使用完整筛选范围的当前有效结论计算覆盖率和正确率,人工优先,失效与对应待确认/未评估/不计入排除;零分母返回null,界面展示原始分子分母和排除构成,前端拒绝不守恒或比例不一致的响应。 验证:make verify退出码0,Go658及前端257通过,91项声明可选集成跳过;隔离PG、三包竞态及合成/真实只读源浏览器验证通过。无新增迁移,不接AI模型,不改绩效,不写源库;本批未推送或部署,8088运行版保持0.1.21。
This commit is contained in:
@@ -41,3 +41,56 @@ describe('manual quality operations', () => {
|
||||
expect(ok).toContain('AI 评估未接入')
|
||||
})
|
||||
})
|
||||
|
||||
it('displays approved rates with raw denominators and exclusions', () => {
|
||||
const html = renderToStaticMarkup(
|
||||
<QAQualitySummaryView data={parseQualitySummary(summaryFixture)} error="" loading={false} onRefresh={() => {}} />
|
||||
)
|
||||
for (const value of [
|
||||
'知识覆盖率',
|
||||
'回答正确率',
|
||||
'50.0%',
|
||||
'100.0%',
|
||||
'分子',
|
||||
'分母',
|
||||
'排除',
|
||||
'部分覆盖',
|
||||
'quality-rates-human-first-v1',
|
||||
])
|
||||
expect(html).toContain(value)
|
||||
expect(html).not.toContain('不计算比例')
|
||||
})
|
||||
it('does not turn zero denominators into 0% or 100%', () => {
|
||||
const v = parseQualitySummary(structuredClone(summaryFixture))
|
||||
v.summary = {
|
||||
total: 0,
|
||||
reviewed: 0,
|
||||
unreviewed: 0,
|
||||
stale: 0,
|
||||
pending_confirmation: 0,
|
||||
rule_uncovered: 0,
|
||||
rates: {
|
||||
policy_version: 'quality-rates-human-first-v1',
|
||||
coverage: { numerator: 0, denominator: 0, percent: null },
|
||||
correctness: { numerator: 0, denominator: 0, percent: null },
|
||||
breakdown: {
|
||||
covered: 0,
|
||||
uncovered: 0,
|
||||
partial: 0,
|
||||
pending: 0,
|
||||
unassessed: 0,
|
||||
excluded: 0,
|
||||
stale: 0,
|
||||
covered_correct: 0,
|
||||
covered_incorrect: 0,
|
||||
covered_pending: 0,
|
||||
covered_unassessed: 0,
|
||||
},
|
||||
},
|
||||
} as typeof v.summary
|
||||
const html = renderToStaticMarkup(
|
||||
<QAQualitySummaryView data={parseQualitySummary(v)} error="" loading={false} onRefresh={() => {}} />
|
||||
)
|
||||
expect(html).toContain('暂无可计算数据')
|
||||
expect(html).not.toContain('0.0%')
|
||||
})
|
||||
|
||||
+40
-1
@@ -421,8 +421,47 @@ export function QAQualitySummaryView({ data, loading, error, onRefresh }: Summar
|
||||
待重新核对 {data.summary.stale} · 人工待确认 {data.summary.pending_confirmation} · 规则命中未覆盖{' '}
|
||||
{data.summary.rule_uncovered}
|
||||
</p>
|
||||
<div className="qa-quality-rates" aria-label="质量比例">
|
||||
{(
|
||||
[
|
||||
['知识覆盖率', data.summary.rates.coverage],
|
||||
['回答正确率', data.summary.rates.correctness],
|
||||
] as const
|
||||
).map(([label, rate]) => (
|
||||
<div key={label}>
|
||||
<strong>
|
||||
{label}:{rate.percent === null ? '暂无可计算数据' : `${rate.percent.toFixed(1)}%`}
|
||||
</strong>
|
||||
<p>
|
||||
分子 {rate.numerator} / 分母 {rate.denominator};排除 {data.summary.total - rate.denominator} 条
|
||||
</p>
|
||||
</div>
|
||||
))}
|
||||
</div>
|
||||
<details className="qa-ops-organization">
|
||||
<summary>查看比例口径和分母构成</summary>
|
||||
<p>
|
||||
覆盖分母:已覆盖 {data.summary.rates.breakdown.covered} + 未覆盖 {data.summary.rates.breakdown.uncovered}{' '}
|
||||
+ 部分覆盖 {data.summary.rates.breakdown.partial};部分覆盖不计完全覆盖。
|
||||
</p>
|
||||
<p>
|
||||
覆盖分母排除:待确认 {data.summary.rates.breakdown.pending}、未评估{' '}
|
||||
{data.summary.rates.breakdown.unassessed}、不计入 {data.summary.rates.breakdown.excluded}、待重新核对{' '}
|
||||
{data.summary.rates.breakdown.stale}。
|
||||
</p>
|
||||
<p>
|
||||
正确性分母仅含已覆盖且判断明确:正确 {data.summary.rates.breakdown.covered_correct} + 错误{' '}
|
||||
{data.summary.rates.breakdown.covered_incorrect};已覆盖中的待确认{' '}
|
||||
{data.summary.rates.breakdown.covered_pending}、未评估 {data.summary.rates.breakdown.covered_unassessed}{' '}
|
||||
不计入,其余未覆盖/部分覆盖等记录也不计入。
|
||||
</p>
|
||||
<p>
|
||||
当前有效人工结论优先;精确规则命中的未覆盖可进入覆盖分母,但不证明回答错误。失效记录暂不参与两项比例,原始问答运行成功也不代表答案正确。
|
||||
</p>
|
||||
<p className="qa-ops-muted">统计规则:{data.summary.rates.policy_version}</p>
|
||||
</details>
|
||||
<p className="qa-ops-muted">
|
||||
AI 评估未接入;人工待确认是已复核中的子集,规则命中可与人工结论不同。不计算比例。读取:{stamp(data.as_of)}
|
||||
AI 评估未接入;人工待确认是已复核中的子集,规则命中可与人工结论不同。读取:{stamp(data.as_of)}
|
||||
</p>
|
||||
</>
|
||||
) : !error && !loading ? (
|
||||
|
||||
@@ -394,3 +394,17 @@
|
||||
grid-template-columns: 100px 1fr;
|
||||
}
|
||||
}
|
||||
|
||||
.qa-quality-rates {
|
||||
display: flex;
|
||||
flex-wrap: wrap;
|
||||
gap: 12px 32px;
|
||||
margin: 12px 0;
|
||||
}
|
||||
.qa-quality-rates > div {
|
||||
min-width: 0;
|
||||
overflow-wrap: anywhere;
|
||||
}
|
||||
.qa-quality-rates p {
|
||||
margin: 4px 0 0;
|
||||
}
|
||||
|
||||
@@ -40,7 +40,32 @@ export const qualityFixture = {
|
||||
effective: fields,
|
||||
}
|
||||
export const summaryFixture = {
|
||||
summary: { total: 3, reviewed: 1, unreviewed: 1, stale: 1, pending_confirmation: 0, rule_uncovered: 2 },
|
||||
summary: {
|
||||
total: 3,
|
||||
reviewed: 1,
|
||||
unreviewed: 1,
|
||||
stale: 1,
|
||||
pending_confirmation: 0,
|
||||
rule_uncovered: 2,
|
||||
rates: {
|
||||
policy_version: 'quality-rates-human-first-v1',
|
||||
coverage: { numerator: 1, denominator: 2, percent: 50 },
|
||||
correctness: { numerator: 1, denominator: 1, percent: 100 },
|
||||
breakdown: {
|
||||
covered: 1,
|
||||
uncovered: 1,
|
||||
partial: 0,
|
||||
pending: 0,
|
||||
unassessed: 0,
|
||||
excluded: 0,
|
||||
stale: 1,
|
||||
covered_correct: 1,
|
||||
covered_incorrect: 0,
|
||||
covered_pending: 0,
|
||||
covered_unassessed: 0,
|
||||
},
|
||||
},
|
||||
},
|
||||
start: '2026-10-14T00:00:00Z',
|
||||
end: '2026-10-15T00:00:00Z',
|
||||
keyword: '',
|
||||
|
||||
@@ -49,3 +49,26 @@ describe('quality evidence contracts', () => {
|
||||
).toHaveLength(1)
|
||||
})
|
||||
})
|
||||
|
||||
describe('confirmed quality rate contract', () => {
|
||||
it('retains raw counts, null rates, scope breakdown and policy version', () => {
|
||||
const result = parseQualitySummary(summaryFixture)
|
||||
expect(result.summary.rates.coverage.numerator).toBe(1)
|
||||
expect(result.summary.rates.coverage.denominator).toBe(2)
|
||||
expect(result.summary.rates.correctness.percent).toBe(100)
|
||||
})
|
||||
it('rejects invalid or inconsistent numerator, denominator, percent and category sums', () => {
|
||||
for (const change of [
|
||||
(r: any) => (r.coverage.numerator = 3),
|
||||
(r: any) => (r.coverage.denominator = 3),
|
||||
(r: any) => (r.coverage.percent = 99),
|
||||
(r: any) => (r.breakdown.stale = 0),
|
||||
(r: any) => (r.correctness.percent = null),
|
||||
(r: any) => (r.policy_version = 'unknown'),
|
||||
]) {
|
||||
const v = structuredClone(summaryFixture)
|
||||
change(v.summary.rates)
|
||||
expect(() => parseQualitySummary(v)).toThrow()
|
||||
}
|
||||
})
|
||||
})
|
||||
|
||||
@@ -45,6 +45,29 @@ export interface QualitySnapshot {
|
||||
created_at: string
|
||||
payload: { id: string; question: string; answer: string; status: string; references: unknown; media: unknown }
|
||||
}
|
||||
export interface QualityRate {
|
||||
numerator: number
|
||||
denominator: number
|
||||
percent: number | null
|
||||
}
|
||||
export interface QualityRates {
|
||||
policy_version: 'quality-rates-human-first-v1'
|
||||
coverage: QualityRate
|
||||
correctness: QualityRate
|
||||
breakdown: {
|
||||
covered: number
|
||||
uncovered: number
|
||||
partial: number
|
||||
pending: number
|
||||
unassessed: number
|
||||
excluded: number
|
||||
stale: number
|
||||
covered_correct: number
|
||||
covered_incorrect: number
|
||||
covered_pending: number
|
||||
covered_unassessed: number
|
||||
}
|
||||
}
|
||||
export interface QualitySummary {
|
||||
summary: {
|
||||
total: number
|
||||
@@ -53,6 +76,7 @@ export interface QualitySummary {
|
||||
stale: number
|
||||
pending_confirmation: number
|
||||
rule_uncovered: number
|
||||
rates: QualityRates
|
||||
}
|
||||
start: string
|
||||
end: string
|
||||
@@ -195,6 +219,56 @@ export function parseQAQuality(value: unknown): QAQuality {
|
||||
for (const key of Object.keys(expected) as (keyof QualityFields)[]) if (expected[key] !== effective[key]) return bad()
|
||||
return { input_hash, input_epoch, revision, state: state as QAQuality['state'], automatic, latest_review, effective }
|
||||
}
|
||||
function qualityRate(value: unknown): QualityRate {
|
||||
const v = obj(value),
|
||||
numerator = num(v.numerator),
|
||||
denominator = num(v.denominator)
|
||||
if (numerator > denominator) return bad()
|
||||
if (denominator === 0) {
|
||||
if (v.percent !== null) return bad()
|
||||
return { numerator, denominator, percent: null }
|
||||
}
|
||||
if (
|
||||
typeof v.percent !== 'number' ||
|
||||
!Number.isFinite(v.percent) ||
|
||||
v.percent < 0 ||
|
||||
v.percent > 100 ||
|
||||
Math.abs(v.percent - (100 * numerator) / denominator) > 1e-8
|
||||
)
|
||||
return bad()
|
||||
return { numerator, denominator, percent: v.percent }
|
||||
}
|
||||
function qualityRates(value: unknown, total: number, stale: number): QualityRates {
|
||||
const v = obj(value),
|
||||
r = obj(v.breakdown)
|
||||
const b = {
|
||||
covered: num(r.covered),
|
||||
uncovered: num(r.uncovered),
|
||||
partial: num(r.partial),
|
||||
pending: num(r.pending),
|
||||
unassessed: num(r.unassessed),
|
||||
excluded: num(r.excluded),
|
||||
stale: num(r.stale),
|
||||
covered_correct: num(r.covered_correct),
|
||||
covered_incorrect: num(r.covered_incorrect),
|
||||
covered_pending: num(r.covered_pending),
|
||||
covered_unassessed: num(r.covered_unassessed),
|
||||
}
|
||||
const coverage = qualityRate(v.coverage),
|
||||
correctness = qualityRate(v.correctness)
|
||||
if (
|
||||
v.policy_version !== 'quality-rates-human-first-v1' ||
|
||||
b.stale !== stale ||
|
||||
num(b.covered + b.uncovered + b.partial + b.pending + b.unassessed + b.excluded + b.stale) !== total ||
|
||||
num(b.covered_correct + b.covered_incorrect + b.covered_pending + b.covered_unassessed) !== b.covered ||
|
||||
coverage.numerator !== b.covered ||
|
||||
coverage.denominator !== b.covered + b.uncovered + b.partial ||
|
||||
correctness.numerator !== b.covered_correct ||
|
||||
correctness.denominator !== b.covered_correct + b.covered_incorrect
|
||||
)
|
||||
return bad()
|
||||
return { policy_version: 'quality-rates-human-first-v1', coverage, correctness, breakdown: b }
|
||||
}
|
||||
export function parseQualitySummary(value: unknown): QualitySummary {
|
||||
const v = obj(value),
|
||||
s = obj(v.summary),
|
||||
@@ -205,6 +279,7 @@ export function parseQualitySummary(value: unknown): QualitySummary {
|
||||
stale: num(s.stale),
|
||||
pending_confirmation: num(s.pending_confirmation),
|
||||
rule_uncovered: num(s.rule_uncovered),
|
||||
rates: qualityRates(s.rates, num(s.total), num(s.stale)),
|
||||
}
|
||||
const start = stamp(v.start),
|
||||
end = stamp(v.end)
|
||||
|
||||
Reference in New Issue
Block a user