉前先记一笔工作区快照**:账本的 HEAD 只在回合边界与交付点前进,不记的话世界线从上一笔快照分出去,看不见这之后写下的文件。采纳时主线上有挡路的未跟踪文件,也按「先存快照再合并」处理(此前只认「本地修改」那一种拒绝)。
170
+ - **开场的残留读数两处都数**:旁路账本里的 `clearai/*` 分支,以及**升级前**留在用户仓库里的那些。
171
+
172
+ ### Changed
173
+
174
+ - 旁路账本的文件数上限(`ledgerMaxFiles`,缺省 20000)从此对 git 工作区也生效:超过上限的大仓库里账本与 git 世界线退化成声明目录,并如实说明原因(此前 git 工作区不受这条限制,因为它写的是用户自己的仓库)。
175
+ - 仓库清理:删除误入版本库的 ClearAI 工作区残留(`lab/`、`products/`、`.tmp-fontdiag/`);案例证据搬到 `docs/cases/runs/`,营销卡片脚本搬到 `tools/marketing/`;`.gitignore` 挡住工作区约定的目录;新增 `tools/check-workspace-residue.mjs` 并接进 CI(不许跟踪工作区产物与根上的 `PROJECT.md`,PR 里不许有 `clearai@local` 的提交)。
176
+
5
177
  ## [0.3.1] — 2026-09-29
6
178
 
7
179
  **两个死结:计划置 blocked 后再也解不开,续跑窗口的阻塞码收不了兵。** 两条都不是措辞问题,是机制自己在文档承诺的出口上焊死了——0.3.0 的「连拦达阈值 ⇒ 置 blocked、停下等人」写得没错,可人按卡上说的三条出路走,一条也走不出去。
package/README.md CHANGED
@@ -16,14 +16,14 @@
16
16
  ClearAI is an **ontology discovery and exploration platform**, built on two core concepts:
17
17
 
18
18
  - **Domain ontology** (what you get) — your project's own vocabulary, the knowledge entries established through the loop, and their graphs. At the end of a research session you hold a continuously growing knowledge structure, retrievable next round by concept.
19
- - **Epistemic loop** (how you get it) — a disciplined seven-stage path: frame, hypothesize, plan, observe, verify, evaluate, record. Every edge is tested by evidence and independent evaluation.
19
+ - **Epistemic loop** (how you get it) — question → judgement (state what would prove it wrong) → a test that could fail → evidence → bounded conclusion → grows into the ontology. Every edge is tested by evidence and independent evaluation.
20
20
 
21
21
  > Other knowledge graphs pile up edges by extraction and assertion; here every edge has to be earned through the loop.
22
22
 
23
23
  ```bash
24
24
  # Install (npm package, prebuilt — no build step, no allowBuilds prompt)
25
- dsh plugin --profile web add clearai-dsh@0.3.1
26
- # or in the app: Plugins → Add plugin → clearai-dsh@0.3.1
25
+ dsh plugin --profile web add clearai-dsh@0.5.0
26
+ # or in the app: Plugins → Add plugin → clearai-dsh@0.5.0
27
27
  ```
28
28
 
29
29
  Restart `dsh web`, then pick **ClearAI** in the preset picker at the top of a new session. That is the whole setup. [Full install notes ↓](#install-and-use)
@@ -34,7 +34,7 @@ Restart `dsh web`, then pick **ClearAI** in the preset picker at the top of a ne
34
34
  <img src="docs/diagrams/ontology-hero.png" alt="The epistemic loop (left) growing a domain ontology (right)" width="1200">
35
35
  </picture>
36
36
 
37
- *Left: the Epistemic Loop — seven stages. Its emerald fact dot is also the first node of the domain ontology on the right. Right: the ontology graph — dark is a concept, light is a value form, emerald an instance; the instance carries two contradictory assertions — **the two readings are tinted amber**, marking that they do not agree. The system reports the conflict; retracting or keeping is a human decision.*
37
+ *Left: the Epistemic Loop. Its emerald fact dot is also the first node of the domain ontology on the right. Right: the ontology graph — dark is a concept, light is a value form, emerald an instance; the instance carries two contradictory assertions — **the two readings are tinted amber**, marking that they do not agree. The system reports the conflict; retracting or keeping is a human decision.*
38
38
 
39
39
  ---
40
40
 
@@ -66,7 +66,7 @@ Most agent loops track one thing: whether the task is done. The Epistemic Loop a
66
66
 
67
67
  *Inside the ring is the instrument's read-out: the L0–L4 axis, the **pre-registered** threshold as a dashed line, and five observations with error bars — the supported one filled, the inconclusive drawn as a dashed circle, the refuted left in place with a slash through it (nothing is deleted). The emerald dot at the opening is the one reading that crossed the threshold and settled as a fact.*
68
68
 
69
- At runtime, the seven stages compress into four beats — plan, execute, observe, reflect. State is derived from the session record with no second store; the tools the model holds contain no field in which it could declare a step complete.
69
+ State is derived from the session record with no second store; the tools the model holds contain no field in which it could declare a step complete, and a goal completes only after independent evaluation. ClearAI does only what the host cannot — the epistemic contract, the domain ontology, presentation; goal continuation, subagents, asking you, deliverable cards and file history all come from DSH itself.
70
70
 
71
71
  ClearAI does **not** claim recursive self-improvement. It provides the epistemic substrate a self-improving system would need. See [Positioning](docs/positioning.md) and the [OpenRSI survey](docs/research-openrsi.md).
72
72
 
@@ -74,16 +74,16 @@ ClearAI does **not** claim recursive self-improvement. It provides the epistemic
74
74
 
75
75
  ## Install and use
76
76
 
77
- **Requirements:** DSH ≥ `0.1.7-alpha.1` — that generation introduced the composition declaration line this preset rides on. Verified against the host's `0.1.7-rc.2` and `0.2.0-rc.1`.
77
+ **Requirements:** DSH ≥ `0.2.0-rc.2` (the current `latest`). The preset is registered through a composition declaration line, which is the only way the host's preset registry finds presets. Verified against the host's `0.2.0-rc.2`, and also checked against `0.2.1-alpha.1`. Hosts older than `0.2.0` are no longer supported.
78
78
 
79
79
  **Recommended — install it in the app, with the version pinned:**
80
80
 
81
- In the sidebar open **Plugins → Add plugin**, enter `clearai-dsh@0.3.1`, and install. That is DSH's own plugin manager: it hands what you type to pnpm, checks that the package declares a bundle and is compatible with this host, and applies it live. (The Settings page **插件列表 / Plugins** is the read-only inventory — installing happens on the sidebar's Plugins page.)
81
+ In the sidebar open **Plugins → Add plugin**, enter `clearai-dsh@0.5.0`, and install. That is DSH's own plugin manager: it hands what you type to pnpm, checks that the package declares a bundle and is compatible with this host, and applies it live. (The Settings page **插件列表 / Plugins** is the read-only inventory — installing happens on the sidebar's Plugins page.)
82
82
 
83
83
  **Or from a terminal — the same install:**
84
84
 
85
85
  ```bash
86
- dsh plugin --profile web add clearai-dsh@0.3.1
86
+ dsh plugin --profile web add clearai-dsh@0.5.0
87
87
  ```
88
88
 
89
89
  This installs the prebuilt package from the npm registry. Nothing is compiled on your machine, so there is no `allowBuilds` grant to approve — the plugin is ready the moment the command returns.
@@ -116,27 +116,27 @@ dsh plugin --profile web add github:Clearailhc/clearai-dsh
116
116
 
117
117
  Git fetches source rather than build artifacts, so pnpm ≥10 will refuse to run the `prepare` script until you add an `allowBuilds` entry to the profile's `pnpm-workspace.yaml`. That grant means *permission for this package's code to execute on your machine at install time* — grant it only if you have read the source, and pin a commit. If you just want to use ClearAI, use the npm install above.
118
118
 
119
- The installer's output follows your system language (`--lang zh|en` overrides it, `doctor` / `seed` / `unseed` take the same flag). Its only runtime dependency is `zod`; the graph stack is bundled into the client half at build time.
119
+ The installer's output follows your system language (`--lang zh|en` overrides it, and `doctor` takes the same flag). Its only runtime dependency is `zod`; the graph stack is bundled into the client half at build time.
120
120
 
121
121
  Restart `dsh web` afterwards (`npx @deepseek-ai/dsh web`), then **create a session and switch to the `ClearAI` mode in the picker at the top**:
122
122
 
123
123
  1. Open `dsh web` and click "New session";
124
124
  2. Click the current mode name at the top (default: **Standard mode**) to open the preset list;
125
125
  3. Pick **ClearAI** — its card reads "利用认识论循环构建可信本体。Build a trustworthy ontology through the epistemic loop.";
126
- 4. Just ask your question. Ordinary Q&A runs as usual; once you set a goal and register hypotheses, the system enters knowledge mode by itself: known facts come to you, gaps stay visible, and conclusions earn their place.
126
+ 4. Just ask your question. Ordinary Q&A runs as usual; once a goal is set and judgements are registered, the system enters knowledge mode by itself: what is already known comes to you, and conclusions earn their place through evidence. Only decisions only you can make are put to you.
127
127
 
128
128
  <picture>
129
- <img src="docs/shots/zh/jepa-ontology.png" alt="The knowledge graph in ClearAI mode" width="820">
129
+ <img src="docs/shots/en/jepa-ontology.png" alt="The Ontology pane in ClearAI mode" width="820">
130
130
  </picture>
131
131
 
132
- *The ontology graph in ClearAI mode — this real session grew 21 concepts and 9 predicates; the same ledger always yields the same picture. (UI shown is Chinese.)*
132
+ *The Ontology pane after the [JEPA world model](docs/cases/jepa-world-model.md) session: the question, the progress rail, the graph, and the conclusions grouped by status.*
133
133
 
134
134
  If pnpm is not on PATH: `npm install -g pnpm` (do not `corepack enable` — it installs a version forwarder that may download a pnpm it cannot launch).
135
135
 
136
136
  From the repository:
137
137
 
138
138
  ```bash
139
- npm test # 18 suites
139
+ npm test # 17 suites
140
140
  node tools/build-package.mjs # assemble dist/ from source
141
141
  node tools/verify-package.mjs # rebuild on the spot, byte-compare
142
142
  node docs/diagrams/build-hero.mjs # redraw the product hero (needs google-chrome)
@@ -148,35 +148,36 @@ node docs/diagrams/build-hero.mjs # redraw the product hero (needs google-chro
148
148
 
149
149
  ## What it looks like
150
150
 
151
- The middle column has two switchable views: **Deliverables** and **Ontology**. The right sidebar: **Worldlines** and **External Brain**.
151
+ One pane in the middle: **Ontology**. One pane on the right: **World Tree**. Next to the input box: "to handle N".
152
152
 
153
- **Ontology** — this view is your knowledge home. At the top, a **graph band**: the ontology graph (what your domain looks like) and the entity graph (what you have actually verified) toggle with one click; clicking a node or edge opens the **knowledge inspector** (definition / relations / assertions / evidence chain / history), and "filter by this" is an explicit action inside the detail view. Below that, the **ontology shelf**: established entries, each with assertion chips (click to see what the term means), boundary, and support level; contradictions surface automatically. The vocabulary maintenance block sits collapsed at the bottom — it auto-expands when a language exists before any sentence does.
153
+ **Ontology** — one page for your four questions. At the top: the question, one line of counts, anything you need to handle ("to handle"), and a small progress rail: judgment → test → verified → in ontology. In the middle, the **graph**: the ontology graph (what your domain looks like) and the entity graph (the concrete things found) toggle with one click, and clicking a node filters by it. Below, the **conclusion list**, one line each, grouped by status: verified, awaiting check, testing, uncertain, refuted, replaced; open one to see which station it reached, how its trust changed, and its basis and scope. Two contradictory conclusions light up; retracting or keeping is your call.
154
154
 
155
- <picture>
156
- <img src="docs/shots/zh/jepa-band.png" alt="The graph band: ontology graph and entity graph" width="820">
157
- </picture>
155
+ **World Tree** — the plan's steps and gates, one line per step; open one to see which judgments it tested and what came out.
158
156
 
159
- *The band — the ontology graph and the entity graph share one deterministic projection, so the same ledger always yields the same picture (captured from a real session: 21 concepts, 9 predicates).*
157
+ **Deliverables** — at close, the artefacts accepted for each step appear as DSH's native deliverable cards; what changed each turn is in DSH's native change cards.
158
+
159
+ **Language** — everything the system writes (tool results, the runtime card, questions to you, the files under `clear/`) follows the language you write in, Chinese or English. The panel follows the UI language.
160
160
 
161
161
  <picture>
162
- <img src="docs/shots/zh/jepa-inspector.png" alt="Knowledge inspector: definition, relations, assertions, evidence chain" width="820">
162
+ <img src="docs/shots/en/jepa-ontology-graph.png" alt="The ontology graph, full screen" width="820">
163
163
  </picture>
164
164
 
165
- *Open any node or edge: definition, relations, assertions, evidence chain, registration and revision history, all in one place. (UI shown is Chinese.)*
166
-
167
- **Worldlines** — when two routes genuinely disagree, they run as separate branches with their own readings; the losing one stays on record, and adoption is a human press.
165
+ *The ontology graph, full screen. The ontology graph and the entity graph come from one deterministic projection of the files under `clear/ontology/`, so the same files always give the same picture.*
168
166
 
169
- **Deliverables** — the middle column keeps "what the plan declared" and "what actually exists on disk" apart.
167
+ <picture>
168
+ <img src="docs/shots/en/ns-refuted.png" alt="A refuted judgment, expanded" width="820">
169
+ </picture>
170
170
 
171
- **External Brain** — skills and memory as DSH-native entries in one merged catalogue.
171
+ *A refuted judgment from the [Navier–Stokes](docs/cases/navier-stokes.md) session: it stopped at the test stage, and the record keeps why.*
172
172
 
173
173
  ---
174
174
 
175
175
  ## Cases
176
176
 
177
- - [Physical-world process experiment](docs/cases/physical-experiment.md) — sensor thermal drift: the full chain from raising terms to a conflict surfacing
178
- - [AI for Science](docs/cases/ai4sci.md) — convergence order of WENO reconstructions, and what "we could not resolve it" honestly means
179
- - [Mathematics](docs/cases/mathematics.md) — keeping finite numerical evidence strictly separate from proof
177
+ Both are real-model sessions run end to end without asking anything; the screenshots replay them in real DSH.
178
+
179
+ - [JEPA world models](docs/cases/jepa-world-model.md): a literature review, a toy experiment judged by an independent evaluator, and an ontology of 21 concepts, 8 relations and 27 entities
180
+ - [Was Navier–Stokes solved?](docs/cases/navier-stokes.md): two popular claims refuted and kept, an intake rejection handled honestly, and an ontology of 24 concepts, 13 relations and 59 entities
180
181
 
181
182
  ---
182
183
 
@@ -185,8 +186,7 @@ The middle column has two switchable views: **Deliverables** and **Ontology**. T
185
186
  - [Positioning](docs/positioning.md) · [Domain ontology design](docs/domain-ontology.md)
186
187
  - [Epistemic loop](docs/epistemic-loop.md) · [Verification ontology](docs/verification-loop.md) · [Loop philosophy](docs/loop-philosophy.md)
187
188
  - [Design principles](docs/design-principles.md) · [Soul map](docs/soul-map.md) · [Glossary](docs/glossary.md)
188
- - [Knowledge-native loop ledger](docs/optimization/knowledge-native-loop.zh-CN.md) (this round: triage / preflight / knowledge gate / graph / inspector; zh-CN)
189
- - [Development plan](docs/optimization/domain-ontology-plan.md) (with real-run evidence from the Hengtong project)
189
+ - [Mechanism truth table](docs/optimization/truth-table.md) · [State machines](docs/optimization/state-machines.md) · [Timing diagrams](docs/optimization/timing-diagrams.md)
190
190
  - [Known gaps](docs/known-gaps.md) · [Authority map](docs/authority-map.md) · [Release verification](docs/release-verification.md)
191
191
 
192
192
  ## Where it sits in DSH
package/README.zh-CN.md CHANGED
@@ -16,14 +16,14 @@
16
16
  ClearAI 是一个**本体发现与探索平台**,核心由两个概念支撑:
17
17
 
18
18
  - **领域本体**(你得到什么)——项目自己的词汇、经循环确立的知识条目、以及它们的图。研究结束时你拿到一个持续生长的知识结构,下一轮按概念检索。
19
- - **认识论循环**(你怎么得到它)——七个阶段的纪律化路径:界定、假设、规划、观测、验证、评估、记录。每条边都要经过证据与独立评估的检验。
19
+ - **认识论循环**(你怎么得到它)——问题 → 判断(写明怎样算错)→ 一次可能失败的检验 → 证据 → 带范围的结论 → 长进本体。每条边都要经过证据与独立评估的检验。
20
20
 
21
21
  > 别的知识图谱靠抽取与断言堆边;这里的每一条边都要通过循环挣得。
22
22
 
23
23
  ```bash
24
24
  # 安装(npm 包,预构建——无需构建步骤,不会触发 allowBuilds 授权)
25
- dsh plugin --profile web add clearai-dsh@0.3.1
26
- # 或在应用里:侧栏「插件」→ 添加插件 → clearai-dsh@0.3.1
25
+ dsh plugin --profile web add clearai-dsh@0.5.0
26
+ # 或在应用里:侧栏「插件」→ 添加插件 → clearai-dsh@0.5.0
27
27
  ```
28
28
 
29
29
  重启 `dsh web`,在新建会话顶部的模式选择器里选 **ClearAI** 即可。这就是全部步骤。[完整安装说明 ↓](#安装与使用)
@@ -34,7 +34,7 @@ dsh plugin --profile web add clearai-dsh@0.3.1
34
34
  <img src="docs/diagrams/ontology-hero.zh-CN.png" alt="认识论循环(左)长出领域本体(右)" width="1200">
35
35
  </picture>
36
36
 
37
- *左:认识论循环——七阶段。绿点是它落定的事实,也是右侧领域本体的第一个节点。右:本体图——深墨是概念,浅墨是值形态,emerald 是实例;实例上挂着两条互相矛盾的断言——**那两个取值染成 amber**,就是「这两条读数对不上」。系统只报出冲突,撤回或维持由人决定。*
37
+ *左:认识论循环。绿点是它落定的事实,也是右侧领域本体的第一个节点。右:本体图——深墨是概念,浅墨是值形态,emerald 是实例;实例上挂着两条互相矛盾的断言——**那两个取值染成 amber**,就是「这两条读数对不上」。系统只报出冲突,撤回或维持由人决定。*
38
38
 
39
39
  ---
40
40
 
@@ -66,7 +66,7 @@ dsh plugin --profile web add clearai-dsh@0.3.1
66
66
 
67
67
  *环内是这台仪器的读数面:横轴是 L0–L4 七个等级,那条虚线是**事先登记进判据**的阈值;五个观测各带误差棒——被支持的填实、无法判定的画虚圈、被推翻的留在原位打一道斜杠(什么都不删)。开口处那颗 emerald 是唯一越过判据、落定成事实的读数。*
68
68
 
69
- 运行时,七个阶段压缩为四拍——计划、执行、观察、反思。状态从会话记录派生,没有第二份存储;模型的工具里没有可以宣告某一步完成的字段。
69
+ 状态从会话记录派生,没有第二份存储;模型的工具里没有可以宣告某一步完成的字段,目标也只能在独立评估通过后完成。ClearAI 只做宿主做不了的那部分——认识论契约、领域本体、呈现;目标续跑、子代理、问人、交付卡片、文件历史都用 DSH 原生的。
70
70
 
71
71
  ClearAI **不**声称递归自我改进。它提供的是自我改进系统所需要的认识论底座。详见[定位](docs/positioning.zh-CN.md)与[OpenRSI 调研](docs/research-openrsi.md)。
72
72
 
@@ -74,16 +74,16 @@ ClearAI **不**声称递归自我改进。它提供的是自我改进系统所
74
74
 
75
75
  ## 安装与使用
76
76
 
77
- **要求:** DSH ≥ `0.1.7-alpha.1`——预设靠的是那一代引入的组合声明行。已在宿主的 `0.1.7-rc.2` 与 `0.2.0-rc.1` 上验过。
77
+ **要求:** DSH ≥ `0.2.0-rc.2`(当前的 `latest`)。预设靠组合里的声明行注册,宿主的预设名册只认这一种。已在宿主的 `0.2.0-rc.2` 上验过,`0.2.1-alpha.1` 也核过。`0.2.0` 以前的宿主不再支持。
78
78
 
79
79
  **推荐——在应用里装,并把版本钉住:**
80
80
 
81
- 侧栏打开**「插件」→ 添加插件**,填 `clearai-dsh@0.3.1`,安装。这就是 DSH 自己的插件管理器:它把你填的东西交给 pnpm,校验这个包声明了组合包、与当前宿主兼容,然后当场生效。(设置里的**插件列表**是**只读清单**;安装入口在侧栏那个「插件」页。)
81
+ 侧栏打开**「插件」→ 添加插件**,填 `clearai-dsh@0.5.0`,安装。这就是 DSH 自己的插件管理器:它把你填的东西交给 pnpm,校验这个包声明了组合包、与当前宿主兼容,然后当场生效。(设置里的**插件列表**是**只读清单**;安装入口在侧栏那个「插件」页。)
82
82
 
83
83
  **或者开终端——同一次安装:**
84
84
 
85
85
  ```bash
86
- dsh plugin --profile web add clearai-dsh@0.3.1
86
+ dsh plugin --profile web add clearai-dsh@0.5.0
87
87
  ```
88
88
 
89
89
  从 npm registry 装预构建产物。本机不跑任何编译,因此不需要批准 `allowBuilds` 授权——命令返回时插件就已经可用。
@@ -116,7 +116,7 @@ dsh plugin --profile web add github:Clearailhc/clearai-dsh
116
116
 
117
117
  Git 拉的是源码而不是构建产物,所以 pnpm ≥10 会拒绝运行 `prepare` 脚本,直到你在该 profile 的 `pnpm-workspace.yaml` 里加上 `allowBuilds` 条目。那条授权的含义是**允许该包代码在安装时于你机器上执行**——只在你读过源码后再授权,并且固定 commit。如果你只是想用 ClearAI,请用上面的 npm 安装。
118
118
 
119
- 安装侧的输出**跟系统语言走**(`--lang zh|en` 可覆盖;`doctor` / `seed` / `unseed` 同样认这个开关)。运行时依赖只有 `zod`——图谱那套栈在构建期就打进客户端了。
119
+ 安装侧的输出**跟系统语言走**(`--lang zh|en` 可覆盖;`doctor` 同样认这个开关)。运行时依赖只有 `zod`——图谱那套栈在构建期就打进客户端了。
120
120
 
121
121
 
122
122
  装完重启 `dsh web`(`npx @deepseek-ai/dsh web`),然后**新建会话,在顶部的模式选择器里切换到 `ClearAI`**:
@@ -124,20 +124,20 @@ Git 拉的是源码而不是构建产物,所以 pnpm ≥10 会拒绝运行 `pr
124
124
  1. 打开 `dsh web`,点「新建会话」;
125
125
  2. 点顶部当前的模式名(默认是「标准模式」),展开预设列表;
126
126
  3. 选 **ClearAI**——卡片描述就一句:「利用认识论循环构建可信本体。Build a trustworthy ontology through the epistemic loop.」;
127
- 4. 像平常一样提问。普通问答照常走;一旦你立了目标并登记假设,系统自动进入知识模式:预检把已知送上来,缺口如实读出来,结论过证据才准入。
127
+ 4. 像平常一样提问。普通问答照常走;一旦立了目标、登记了判断,系统自动进入知识模式:已知送上来,结论过证据才准入。只有你能做的决定才会来问你。
128
128
 
129
129
  <picture>
130
- <img src="docs/shots/zh/jepa-ontology.png" alt="ClearAI 模式下的知识图谱" width="820">
130
+ <img src="docs/shots/zh/jepa-ontology.png" alt="ClearAI 模式下的「本体」一格" width="820">
131
131
  </picture>
132
132
 
133
- *ClearAI 模式下的本体图——这场真实会话长出 21 个概念、9 条谓词;同一份账本永远得到同一张图。*
133
+ *[JEPA 世界模型](docs/cases/jepa-world-model.zh-CN.md)那场会话之后的「本体」一格:问题、进度轨、图,以及按状态分组的结论。*
134
134
 
135
135
  如果 PATH 上没有 pnpm:`npm install -g pnpm`(别用 `corepack enable`——它装的是版本转发器,可能下载一个自己启动不了的 pnpm)。
136
136
 
137
137
  从仓库开发:
138
138
 
139
139
  ```bash
140
- npm test # 18 份套件
140
+ npm test # 17 份套件
141
141
  node tools/build-package.mjs # 由源装配 dist/
142
142
  node tools/verify-package.mjs # 现场重建并逐字节比对
143
143
  node docs/diagrams/build-hero.mjs # 重画产品主图(需 google-chrome)
@@ -149,35 +149,36 @@ node docs/diagrams/build-hero.mjs # 重画产品主图(需 google-chrome)
149
149
 
150
150
  ## 它长什么样
151
151
 
152
- 中栏两格可切:**产物**与**本体**。右栏:**世界树**与**外脑**。
152
+ 中栏一格:**本体**。右栏一格:**世界树**。输入框旁边是「待处理 N」。
153
153
 
154
- **本体**——这一格是你的知识主场。顶部是**图带**:本体图(你的领域长什么样)与实体图(已经验证出了什么)一键切换;点节点或边打开**知识 Inspector**(定义 / 关系 / 断言 / 证据链 / 历史),「按此过滤」是详情里的显式动作。下面是**本体货架**:已确立的条目,每条带断言芯片(点开看这个词什么意思)、边界与等级;互相矛盾的自动亮出来。词汇维护区收在最底下——语言先于句子时它自动展开。
154
+ **本体**——一页回答你的四个问题。最上面一行问题、一行计数、要你处理的事(「待处理」),和一条小进度轨:判断 → 检验 → 已验证 → 入本体。中间是**图**:本体图(你的领域长什么样)与实体图(找到了哪些具体东西)一键切换,点节点即按它过滤。下面是**结论清单**,一条一行,按状态分组:已验证、待核验、验证中、不确定、已推翻、已替换;点开一条看它走到了哪一站、可信度怎么变过来的、依据与范围。互相矛盾的两条会亮出来,撤回或维持由你决定。
155
155
 
156
- <picture>
157
- <img src="docs/shots/zh/jepa-band.png" alt="本体图带:本体图与实体图切换" width="820">
158
- </picture>
156
+ **世界树**——计划的步骤与门,每步一行,点开看它检验了哪条判断、结果是什么。
159
157
 
160
- *图带——本体图与实体图共用同一份确定性投影,同一份账本永远得到同一张图;素材来自一场真实会话(21 概念 · 9 谓词)。*
158
+ **交付**——结案时,各步收下的产物以 DSH 原生交付卡片出现;每轮改了哪些文件,看 DSH 原生的改动卡片。
159
+
160
+ **语言**——系统自己写的话(工具结果、运行态卡、问你的话、`clear/` 下的文件)都跟着你说话的语言走,中文或英文。面板跟着界面语言。
161
161
 
162
162
  <picture>
163
- <img src="docs/shots/zh/jepa-inspector.png" alt="知识 Inspector:定义、关系、断言、证据链" width="820">
163
+ <img src="docs/shots/zh/jepa-ontology-graph.png" alt="本体图全屏" width="820">
164
164
  </picture>
165
165
 
166
- *点开任一节点或边:定义、关系、断言、证据链、登记与修订史,一处读完。*
167
-
168
- **世界树**——两条路线真的分歧时,各自独立跑、各自带读数;落选的那条留在记录里,采纳是人按的那一下。
166
+ *本体图全屏。本体图与实体图来自 `clear/ontology/` 下文件的同一份确定性投影,同样的文件永远得到同一张图。*
169
167
 
170
- **产物**——中栏把「计划声明交付的」与「盘上真有的」分开摆,不许混为一谈。
168
+ <picture>
169
+ <img src="docs/shots/zh/ns-refuted.png" alt="展开一条被推翻的判断" width="820">
170
+ </picture>
171
171
 
172
- **外脑**——技能与记忆以 DSH 原生条目的形式出现在同一张合并目录里。
172
+ *[Navier–Stokes](docs/cases/navier-stokes.zh-CN.md) 那场会话里一条被推翻的判断:它停在检验这一站,记录留着原因。*
173
173
 
174
174
  ---
175
175
 
176
176
  ## 案例
177
177
 
178
- - [物理世界工艺实验](docs/cases/physical-experiment.zh-CN.md)——传感器热漂移:从立词到冲突现形的完整链路
179
- - [AI for Science](docs/cases/ai4sci.zh-CN.md)——WENO 重构的收敛阶,以及「分辨不出来」意味着什么
180
- - [数学探索](docs/cases/mathematics.zh-CN.md)——把有限数值证据与形式证明严格分开
178
+ 两个都是真模型从头跑到尾、中途没问人的会话;截图是在真 DSH 里回放它们拍的。
179
+
180
+ - [JEPA 世界模型](docs/cases/jepa-world-model.zh-CN.md):文献综述、交给独立评估者判的玩具实验,以及 21 个概念、8 种关系、27 个实体的本体
181
+ - [Navier–Stokes 被解决了吗?](docs/cases/navier-stokes.zh-CN.md):两条流行说法被推翻并保留,一次准入被拦后如实处理,以及 24 个概念、13 种关系、59 个实体的本体
181
182
 
182
183
  ---
183
184
 
@@ -186,8 +187,7 @@ node docs/diagrams/build-hero.mjs # 重画产品主图(需 google-chrome)
186
187
  - [定位](docs/positioning.zh-CN.md) · [领域本体设计](docs/domain-ontology.zh-CN.md)
187
188
  - [认识论循环](docs/epistemic-loop.zh-CN.md) · [验证本体](docs/verification-loop.zh-CN.md) · [循环哲学](docs/loop-philosophy.zh-CN.md)
188
189
  - [设计原则](docs/design-principles.zh-CN.md) · [灵魂映射](docs/soul-map.zh-CN.md) · [术语表](docs/glossary.zh-CN.md)
189
- - [知识原生循环执行账本](docs/optimization/knowledge-native-loop.zh-CN.md)(本轮:分诊 / 预检 / 知识门 / 图 / Inspector)
190
- - [开发计划](docs/optimization/domain-ontology-plan.zh-CN.md)(含亨通真跑读数)
190
+ - [机制真值表](docs/optimization/truth-table.zh-CN.md) · [状态机](docs/optimization/state-machines.zh-CN.md) · [时序图](docs/optimization/timing-diagrams.zh-CN.md)
191
191
  - [已知缺口](docs/known-gaps.zh-CN.md) · [权威归属](docs/authority-map.zh-CN.md) · [发布验收](docs/release-verification.zh-CN.md)
192
192
 
193
193
  ## 它落在 DSH 的哪一层