Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
36 changes: 36 additions & 0 deletions docs/architecture/rfcs/monorepo-distribution-split-v0.md
Original file line number Diff line number Diff line change
Expand Up @@ -4169,3 +4169,39 @@ File/SQLite negative cases guard this boundary without changing valid activation
Full task-artifact success through this entry, installed native App/model
adoption, compaction retention, complete D2/D5 closure, comparable total cost,
V1–V5/V10, R1/M3 and C1 default cutover remain unqualified.

### 2026-11-15 — First-Goal reply adopted by the original governed Task

- **Source:** `main@74ca883d5bd3`. The same source combines without conflict
with C1 #8266 at `2c099be41e68`; the isolated combined tree passes 287
native-session, work-context, first-read and Turn-settlement tests using
isolated state and synthetic provider/model fixtures. This does not approve
C1 or qualify its default cutover.
- **Observed product path:** the real Chat discovery/action/store creates one
Goal and its initial Task from a synthetic reviewed request. The current
Host model reads the retained request and input through a temporary stdio
relay, produces an eight-group read-only report, and the real packaged Goal
conversation renders it. Independent CSV/Decimal checks verify groups and
totals. A real Chat restart preserves the original Session, reply and action;
retrying that action dispatches no new Chat Turn.
- **Governed adoption:** the fixture caller declares an independent argv
validator through the existing Task owner, then the current model adopts
that original reply through the existing Core + SQLite + generic Host route.
It uses the same Goal, Agent and initial Task. SIGKILL after durable
`typed_result` resumes the original Turn without another Core Host invocation.
A damaged reply fails validation with the Task open and zero debit; restoring
the checked reply completes that Task with one debit, retained on replay.
- **Remaining entry gap:** the first Chat request is still a read-only planning
conversation with a separate explicit LoopX-mode control. Its observed model
input is 24,372 UTF-8 bytes; this is one observation, not a cost comparison.
The fixture's explicit validation declaration and governed adoption are not
automatic product behavior. Reuse the registered managed-session, Task
validation/admission and C1 owners for that missing integration; do not infer
write authority or weaken independent validation to activate it.
- **Boundary:** these source and transport checks strengthen first-result
readback and original-Task recovery evidence. They do not qualify one-goal
native onboarding, installed native App/model adoption, model compaction
retention, complete D2/D5 closure, comparable full costs, V1–V5/V10, R1/M3 or
C1 cutover. No runtime, schema, activation, permission or public evidence
policy changes; one-run relay/validator files remain private qualification
artifacts rather than a new shipped execution pipeline or CI fixture.
27 changes: 27 additions & 0 deletions docs/architecture/rfcs/monorepo-distribution-split-v0.zh-CN.md
Original file line number Diff line number Diff line change
Expand Up @@ -3239,3 +3239,30 @@ R1/M3、C1 切换和原目标回归仍未完成资格验证。
创建和恢复 owner。没有新增共享 schema、授权、替代执行管线或 C1 组合 owner。
通过该入口交付完整任务成果、已安装原生 App/模型采用、compaction 保留、完整
D2/D5 闭包、可比总成本、V1–V5/V10、R1/M3 和 C1 默认切换仍未完成资格验证。

### 2026-11-15 — 首个 Goal 的回复被原受治理 Task 采用

- **来源:**`main@74ca883d5bd3`。同一来源与 C1 #8266 的 `2c099be41e68`
无冲突组合,隔离组合树通过 287 项原生会话、work-context、首读和 Turn 结算
测试;这些测试使用隔离状态和合成 provider/model fixtures。这不构成 C1 的
批准或默认切换资格。
- **实际产品路径:**真实 Chat discovery/action/store 从合成的已审阅请求创建
一个 Goal 和初始 Task。当前 Host 模型通过临时 stdio relay 读取保留的请求和
输入,产出八组只读报告;实际打包的 Goal 对话页面显示该报告。独立 CSV/Decimal
检查核验每组及总计。真实 Chat 重启保留原 Session、回复及操作;重试原操作
不会再次派发 Chat Turn。
- **受治理采用:**测试调用方通过既有 Task owner 声明独立 argv 校验器,当前
模型再沿既有 Core + SQLite + generic Host 路径采用原回复。Goal、Agent 和
初始 Task 保持一致。在 durable `typed_result` 后注入 SIGKILL,同一个 Turn
不再次调用 Core Host 即可恢复。损坏回复被校验拒绝,Task 保持开放且零扣额;
恢复已校验回复后,原 Task 完成且只扣一次,重放仍保留该单次扣额。
- **剩余入口缺口:**首个 Chat 请求仍是只读规划对话,另有明确的 LoopX 模式
控件。本次观察到的模型输入为 24,372 个 UTF-8 字节;这是单次观察,不是成本
对照。测试中明确声明校验器及受治理采用的过程不是产品自动行为。缺失集成应
复用注册的 managed-session、Task 校验/准入及 C1 owner;不得推断写权限或
削弱独立校验来激活该路径。
- **边界:**这些来源与传输检查增强首个结果读回、原 Task 恢复的证据,不构成
一个目标即可启动的原生入口、已安装原生 App/模型采用、模型 compaction 保留、
完整 D2/D5 闭包、可比完整成本、V1–V5/V10、R1/M3 或 C1 切换的资格验证。
没有改变运行时、schema、激活、权限或公共证据政策;单次 relay/校验器文件
保留为私有资格验证产物,没有新增已交付执行管线或 CI fixture。
Loading