Jev(TypeSafe System One Model)精读与机制映射(200-201):闭合输出空间 × 校准领地 × 快慢分工 - #1171
Merged
Merged
Conversation
- 190 精读笔记:「快递分拣中心」单射比喻贯穿 M1-M4(闭合输出空间合法≠正确 / 一次编码·分支隔离·选项读出 / RLCD 校准与 confidence 固定公式 / 快慢分工编排);多源对账(官方文档与评测 + adapter 钉点 e1d4cc9 + Hume 逆向探针 + kev b8aa777 / laya 76361c8 开源复刻 + 三组第三方基准 + 6 篇公众号)逐级标注【一/二/三/四】证据分级;五条底层规律 + 三个争议 + 双 Agent 费曼对抗实录(含同构变式复考全绿)+ 厂商数字跨源口径冲突表 - 191 机制映射:16 条(✅4/🔶6/⏸6),最大真增量=「闭合输出空间」纪律而非接入模型(自动作答不校验选项成员 / Judge·PlanReviewer 解析失败静默降 0 分 / global_search「无相关信息」哨兵漏入 reduce);直接接入 Jev 暂缓于 CJK·闭源·模型类型三门槛 - 最小原型 assets/jev_lab.py(纯标准库 399 行,--selftest + B1-B6 六次破坏性实验,实测日志回灌笔记) - archify 两相管线图入四件套:jev--one-pass-decision(dataflow)/ jev--fast-slow-harness(workflow),validate 0/0 + visual-check 四视口通过 - 索引同步:knowledge-map / research readme §五 / _category_ / mermaid README 登记;CHANGELOG Added;取证副产物 7 处漂移登记 ISSUE-196(3 个未注册 task key + auto_answer 死配置等);顺手修复 CHANGELOG 一处指向已外置 to-video 仓的失效相对链接 🤖 Generated with [Claude Code](https://github.com/claude), [CodeX](https://openai.com), [Gemini](https://github.com/apps/gemini-code-assist) Co-Authored-By: Aurelius Huang<threefish.ai@gmail.com>
…190-191、ISSUE-198 让位 196/197); - merge origin/feature/1.x.x(#1170 Hermes Agent 精读已占用 agent-harness 190-191 与 ISSUE-196/197) - Jev 精读笔记/映射 190/191 → 200/201,全部交叉引用(知识索引、research readme、mermaid README、CHANGELOG、issue、.mmd 溯源头)同步改号;本仓编号纪律为「新主题开新十位段」,跨目录同号双主题易混淆故让位 - ISSUE-196 → ISSUE-198;CHANGELOG 双方 Added 条目并存 🤖 Generated with [Claude Code](https://github.com/claude), [CodeX](https://openai.com), [Gemini](https://github.com/apps/gemini-code-assist) Co-Authored-By: Aurelius Huang<threefish.ai@gmail.com>
- 201 #8「190 §5 的原型实测」→「200 §5」:被引数据(自称错误率 0.7%/实际 12%)出自 200 §5 的 S8 输出,190 §5 是 Hermes 学习闭环章; - ISSUE-198 两处「191 §D3/D4/D6」「191 §落地建议」→ 201:D 编号漂移表与落地建议汇总均在 Jev 映射(201),191 为 Hermes 映射; - 200 §3 原型实景块补齐引文:S1 补第三条 422、S4/B2 省略处补「...」标记、B2 拆后 5 行完整引用(含陷阱单越界行),使 §10 表「5 单 3 单越界」可从引文验证;全部非省略行经与实际运行输出逐字比对命中。 🤖 Generated with [Claude Code](https://github.com/claude), [CodeX](https://openai.com), [Gemini](https://github.com/apps/gemini-code-assist) Co-Authored-By: Aurelius Huang<threefish.ai@gmail.com>
ThreeFish-AI
added a commit
that referenced
this pull request
Sep 24, 2026
…,PR diff 只保留视频增量); # Conflicts: # docs/.agents/knowledge-map.md # docs/assets/mermaid/README.md
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
概要
以
/guided-learn导师式精读拆解 TypeSafe AI 的首个「System One Model」Jev(jev-1.13.0,闭源):不生成文字、只对预定义类型问题(noul是非 /choice≤255 选项 /score≤10 级)输出带校准概率的决策。全套沉淀入docs/research/agent-infra/。交付物
grep -n实测。最大真增量是**「闭合输出空间」纪律而非接入模型**:Claude Code 自动作答不校验答案属于选项、Judge/PlanReviewer 解析失败静默降 0 分、global_search「无相关信息」哨兵漏入 reduce;直接接入 Jev ⏸ 暂缓(官方自认 CJK 较弱 / 闭源 / 本仓模型类型枚举无位),触发条件=可自托管 + 中文校准经本仓数据验证(候选 kev/laya)。--selftest十场景 +--break B1..B6六次破坏性实验(拆隔离→暗号串线 / 拆闭合空间→越界标签 / 拆温度→自称错 0.7% 实际 12% / OOD 沿用旧温度→错 23.1% / 拆共享前缀→12.5×→1.0× / 互斥拆双 Noul→和 1.85 同时执行),实测日志回灌笔记。jev--one-pass-decision(推理相 dataflow)、jev--fast-slow-harness(编排相 workflow)。_category_/ mermaid README / CHANGELOG 同步;取证副产物 7 处漂移(3 个未注册 task key、auto_answer_model死配置等)登记 ISSUE-198。与 #1170 的编号协调
feature/1.x.x 已合入 Hermes 精读(agent-harness 190-191 + ISSUE-196/197),本 PR 将 Jev 精读从 190/191 重编号为 200-201、ISSUE-196 → ISSUE-198(编号纪律「新主题开新十位段」,跨目录同号双主题易混淆),全部交叉引用同步改号;另顺手修复 CHANGELOG 中一处指向已外置 to-video 仓的失效相对链接。
验证
uv run --no-project python docs/research/agent-infra/assets/jev_lab.py --selftest→SELFTEST PASSED ✔;B1–B6 各自产出可复现退化日志,笔记引用逐字比对一致;🤖 Generated with Claude Code