mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 20:01:57 +00:00
ThearchyHelios
1195c4abcb
复核 W4R-03（a×c）：召回在原答案的运行上追加 ledger.recalled、ledger.basis_changed，以前只在运行时间线（decode）里，问数据的
轮次底部、证据面板「口径」一步、账本接口都不带——使用者被标了「依据有变」、或者口径升版后「已按 v2 核对，不影响」，重开对话
什么也看不到，得点开「过程」才翻得出来，而使用者进不了记录页。W4-a 已在账本接口加上 recall（ledger/recall.answer_marks：
checked / basis / tasks，几种 state 都带），这边照它画：
- typesLedger：RunLedger.recall（LedgerRecallMarks）
- run/LedgerStep 新增 RecallMarks（问数据的轮次和「口径」一步共用）：「已按口径「X」vN 核对，不影响」（同一份口径核对过
几次只写最近那一版，数变了的不在这里写）；依据有变按两类各一行——数据目录（catalog / values / catalog_edit）和表结构
（structure），认不出的种类只说「答案的依据有变」不猜，项数相加，按新目录新出现错误级问题的写几条；看的人手上有说的是
这个答案的召回待办时写一句（口径升到 vN、数变了 / 依据有变检查出新问题）并链到 /apps/inbox/{待办}
- LedgerTurn：比对过的卡里写在摘要下面；没比对、不比对的答案（state 不是 matched）只要带着标记就单独写一张小卡
- LedgerPart（证据面板「口径」一步）：有标签的写在脚注前；没标签但答案带着标记的，这一步只写标记
复核 W4R-10（多包）：细案 02 做法 10 的 E2「说明有用 / 看不懂」以前没有入口。W4-a 已加 POST /ledger/answers/{run}/reconcile/
{对账运行}/feedback（{verdict: useful | unclear}，记成对账运行上的封存后事件、统计 E2 数它），账本接口的 reconciled 每一项
带 feedback {mine, useful, unclear}。这边：
- api.caliberLedger.feedback；typesLedger 加 LedgerFeedback、LedgerFeedbackOut
- run/LedgerReconcile 的对账卡：做完的卡问一句「「差在哪」的说明有帮助吗？」，两个按钮，看的人选过的按下（aria-pressed），
写各几个人（服务端给的数），可以改主意；交回来的反馈先画上，账本再取一次让别处挂着的同一张卡跟着变；回收站里只看不反馈；
没做完的卡不问
- 问数据的轮次里卡片上方没有「差在哪」，卡里补写一份（showWhere），反馈问的就是它；证据面板里它本来就在卡片上方，不重复
检查（写请求都在浏览器层拦下）：
- check-evidence 新一段 ledger-recall：比对过的答案「口径」一步写核对过（只写最近那一版）、依据有变（项数相加、新问题条数）、
链到待办；不比对的答案这一步只写依据有变
- check-chat 新一段 recall：轮次底部同上，另核对做完的对账卡写「差在哪」和反馈人数；不比对的答案单独一张小卡写表结构有变
- check-ledger 对账一段：没做完不问；点「看不懂」请求体只有 verdict、按下、写各几个人；改点「说明有用」再记一条、按下的换过来
- 共用夹具 ledger-fixtures 加 RECALL
把这几个文件的改动拿掉回跑：check-chat recall 6 项、check-evidence ledger-recall 4 项、check-ledger 对账反馈 2 项都不过；
放回去全过
134 lines
9.4 KiB
JavaScript
134 lines
9.4 KiB
JavaScript
// 口径账本的检查夹具（check-ledger、check-evidence、check-chat 共用）：GET /api/runs/{id}/ledger 的答复形状照服务端
|
||
// app/ledger/view.py 的 run_view（labels.view_label、issue.Plan.action、reconcile.actions_of / reconciled_view），
|
||
// 挂在可点击证据的夹具（frontend/src/run/__tests__/evidence-doc.json，compose_doc 真跑出来的周报，只用通用名）上：
|
||
// s6「45,678.5元」来自受管口径「周报口径」v2，等价已证明；s10「1,234单」来自已发布口径「订单口径」v1，助手选用；
|
||
// s8「8.7%」相关口径「订单口径」v1（差在哪：少了一个筛选条件），可以对账；s12「37.0元」这次的推理；
|
||
// s23「2.35%」没有比对（比对超时）。
|
||
// 口径名、表名、数都是虚构的通用名。
|
||
|
||
/** 口径的几份（calibers） */
|
||
export const WEEKLY = { caliber: 'wf-weekly@2', workflow_id: 'wf-weekly', version: 2, level: 'governed', level_text: '受管',
|
||
workflow_name: '周报口径', graph_hash: 'aa11' }
|
||
export const ORDERS = { caliber: 'wf-orders@1', workflow_id: 'wf-orders', version: 1, level: 'published', level_text: '已发布',
|
||
workflow_name: '订单口径', graph_hash: 'bb22' }
|
||
|
||
const caliberLabel = (c, how, extra = {}) => ({
|
||
kind: 'caliber', how, caliber: c.caliber, workflow_id: c.workflow_id, version: c.version, level: c.level,
|
||
level_text: c.level_text, workflow_name: c.workflow_name, ...extra,
|
||
})
|
||
|
||
export const DIFF = { items: [{ aspect: 'filter', a: null, b: "status = '[已遮罩]'", text: '口径只保留状态为已支付的行，这次没有这个条件' }],
|
||
fallback: false, text_diff: [] }
|
||
|
||
/** 一份比对过的账本。whole：整题判断（缺省不等价）；actions、issued、reconciled 可以换 */
|
||
export function ledgerOf(runId, { whole, actions, issued = [], reconciled = [], retro = false, labels } = {}) {
|
||
return {
|
||
run_id: runId, state: 'matched', note: '封存后比对，不在封存范围内', post_seal: true, artifact: 'cm-1', algo: 'v1',
|
||
trigger: retro ? 'on_demand' : 'sealed', retro, matched_at: '2026-10-15T01:00:00+00:00',
|
||
...(retro ? { retro_note: '按现在的口径回看' } : {}),
|
||
calibers: [WEEKLY, ORDERS],
|
||
queries: [
|
||
{ artifact: 'q-gmv', node_id: 'ag', source: 'shop', tables: ['orders'], status: 'proven',
|
||
candidates: [{ ...WEEKLY, node_id: 'fetch', status: 'proven', how: 'independent' }] },
|
||
{ artifact: 'q-orders', node_id: 'ag', source: 'shop', tables: ['orders'], status: 'related',
|
||
candidates: [{ ...ORDERS, node_id: 'fetch', status: 'related', comparable: true, via: 'heuristic', diff: DIFF,
|
||
binding: [{ field: 'start', value: '2026-09-01' }] }] },
|
||
],
|
||
reports: { write: { labels: labels ?? {
|
||
s6: caliberLabel(WEEKLY, 'independent', { binding: [{ field: 'week', value: '2026-W37' }], card: 'caliber', metric: 'gmv',
|
||
queries: ['q-gmv'] }),
|
||
s10: caliberLabel(ORDERS, 'composed', { binding: [{ field: 'store', masked: true, value: '已遮罩' }], card: 'caliber',
|
||
metric: 'orders', queries: ['q-gmv'] }),
|
||
s8: { kind: 'related', related: [{ caliber: ORDERS.caliber, workflow_id: ORDERS.workflow_id, version: 1, level: 'published',
|
||
level_text: '已发布', workflow_name: '订单口径', node_id: 'fetch' }], queries: ['q-orders'] },
|
||
s12: { kind: 'reasoning', query: 'q-gmv' },
|
||
s23: { kind: 'unknown', reason: '比对超过时限，没有做完', query: 'q-gmv' },
|
||
} } },
|
||
primary: 'write',
|
||
whole: whole ?? { equivalent: false, reason: '有的数是这次的推理', counts: { caliber: 2, related: 1, reasoning: 1, unknown: 1 } },
|
||
actions: actions ?? [RECONCILE],
|
||
issued, reconciled,
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 答案身上的召回标记（服务端 ledger/recall.answer_marks）：周报口径升到 v3 重算过、数没变（同一份口径核对过两次，只写最近
|
||
* 那一版）；目录里影响数字的两项被驳回、按新目录检查新出现 1 条错误级问题，另有一次出现了新码值（同归「数据目录」那一类，
|
||
* 项数相加）；看的这个人手上有一条说的是这个答案的 recall 待办（订单口径升到 v2，数变了）
|
||
*/
|
||
export const RECALL = {
|
||
checked: [
|
||
{ workflow_id: 'wf-weekly', workflow_name: '周报口径', from_version: 1, to_version: 2, recall_run: 'run-rc0', changed: false,
|
||
at: '2026-10-14T01:00:00+00:00' },
|
||
{ workflow_id: 'wf-weekly', workflow_name: '周报口径', from_version: 2, to_version: 3, recall_run: 'run-rc1', changed: false,
|
||
at: '2026-10-15T01:00:00+00:00' },
|
||
],
|
||
basis: [
|
||
{ kind: 'catalog', items: 2, new_errors: 1, at: '2026-10-15T02:00:00+00:00' },
|
||
{ kind: 'values', items: 1, new_errors: 0, at: '2026-10-15T03:00:00+00:00' },
|
||
],
|
||
tasks: [{ id: 'tk-recall-1', reason: 'caliber', to_version: 2 }],
|
||
}
|
||
|
||
/** 看差多少（s8 那条查询、订单口径）：参数是推断的，要先核对 */
|
||
export const RECONCILE = {
|
||
kind: 'reconcile', query: 'q-orders', caliber: ORDERS.caliber, workflow_id: ORDERS.workflow_id, version: 1, node_id: 'fetch',
|
||
workflow_name: '订单口径', level_text: '已发布', needs_check: true, check_reason: '参数按这次的 SQL 推断，请核对',
|
||
binding: [{ field: 'start', value: '2026-09-01' }], fallback: false,
|
||
fields: [{ name: 'start', required: true, has_default: false }, { name: 'store', required: false, has_default: true }],
|
||
}
|
||
|
||
/** 按口径正式出具（整题等价于周报口径）：搭建者能一键，不自动的原因写在 auto_reason */
|
||
export const ISSUE = {
|
||
kind: 'issue', workflow_id: WEEKLY.workflow_id, version: 2, caliber: WEEKLY.caliber, workflow_name: '周报口径', how: 'independent',
|
||
level: 'governed', level_text: '受管', available: true, auto: false, auto_reason: '设置里关了自动按口径出具',
|
||
}
|
||
|
||
/** 整题等价时的标签（三个数都来自周报口径） */
|
||
export const WHOLE_LABELS = {
|
||
s6: caliberLabel(WEEKLY, 'independent', { binding: [{ field: 'week', value: '2026-W37' }], queries: ['q-gmv'] }),
|
||
s10: caliberLabel(WEEKLY, 'independent', { binding: [{ field: 'week', value: '2026-W37' }], queries: ['q-gmv'] }),
|
||
s12: caliberLabel(WEEKLY, 'independent', { binding: [{ field: 'week', value: '2026-W37' }], queries: ['q-gmv'] }),
|
||
}
|
||
export const WHOLE = { equivalent: true, caliber: WEEKLY.caliber, workflow_id: WEEKLY.workflow_id, version: 2, level: 'governed',
|
||
level_text: '受管', workflow_name: '周报口径', how: 'independent', binding: [{ field: 'week', value: '2026-W37' }],
|
||
missing: [], missing_required: [], counts: { caliber: 3, related: 0, reasoning: 0, unknown: 0 } }
|
||
|
||
/** 对过的一次账（单行：一组对照）。status 可以换 */
|
||
export const reconciledOf = (status = 'succeeded') => ({
|
||
run_id: 'run-rec-1', status, query: 'q-orders', caliber: ORDERS.caliber, workflow_id: ORDERS.workflow_id, version: 1,
|
||
inputs_from: 'checked', rows: null, card: 'reconcile',
|
||
groups: [{ label: '订单数', ours: 'ours_0', caliber: 'cal_0', diff: 'diff_0', rel: 'rel_0' }], created_at: '2026-10-15T02:00:00Z',
|
||
})
|
||
|
||
/** 对账卡上四个指标的展示值（/evidence/metrics/reconcile/{id} 的 rendered） */
|
||
export const RECONCILE_VALUES = { ours_0: '1,234单', cal_0: '1,180单', diff_0: '54单', rel_0: '4.58%' }
|
||
|
||
/** 对账运行的指标出处（形状同片段接口：chain、seal），给 EvidenceDrawer 画 */
|
||
export const metricDetail = (runId, metric) => ({
|
||
run_id: runId, node: 'reconcile', metric, rendered: RECONCILE_VALUES[metric] ?? '—', eid: `ev:metric:${metric}`,
|
||
chain: [{ step: 'metric', alias: 'm', metric, name: metric, value: 1, rendered: RECONCILE_VALUES[metric], node_id: 'reconcile',
|
||
caliber: '口径对账', version: 'v1', expression: 'cell(nodes.ours_q1, 0, \'orders\')', sealed: true, inputs: [] }],
|
||
seal: { sealed: true, ok: true, covered: true }, redacted: { columns: [] },
|
||
})
|
||
|
||
/** 认证队列：三类来源各一项 */
|
||
export const QUEUE = {
|
||
items: [
|
||
{ id: 'shape:aa11', kind: 'shape', source: 'shop', shape: 'SELECT SUM(amount) FROM orders WHERE paid_at >= ? AND status = ?',
|
||
runs: 4, conversations: 3, question: '上周销售额是多少',
|
||
turns: [{ run_id: 'run-s1', conversation_id: 'c-1', turn_id: 't-1', question: '上周销售额是多少' },
|
||
{ run_id: 'run-s2', conversation_id: 'c-2', turn_id: 't-2', question: '这周销售额是多少' }],
|
||
params: [{ index: 0, distinct: 3 }], constants: 1, related: [{ caliber: ORDERS.caliber, workflow_id: ORDERS.workflow_id, version: 1, level: 'published' }],
|
||
representative: { run_id: 'run-s1', query: 'q-1' }, peers: ['run-s2', 'run-s3'], frozen_into: [] },
|
||
{ id: 'cluster:bb22', kind: 'cluster', question: '各门店的退款率', count: 3,
|
||
turns: [{ run_id: 'run-c1', turn_id: 't-c1', question: '各门店的退款率' }, { run_id: 'run-c2', turn_id: 't-c2', question: '门店退款率怎么样' }],
|
||
frozen_into: [] },
|
||
{ id: 'reused:cc33', kind: 'reused', question: '本月新客数', uses: { exports: 2, app: false, extracted: false, canvas: true },
|
||
turns: [{ run_id: 'run-r1', conversation_id: 'c-3', turn_id: 't-r1', question: '本月新客数' }],
|
||
representative: { run_id: 'run-r1' }, frozen_into: [{ workflow_id: 'wf-new-old', name: '本月新客数（固化的口径草稿）', status: 'draft' }] },
|
||
],
|
||
considered: { runs: 120, matched: 100, questions: 90 },
|
||
threshold: { runs: 3, conversations: 2, exports: 2 },
|
||
}
|