mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 04:41:57 +00:00
证据接口的核查状态类型挪进 types.ts,证据面板的反方审稿改用单句展示:EvidenceDualCheck、EvidenceUnitChallenge、输入步骤的 masked / mask_unverified、证据图的 challenge_verdicts 照服务端逐字核对;审批卡上的概况类型补上 unreviewed、degrades。ChallengeSummary 加单句模式(scope=claim,只画这一句的判断、它可能没反查成的说明、展开看理由和查询,不画报告级的标题和计数)和显式的「不谈出具」开关(issuance=false,不预告降档、也不为此取运行类别),证据面板改用这两样,去掉传 runId 为空的做法;check-evidence 的核查一节跟着改成单句的断言
This commit is contained in:
@@ -3489,7 +3489,8 @@ await section('verify', '期 4:指标步骤的双盲复算、取自遮罩列
|
||||
check('输入来源:没遮罩的字段照旧', verified.includes('与快照一致') && !verified.includes('遮罩'), verified.replace(/\s+/g, ' '))
|
||||
await q.ctx.close()
|
||||
|
||||
// ---- 反方审稿:整句面板、片段面板都看得到这一句的判断。复用审批卡的展示,概况是整份报告的真实计数,不预告出具
|
||||
// ---- 反方审稿:整句面板、片段面板都看得到这一句的判断。用审批卡那一块的单句展示:只画这一句的判断,不画报告级的
|
||||
// 标题和计数,不预告出具
|
||||
const u = fxj.formal.units
|
||||
const challenge = (verdict) => ({ verdict, node_id: 'chal', artifact: CH_ART, on_refuted: 'degrade', report_label: '写周报',
|
||||
counts: CH_COUNTS })
|
||||
@@ -3528,10 +3529,12 @@ await section('verify', '期 4:指标步骤的双盲复算、取自遮罩列
|
||||
await c.page.waitForSelector('[data-evidence-panel] [data-ev-challenge] [data-challenge-item]', { timeout: 4000 }).catch(() => {})
|
||||
const text = await blockText()
|
||||
if (w === 1280) {
|
||||
check('整句面板:反方审稿的判断(被推翻)挂在这一句下,写明是哪份报告的审稿', await block.getAttribute('data-ev-challenge', { timeout: 3000 }).catch(() => null) === 'refuted'
|
||||
&& text.includes('报告「写周报」的反方审稿') && text.includes('被推翻') && text.includes('新客首单'), text)
|
||||
check('……概况是整份报告的真实计数(服务端给的),不凭这一句拼', text.includes('经受住 3 句') && text.includes('被削弱 1 句')
|
||||
&& text.includes('被推翻 2 句') && text.includes('无法判断 0 句') && await block.locator('[data-challenge-more="5"]').count() === 1, text)
|
||||
check('整句面板:「反方审稿」一节写这一句的判断(被推翻),用单句展示', await block.getAttribute('data-ev-challenge', { timeout: 3000 }).catch(() => null) === 'refuted'
|
||||
&& text.startsWith('反方审稿') && await block.locator('[data-challenge-scope="claim"]').count() === 1
|
||||
&& (await block.locator('[data-challenge-tag="refuted"]').innerText().catch(() => '')).trim() === '被推翻', text)
|
||||
check('……单句只画这一句:不画报告级的标题、四种判断的计数、别的句子,也不再抄一遍这一句的原文',
|
||||
await block.locator('[data-challenge-counts], [data-challenge-items], [data-challenge-more], [data-challenge-show-all]').count() === 0
|
||||
&& !text.includes('报告「写周报」的反方审稿') && !/\d+ 句/.test(text) && !text.includes('增长主要来自新客首单'), text)
|
||||
check('……运行已经结束:不预告「出具会降档」,降没降档看报告的出具横幅', await block.locator(ISSUANCE_LINES).count() === 0
|
||||
&& !text.includes('降档'), text)
|
||||
check('……判断由反方模型给出,并非系统核对', text.includes('并非系统核对'), text)
|
||||
@@ -3539,12 +3542,12 @@ await section('verify', '期 4:指标步骤的双盲复算、取自遮罩列
|
||||
await c.page.waitForSelector('[data-evidence-panel] [data-challenge-detail-state="ok"]', { timeout: 4000 }).catch(() => {})
|
||||
const why = await block.locator('[data-challenge-rationale]').innerText().catch(() => '')
|
||||
check('……点开看得到反方的理由(按审稿记录取)', why.includes(CH_RATIONALE) && artifactHits.length === 1, why)
|
||||
// 被削弱的那一句:概况照样是整份报告的数,不出现凭这一句拼的「被推翻 0 句」
|
||||
// 被削弱的那一句:只写它自己的判断,不写整份报告的计数(以前借报告级的展示,读着像这一句「被推翻 0 句」)
|
||||
await badge(c.page, J, u.ok).click()
|
||||
await c.page.waitForSelector('[data-evidence-panel] [data-ev-challenge="weakened"]', { timeout: 4000 }).catch(() => {})
|
||||
const weak = await blockText()
|
||||
check('被削弱的那一句:概况写整份报告的计数(被推翻 2 句),不写凭这一句拼的「被推翻 0 句」', weak.includes('被推翻 2 句')
|
||||
&& !weak.includes('被推翻 0 句') && !weak.includes('被削弱 1 句 · 被推翻 0 句'), weak)
|
||||
check('被削弱的那一句:只写这一句的判断(被削弱),不写整份报告的计数', weak.includes('被削弱') && !weak.includes('被推翻')
|
||||
&& !/\d+ 句/.test(weak), weak)
|
||||
// 片段面板:点这一句里的数字,同样看得到判断;正式运行里双盲复算记的是出具缺口
|
||||
await c.page.locator(`${J} [data-seg="s6"]`).click()
|
||||
await c.page.waitForSelector('[data-evidence-panel] [data-ev-challenge="weakened"]', { timeout: 4000 }).catch(() => {})
|
||||
@@ -3567,7 +3570,7 @@ await section('verify', '期 4:指标步骤的双盲复算、取自遮罩列
|
||||
await c.page.locator(`${JP} [data-seg="s14"]`).click()
|
||||
await c.page.waitForSelector('[data-evidence-panel] [data-ev-challenge]', { timeout: 4000 }).catch(() => {})
|
||||
check('没开结论句裁判的报告:点句中的数字看得到反方审稿判断', await block.getAttribute('data-ev-challenge', { timeout: 3000 }).catch(() => null) === 'refuted'
|
||||
&& (await blockText()).includes('退款 29 单'), await blockText())
|
||||
&& (await block.locator('[data-challenge-tag="refuted"]').count()) === 1, await blockText())
|
||||
} else {
|
||||
const wide = await c.page.evaluate(() => {
|
||||
const el = document.querySelector('[data-evidence-panel]')
|
||||
|
||||
Reference in New Issue
Block a user