M6(2/n): 治理台九功能 + SSE 订阅层 + 读/治分权
治理台按 §9.1 拆成九个 gov-* feature,共用件落 shared/gov:
· 外壳(左轨/顶栏/面包屑/页签/整页下钻)自原型 Governance.tsx 平移,
DOM 与 g-* 类名原样;路由状态即应用状态,区/页签/下钻栈全编码进 hash 可深链
· 编排画布 diagram-js 画布件 1:1 平移(flow-layout/flow-node-paint/FlowDiagram),
新增 pkg-to-graph 把 ProcessPackage 转成画布图模型;统一报价 v1 全图 12 节点 15 边渲染通过
· 后端补四路由:审批单 / 工作流下钻 / 通道实例 / 度量仪表
SSE 订阅层(§9.2):app 层单一连接,按信封目标失效缓存;
组件不再自行轮询(去掉 refetchInterval)。广播只说「哪个目标动了」,不带业务载荷,
且在投递已落库已送达之后才发 —— 不是第二条投递链路。
两处实质修正:
· 治理路由一律要 govern → 读路由改要 read。客体三类 × 动作三种的分法就是为此:
BD 有 task:read 能看在途、动不了控制事件;一律 govern 等于把三种动作压成一种
· 控制事件/恢复/改派原记为 engine-transition(归机器产出)→ 新增 control 条目种类,
归人裁决侧。归侧是按动作类型的纯推导,所以「谁干的」必须体现在 kind 上;
否则人的叫停会被算进自主率的机器分子
判据 ⑩ 治理台一半跑通:在途监控 / 叫停 / 恢复,治理动作留痕可查。
无权查看不再画成空清单,按岗位说明缺哪一项。
gates: tsc / eslint / depcruise(286 模块)/ api-report / 360 测试 全绿
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KeDBx2PHLavbURXB8LJJ7m
This commit is contained in:
Claude Opus 5
@@ -0,0 +1,115 @@
|
||||
/**
|
||||
* feature `gov-memory` —— 记忆管理:度量仪表、活动记忆四层
|
||||
*
|
||||
* 自 `~/coding/JCAI_PRD/governance/_components/Governance.tsx` 平移:
|
||||
* DOM 结构、类名(`g-*` + 色调 `t-*`)与文案原样保留,mock 常量换成服务端投影(README 平移规则 1/2/4)。
|
||||
* 区首页只负责导航,治理动作一律落在下钻详情页的**就地确认带**(影响预览三必答)。
|
||||
*/
|
||||
import { useState } from 'react';
|
||||
import { Chip, Clear, Kpi, NList, NRow, Note, PageHead, QueryGate } from '@web-shared/gov/primitives.tsx';
|
||||
import type { Nav } from '@web-shared/gov/gov-route.ts';
|
||||
import {
|
||||
useGovActivities,
|
||||
useGovMemory,
|
||||
useGovMetrics,
|
||||
} from '@web-shared/gov/gov-api.ts';
|
||||
|
||||
export type HomeProps = { tab: string; nav: Nav };
|
||||
|
||||
export function MemoryHome({ tab, nav }: HomeProps) {
|
||||
const metrics = useGovMetrics();
|
||||
const activities = useGovActivities();
|
||||
const [picked, setPicked] = useState<string | undefined>(undefined);
|
||||
const memory = useGovMemory(picked);
|
||||
|
||||
if (tab === 'layers') {
|
||||
return (
|
||||
<>
|
||||
<Note>
|
||||
活动记忆四层:<b>事实层 / 模式层 / 裁决层 / 例外层</b>。写入唯蒸馏,条目必挂证据链;
|
||||
记忆不载权力——强制力只经候选增量 + 回放 + 人放行进入业务模型。
|
||||
</Note>
|
||||
<div className="g-picker">
|
||||
{(activities.data?.activities ?? []).map((a) => (
|
||||
<button key={a.activityId} className={'g-pick' + (picked === a.activityId ? ' on' : '')} onClick={() => setPicked(a.activityId)}>
|
||||
{a.name}
|
||||
</button>
|
||||
))}
|
||||
</div>
|
||||
<NList>
|
||||
{(memory.data?.entries ?? []).map((e) => (
|
||||
<NRow
|
||||
key={e.id}
|
||||
tone={e.incrementState === undefined ? 'brand' : 'warn'}
|
||||
ico="layers"
|
||||
title={e.statement}
|
||||
sub={`${e.layer} · ${e.createdAt.slice(0, 16).replace('T', ' ')}`}
|
||||
cols={[{ k: '证据', v: e.evidenceCount }]}
|
||||
right={e.incrementState !== undefined ? <Chip tone="warn">候选 · {e.incrementState}</Chip> : <Chip tone="ok">已成条目</Chip>}
|
||||
onClick={() => nav.push({ k: 'mem', id: e.id })}
|
||||
/>
|
||||
))}
|
||||
{picked !== undefined && (memory.data?.entries ?? []).length === 0 && (
|
||||
<Clear>该活动尚无记忆条目——终态封卷触发蒸馏钩子后产出。</Clear>
|
||||
)}
|
||||
</NList>
|
||||
</>
|
||||
);
|
||||
}
|
||||
|
||||
const m = metrics.data;
|
||||
const pct = (v: number | undefined) => Math.round((v ?? 0) * 100);
|
||||
return (
|
||||
<QueryGate query={metrics}>
|
||||
<>
|
||||
<div className="g-kpis">
|
||||
<Kpi ico="bot" tone="auto" label="自主率" value={pct(m?.autonomy)} unit="%" dial={pct(m?.autonomy)} sub="机器产出 ÷(机器 + 人)" />
|
||||
<Kpi ico="rotate-ccw" tone="warn" label="纠正率" value={pct(m?.correction)} unit="%" dial={pct(m?.correction)} sub="被纠正批次 ÷ 提交批次" />
|
||||
<Kpi
|
||||
ico="gauge"
|
||||
tone="brand"
|
||||
label="有效自主率"
|
||||
value={pct(m?.effectiveAutonomy)}
|
||||
unit="%"
|
||||
dial={pct(m?.effectiveAutonomy)}
|
||||
sub="自主率 ×(1 − 纠正率)"
|
||||
/>
|
||||
<Kpi ico="hand" label="打扰率" value={(m?.interruption ?? 0).toFixed(1)} sub="每单任务调用人类工具次数" />
|
||||
<Kpi ico="git-commit" tone="ok" label="DMN 覆盖率" value={pct(m?.dmnCoverage)} unit="%" sub="确定性求值占判断次数" />
|
||||
</div>
|
||||
<Note>
|
||||
自主率与纠正率<b>成对读</b>,复合为有效自主率——单看自主率会奖励「猛冲后返工」。度量只是仪表,不载权力。
|
||||
</Note>
|
||||
</>
|
||||
</QueryGate>
|
||||
);
|
||||
}
|
||||
|
||||
/* ---------------- ⑨ 系统管理 ---------------- */
|
||||
|
||||
export function MemPage({ id }: { id: string }) {
|
||||
const activities = useGovActivities();
|
||||
const first = (activities.data?.activities ?? [])[0]?.activityId;
|
||||
const memory = useGovMemory(first);
|
||||
const e = (memory.data?.entries ?? []).find((x) => x.id === id);
|
||||
if (e === undefined) return <Note>记忆条目需从对应活动进入。</Note>;
|
||||
return (
|
||||
<>
|
||||
<PageHead
|
||||
ico="layers"
|
||||
kicker={`活动记忆 · ${e.layer}`}
|
||||
title={e.statement}
|
||||
chips={
|
||||
<>
|
||||
<Chip ico="link">{e.evidenceCount} 条证据</Chip>
|
||||
{e.incrementState !== undefined && <Chip tone="warn">候选增量 · {e.incrementState}</Chip>}
|
||||
</>
|
||||
}
|
||||
desc="写入唯蒸馏,条目必挂证据链;记忆不载权力——强制力只经候选增量 + 回放 + 人放行进入业务模型。"
|
||||
/>
|
||||
</>
|
||||
);
|
||||
}
|
||||
|
||||
/* ---------------- 系统 ---------------- */
|
||||
|
||||
Reference in New Issue
Block a user