Archived
凭证到位后 `w05` 五案例连跑六轮,逼出八条**浏览器走查逮不到**的缺陷 —— 脚本每张卡都先空手交一次、每个岔路口都按序代答,人手点的时候永远不会去试 「空着交上去会怎样」。末轮:跑到的四例全部 released,三例被拒 0 (第五例停在大模型账户余额不足,不是代码问题)。 平台侧(六条): - L 数据工单的提交闸没牙 —— 卡没绑表就回落到必交付物,而数据工单要人补的是 `parts.*` 主数据,不在交付物名单里:闸盯着一堆无关的表、一格不缺,空手交卷一路放行。 契约新增 `ExecutionWorkOrder.gateLedgers`(这张卡该拦哪几张表,发卡的人说了算), `gateAddressesOf` 变三级取一:绑的表 > 卡自报 > 交付物。 - M `finish` 撞 doneWhen 为假的拒回只谈「行与列」,而缺的是一次**工具调用**; 现在摆出判据原文 + 那几个名字此刻的读数,`tool_*_calls` 读到 0 的单独点名(`unmetWhy`)。 - N 人点了决策卡,模型每拍再发一张,连拒五拍熔断。把「他选的是 approve」写进拒回**没用** —— 下一拍它又从头翻一次会滚走的窗口。改结构:`answeredCardFact` 把 「这张卡已答过 → 他选的是什么」做成每拍都在的事实行。 - O 模型传 `requirement: null`,工具 `null.trim()` 当场崩(边界上一律经 `text()` 归一)。 - P 拒回里的 `count=(没写过)` 是 S-FEEL 函数名被当成台账名 —— 假线索模型会去追。 - Q/S 零行 = 零缺,闸与哨兵一起放过空手交卷的数据工单(`emptyGate`:卡自报的表一行都没有 = 没完成);「早收」头一次只落观察、第二次起才算被拒(与「重发已答过的卡」同一把尺)。 活动侧(1.2.7 → 1.2.9):三张执行工单各自声明 `gateLedgers`;说明书③补「发卡前骨架必须先有行」、 ⑤补这张卡的前提(一颗 absent 都不剩);哨兵 `parts_profile_rows >= created` 逐颗对账; S0_bind 写明发录入卡那一拍别抢写需求行。 e2e 脚本加 `JCAI_E2E_BEARER` 直供旁路(服务账号 client_credentials 仍是正路)。 门禁 typecheck / lint / depcruise / test(2091)全绿;api:check 的差异仍只在 org/* 与 api/* (工作树里先前那批鉴权改动),card.d.ts 是本轮的纯新增字段。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FJcX8A6TWp4GRs41TYryEh
291 lines
17 KiB
TypeScript
291 lines
17 KiB
TypeScript
/**
|
|
* 步骤剧本与呈现面 —— 这段错了不报错,只是活动开始转圈 / 界面开始撒谎(与 E04 / I01 同一套惯例)。
|
|
*/
|
|
import { describe, expect, it } from 'vitest';
|
|
import { evaluateCondition } from '@sfeel';
|
|
import { ASSESSMENT_PHASES, ASSESSMENT_STEPS, CMMN_XML, NARRATIVE } from '../src/case-model.ts';
|
|
import { manifest, D } from '../src/manifest.ts';
|
|
|
|
const ITEM_IDS = [...CMMN_XML.matchAll(/<(?:humanTask|decisionTask|stage|milestone)\s+id="([^"]+)"/g)].map((m) => m[1] ?? '');
|
|
|
|
function doneOf(item: string): string {
|
|
const found = ASSESSMENT_STEPS.find((s) => s.item === item)?.doneWhen;
|
|
if (found === undefined) throw new Error(`${item} 没有 doneWhen`);
|
|
return found;
|
|
}
|
|
/** 谓词写坏了引擎当作「没算出来」——测里也照这个口径 */
|
|
function safeEval(expression: string, scope: Record<string, unknown>): boolean {
|
|
try {
|
|
return evaluateCondition(expression, scope);
|
|
} catch {
|
|
return false;
|
|
}
|
|
}
|
|
/** 开局局面:入参 + 所有基座读数归零 */
|
|
const fresh = (input: Record<string, unknown>): Record<string, unknown> => ({
|
|
...input,
|
|
demand_request_rows: 0,
|
|
demand_request_missing: 0,
|
|
[`${D}_verdict_rows`]: 0,
|
|
[`${D}_pn_check_rows`]: 0,
|
|
[`${D}_pn_check_missing`]: 0,
|
|
[`${D}_pn_check_exists_status_absent`]: 0,
|
|
[`${D}_pn_check_exists_status_abandoned`]: 0,
|
|
[`${D}_input_gap_rows`]: 0,
|
|
[`${D}_input_gap_missing`]: 0,
|
|
input_confirmed_total: 0,
|
|
review_approved_total: 0,
|
|
landing_prepared_total: 0,
|
|
substitution_relation_rows: 0,
|
|
substitution_relation_missing: 0,
|
|
substitution_relation_analysis_status_fae_confirmed: 0,
|
|
substitution_difference_item_rows: 0,
|
|
substitution_difference_item_missing: 0,
|
|
substitution_difference_item_distinct_substitution_relation_id: 0,
|
|
substitution_difference_item_difference_status_match: 0,
|
|
substitution_difference_item_difference_status_mismatch: 0,
|
|
substitution_difference_item_difference_status_unknown: 0,
|
|
substitution_difference_item_difference_status_conditional: 0,
|
|
tool_part_substitution_assessment_compare_parameters_calls: 0,
|
|
tool_part_substitution_assessment_recommend_calls: 0,
|
|
});
|
|
|
|
describe('剧本与案例模型对得上', () => {
|
|
it('每一条剧本都指向一个真实存在的计划项,图上每一步也都有剧本', () => {
|
|
for (const step of ASSESSMENT_STEPS) expect(ITEM_IDS, `剧本里的 ${step.item} 在 CMMN 里不存在`).toContain(step.item);
|
|
for (const id of ITEM_IDS) expect(ASSESSMENT_STEPS.map((s) => s.item), `计划项 ${id} 没有剧本`).toContain(id);
|
|
});
|
|
|
|
it('清单里挂的就是这份剧本与这份阶段', () => {
|
|
expect(manifest.assets.businessModel.steps).toBe(ASSESSMENT_STEPS);
|
|
expect(manifest.assets.businessModel.phases).toBe(ASSESSMENT_PHASES);
|
|
});
|
|
});
|
|
|
|
describe('剧本声明的读写面 ⊆ 清单声明', () => {
|
|
const declaredAddresses = new Set(manifest.declarations.ledger.map((d) => `${d.domain}.${d.object}`));
|
|
const declaredTools = new Set(manifest.declarations.tools.map((t) => t.name));
|
|
|
|
it('steps[].writes / reads(台账那几个)全落在声明过的对象上', () => {
|
|
for (const step of ASSESSMENT_STEPS) {
|
|
for (const address of step.writes) expect(declaredAddresses.has(address), `${step.item} 要写 ${address}`).toBe(true);
|
|
for (const address of step.reads.filter((r) => r.includes('.') && !r.startsWith('case:'))) {
|
|
expect(declaredAddresses.has(address), `${step.item} 要读 ${address}`).toBe(true);
|
|
}
|
|
}
|
|
});
|
|
|
|
it('steps[].tools 全部是声明过的工具;只有 S3 调工具', () => {
|
|
for (const step of ASSESSMENT_STEPS) {
|
|
for (const tool of step.tools) expect(declaredTools.has(tool), `${step.item} 要调 ${tool}`).toBe(true);
|
|
if (step.item !== 'S3_assess') expect(step.tools, step.item).toEqual([]);
|
|
}
|
|
expect(ASSESSMENT_STEPS.find((s) => s.item === 'S3_assess')?.tools).toHaveLength(2);
|
|
});
|
|
|
|
it('customer_approval_status 在任何剧本里都不许写(只读边界)', () => {
|
|
for (const step of ASSESSMENT_STEPS) expect(step.guide, step.item).not.toMatch(/customer_approval_status\s*=/);
|
|
});
|
|
});
|
|
|
|
describe('doneWhen —— 开局不许真,干完必须真', () => {
|
|
it('**开局一步都不该被自动收掉** —— 五条样例的入参形状逐一试', () => {
|
|
for (const sample of manifest.samples) {
|
|
for (const step of ASSESSMENT_STEPS) {
|
|
if (step.doneWhen === undefined) continue;
|
|
expect(safeEval(step.doneWhen, fresh(sample.input as Record<string, unknown>)), `样例「${sample.name}」下 ${step.item} 开局就为真`).toBe(false);
|
|
}
|
|
}
|
|
});
|
|
|
|
it('S0:两颗候选 + 原 PN = 3 行核对齐、需求与结论都写了 → 收得掉;少写一颗 → 收不掉', () => {
|
|
const base = { candidateInternalPns: ['XL2596S-5.0', 'LM2596S-ADJ'], demand_request_rows: 1, demand_request_missing: 0, [`${D}_verdict_rows`]: 1, [`${D}_pn_check_missing`]: 0 };
|
|
expect(safeEval(doneOf('S0_bind'), { ...base, [`${D}_pn_check_rows`]: 3 })).toBe(true);
|
|
expect(safeEval(doneOf('S0_bind'), { ...base, [`${D}_pn_check_rows`]: 2 })).toBe(false);
|
|
});
|
|
|
|
it('S2:未放弃的候选各一条关系且每条都有明细 → 收得掉;放弃一颗后关系数随之减一', () => {
|
|
const two = { candidateInternalPns: ['A', 'B'], [`${D}_pn_check_exists_status_abandoned`]: 0, substitution_difference_item_rows: 14 };
|
|
expect(safeEval(doneOf('S2_normalize'), { ...two, substitution_relation_rows: 2, substitution_difference_item_distinct_substitution_relation_id: 2 })).toBe(true);
|
|
expect(safeEval(doneOf('S2_normalize'), { ...two, substitution_relation_rows: 2, substitution_difference_item_distinct_substitution_relation_id: 1 })).toBe(false);
|
|
expect(safeEval(doneOf('S2_normalize'), { ...two, [`${D}_pn_check_exists_status_abandoned`]: 1, substitution_relation_rows: 1, substitution_difference_item_distinct_substitution_relation_id: 1 })).toBe(true);
|
|
});
|
|
|
|
it('S3:四态之和 = 明细行数、必填齐、两枚工具都调过 → 收得掉;工具没调过 → 收不掉', () => {
|
|
const ok = {
|
|
substitution_difference_item_missing: 0,
|
|
substitution_relation_missing: 0,
|
|
substitution_difference_item_rows: 7,
|
|
substitution_difference_item_difference_status_match: 4,
|
|
substitution_difference_item_difference_status_mismatch: 1,
|
|
substitution_difference_item_difference_status_unknown: 1,
|
|
substitution_difference_item_difference_status_conditional: 1,
|
|
tool_part_substitution_assessment_compare_parameters_calls: 1,
|
|
tool_part_substitution_assessment_recommend_calls: 1,
|
|
};
|
|
expect(safeEval(doneOf('S3_assess'), ok)).toBe(true);
|
|
expect(safeEval(doneOf('S3_assess'), { ...ok, tool_part_substitution_assessment_recommend_calls: 0 })).toBe(false);
|
|
expect(safeEval(doneOf('S3_assess'), { ...ok, substitution_difference_item_difference_status_match: 3 })).toBe(false);
|
|
});
|
|
|
|
it('S5:旗 + 全部关系 FAE 确认 + P-01 回写齐 → 收得掉;缺一头都收不掉', () => {
|
|
const ok = { landing_prepared_total: 1, substitution_relation_analysis_status_fae_confirmed: 2, substitution_relation_rows: 2, substitution_relation_missing: 0, demand_request_missing: 0 };
|
|
expect(safeEval(doneOf('S5_prepare_landing'), ok)).toBe(true);
|
|
expect(safeEval(doneOf('S5_prepare_landing'), { ...ok, substitution_relation_analysis_status_fae_confirmed: 1 })).toBe(false);
|
|
expect(safeEval(doneOf('S5_prepare_landing'), { ...ok, demand_request_missing: 1 })).toBe(false);
|
|
});
|
|
});
|
|
|
|
describe('说明书与剧本的骨架', () => {
|
|
it('五张卡的骨架都住在说明书里(压拍之后只有说明书还在)', () => {
|
|
expect(NARRATIVE).toContain('"kind":"entry","itemId":"S0_bind"');
|
|
expect(NARRATIVE).toContain('"kind":"decide","itemId":"S0d_p08_master"');
|
|
expect(NARRATIVE).toContain('"title":"数据工单 · 芯片主数据创建(P-08)');
|
|
expect(NARRATIVE).toContain('"title":"数据工单 · 原厂技术资料创建(P-10)');
|
|
expect(NARRATIVE).toContain('"title":"评估输入与资料补齐确认');
|
|
expect(NARRATIVE).toContain('"kind":"decide","itemId":"S4_review"');
|
|
});
|
|
|
|
it('剧本里不引反例原话;S3 固定三拍且写明「工具后同拍作废」', () => {
|
|
const s3 = ASSESSMENT_STEPS.find((s) => s.item === 'S3_assess')?.guide ?? '';
|
|
expect(s3).toContain('第一拍');
|
|
expect(s3).toContain('第二拍');
|
|
expect(s3).toContain('第三拍');
|
|
expect(s3).toContain('工具调用后同拍其余动作作废');
|
|
});
|
|
|
|
it('S3 第三拍写盘前先自检两枚工具的调用次数(2026-08-27 实测:漏调 recommend 会让 doneWhen 恒假、模型死循环重写同样的行)', () => {
|
|
const s3 = ASSESSMENT_STEPS.find((s) => s.item === 'S3_assess')?.guide ?? '';
|
|
const thirdPass = s3.slice(s3.indexOf('第三拍'));
|
|
expect(thirdPass).toContain('tool_part_substitution_assessment_recommend_calls');
|
|
expect(thirdPass).toContain('tool_part_substitution_assessment_compare_parameters_calls');
|
|
expect(thirdPass).toMatch(/自检|先.*找这两个数|必须\s*>\s*0/);
|
|
// 第二拍那一段要点出「推荐等级只能由工具算,不调它收不了步」,别让模型自己拍推荐等级
|
|
const secondPass = s3.slice(s3.indexOf('第二拍'), s3.indexOf('第三拍'));
|
|
expect(secondPass).toContain('推荐等级只能由它算');
|
|
expect(secondPass).toMatch(/收不了步|收不掉/);
|
|
});
|
|
|
|
it('S0 剧本写明都在库时数据工单两项本就不会进「当下可为」;S0d_p08_master 剧本按候选总数判断该不该问人(=1 必建、≥2 才问、原 PN 硬阻塞)', () => {
|
|
const s0 = ASSESSMENT_STEPS.find((s) => s.item === 'S0_bind')?.guide ?? '';
|
|
expect(s0).toContain('当下可为');
|
|
expect(s0).toContain('不要对计划项用');
|
|
const s0d = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master')?.guide ?? '';
|
|
expect(s0d).toContain('硬阻塞');
|
|
expect(s0d).toContain('候选总数 = 1');
|
|
});
|
|
|
|
it('S0d_p08_master 剧本按候选总数(不是「不在库的颗数」)判断要不要问人 —— 候选总数 ≥2 时逐颗不在库的都要问(2026-08-27 实测逼出的措辞修正)', () => {
|
|
const s0d = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master')?.guide ?? '';
|
|
expect(s0d).toContain('候选总数');
|
|
expect(s0d).toContain('不在库');
|
|
expect(s0d).toMatch(/候选总数\s*≥\s*2/);
|
|
// 反面提示:「不在库的只有一颗」不该被读成跳过决策卡的理由
|
|
expect(s0d).toContain('不是跳过决策卡的理由');
|
|
});
|
|
|
|
it('s_bound 哨兵真的检查了「有料号不在库」这件事(都在库时门不该开;退回旧写法会红)', () => {
|
|
expect(CMMN_XML).toContain('sub_eval_pn_check_exists_status_absent > 0');
|
|
const sBoundBlock = /<sentry id="s_bound">[\s\S]*?<\/sentry>/.exec(CMMN_XML)?.[0] ?? '';
|
|
expect(sBoundBlock).toContain('caseFileItemRef="sub_eval_pn_check_exists_status_absent"');
|
|
expect(sBoundBlock).toContain('sub_eval_pn_check_exists_status_absent > 0');
|
|
});
|
|
|
|
it('说明书与全部剧本里不会教模型对计划项调用 disable(允许出现在「不要 disable」这类禁止句里)', () => {
|
|
// 只逮「disable + 计划项 id」这种祈使形(教模型真的去调用它);
|
|
// 「不要…用 `disable`」「Disabled 只能从 Enabled 来」这类禁止句/状态名不含「disable + id」,不会被误伤。
|
|
const disableCallPattern = /disable\s+(S\d[a-zA-Z_]*|T_release)/;
|
|
expect(NARRATIVE).not.toMatch(disableCallPattern);
|
|
for (const step of ASSESSMENT_STEPS) expect(step.guide, step.item).not.toMatch(disableCallPattern);
|
|
});
|
|
|
|
it('S0d 拆成两项各自绑自己的 view(P-10 那份件不会永远推不出来),两项都没有 doneWhen', () => {
|
|
const p08 = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master');
|
|
const p10 = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p10_document');
|
|
expect(p08?.view).toBe('p08-master-create');
|
|
expect(p10?.view).toBe('p10-document-create');
|
|
expect(p08?.doneWhen).toBeUndefined();
|
|
expect(p10?.doneWhen).toBeUndefined();
|
|
});
|
|
|
|
it('候选放弃到只剩最后一颗时不许再放弃(至少留一颗,防止全放弃死锁)', () => {
|
|
const s0d = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master')?.guide ?? '';
|
|
expect(s0d).toContain('至少留一颗候选');
|
|
});
|
|
|
|
it('P-08 交回之后先 finish 再发 P-10 卡,顺序写死进两段 guide(2026-08-27 实测:漏收步会让工作区画面停在 P-08 那两张表,P-10 卡发出后人却看不见资料表、提交闸拦死)', () => {
|
|
const s0d = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master')?.guide ?? '';
|
|
expect(s0d).toContain('finish S0d_p08_master');
|
|
expect(s0d).toMatch(/顺序不能反/);
|
|
expect(s0d).toContain('p08-master-create');
|
|
const s0d10 = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p10_document')?.guide ?? '';
|
|
expect(s0d10).toContain('finish S0d_p08_master');
|
|
expect(s0d10).toMatch(/若它还开着|先.*finish/);
|
|
});
|
|
|
|
it('「发卡前清单必须先有行」写在说明书里,不只在剧本里(2026-08-27 实测:模型把 P-10 收步与 S1 发卡压成一拍,那一拍 S1 剧本没摊开,卡发出去时 input_gap 一行都没有,人对着空表无从下手)', () => {
|
|
expect(NARRATIVE).toContain('sub_eval.input_gap` 必须先有行');
|
|
expect(NARRATIVE).toMatch(/压拍/);
|
|
expect(NARRATIVE).toMatch(/顺序不能反/);
|
|
});
|
|
|
|
it('存在性核对的剧本教的是「填得出依据才算在库」,不是「你看着办」(结构治,配 found_pn_ref 那道墙)', () => {
|
|
const g = ASSESSMENT_STEPS.find((s) => s.item === 'S0_bind')?.guide ?? '';
|
|
expect(g).toContain('found_pn_ref');
|
|
expect(g).toMatch(/查无此值当场拒/);
|
|
expect(g).toMatch(/名单里没有就是没有/);
|
|
expect(g).toMatch(/把这一行改成 absent/);
|
|
});
|
|
|
|
it('数据工单的名单只认 absent,「建哪几颗」与「要不要问人」分开写(2026-08-27 实测:原 PN 缺库那单,模型把在库的候选派成了数据工单,缺的那颗没人建)', () => {
|
|
const g = ASSESSMENT_STEPS.find((s) => s.item === 'S0d_p08_master')?.guide ?? '';
|
|
expect(g).toMatch(/问题一 · 建哪几颗/);
|
|
expect(g).toMatch(/问题二 · 建之前要不要先问人/);
|
|
expect(g).toMatch(/exists_status = "found" 的一行都不建/);
|
|
});
|
|
|
|
it('数据录入卡自带 basis,不让基座兜上内部家具(2026-08-27 实测:没声明 basis 的那张卡,参考信息栏摆的是「这一步 / 当下可为 / 目标缺口」加九个计划项全名)', () => {
|
|
const i = NARRATIVE.indexOf('"kind":"entry"');
|
|
expect(i).toBeGreaterThan(-1);
|
|
const block = NARRATIVE.slice(i, i + 900);
|
|
expect(block).toContain('"basis"');
|
|
expect(block).toContain('触发原因');
|
|
});
|
|
|
|
it('三张执行工单各自声明 gateLedgers —— 零输入卡不声明,闸就盯着与它无关的表(2026-08-27 e2e 实测:两张数据工单空手交卷都放行了)', () => {
|
|
for (const [item, ledger] of [
|
|
['芯片主数据创建', '"gateLedgers":["parts.profile","parts.parameter"]'],
|
|
['原厂技术资料创建', '"gateLedgers":["parts.document"]'],
|
|
['评估输入与资料补齐确认', '"gateLedgers":["sub_eval.input_gap"]'],
|
|
] as const) {
|
|
expect(NARRATIVE, item).toContain(ledger);
|
|
}
|
|
});
|
|
|
|
it('数据工单也要「先写骨架再发卡」写进说明书,且哨兵不认零行主数据(2026-08-27 e2e:模型压拍没写骨架就发卡,人打开是空表,零行同时骗过闸与哨兵,一路走到 S2 才被引用检查拦下)', () => {
|
|
expect(NARRATIVE).toContain('必须先有骨架行');
|
|
expect(NARRATIVE).toMatch(/压拍/);
|
|
// 标了几颗建好,主数据就得有几行 —— 逐颗对账(第一刀写成「≥1 行就算数」太松,三候选那单从缝里过去了)
|
|
expect(CMMN_XML).toContain('parts_profile_rows >= sub_eval_pn_check_exists_status_created');
|
|
});
|
|
|
|
it('评估输入那张卡的**前提**写在说明书的卡骨架上(2026-08-27 e2e:两颗缺库,模型只问只建了第一颗,CN 还 absent 就把 S1 的卡发了出去 —— 发卡的授权面是调人预算,绕得过哨兵)', () => {
|
|
expect(NARRATIVE).toContain('一颗 absent 都不剩');
|
|
expect(NARRATIVE).toMatch(/有几颗 absent 就要办几颗/);
|
|
});
|
|
|
|
it('发数据录入卡那一拍不许顺手写需求行(客户还没落草稿层,引用检查必拒)', () => {
|
|
const g = ASSESSMENT_STEPS.find((s) => s.item === 'S0_bind')?.guide ?? '';
|
|
expect(g).toMatch(/那一拍只发卡/);
|
|
expect(g).toContain('引用检查');
|
|
});
|
|
|
|
it('五段进度,每段名 ≤10 字;除放行外每项恰属一段', () => {
|
|
expect(ASSESSMENT_PHASES).toHaveLength(5);
|
|
for (const p of ASSESSMENT_PHASES) expect(p.name.length, p.name).toBeLessThanOrEqual(10);
|
|
const all = ASSESSMENT_PHASES.flatMap((p) => p.items);
|
|
expect(new Set(all).size).toBe(all.length);
|
|
expect([...all].sort()).toEqual([...ITEM_IDS.filter((i) => i !== 'T_release')].sort());
|
|
});
|
|
});
|