65653e5c01
build / build (push) Successful in 1m14s
已覆盖只允许正确或错误,未覆盖可待确认,缺引用不再阻止正确结论。 复用统一事务写入与不可变审计重评旧待确认,保留原问答和人工结论。 验证:全仓 make verify、真实隔离 PostgreSQL 全包与 race 回归通过。
44 lines
2.5 KiB
Go
44 lines
2.5 KiB
Go
package qaopsai
|
|
|
|
import (
|
|
"testing"
|
|
)
|
|
|
|
func TestAssessmentRequiresApprovedLabelsAndVerifiableQuotedEvidence(t *testing.T) {
|
|
input := AssessmentInput{Question: "MG400 如何连接?", Answer: "fixture answer", NeededFields: []string{"product_category"}}
|
|
raw := `{"fields":{"product_category":{"value":"MG400","basis":"semantic_only","reason":"原问题明确给出型号","quotes":[{"source":"question","text":"MG400","reference_index":0}]}}}`
|
|
got, err := DecodeAssessment(raw, input)
|
|
if err != nil || got.Fields["product_category"].Value != "MG400" {
|
|
t.Fatalf("valid assessment rejected: %+v %v", got, err)
|
|
}
|
|
for _, bad := range []string{
|
|
`{"fields":{"product_category":{"value":"CR5","basis":"semantic_only","reason":"fixture","quotes":[]}}}`,
|
|
`{"fields":{"product_category":{"value":"MG400","basis":"semantic_only","reason":"fixture","quotes":[{"source":"answer","text":"fabricated quote","reference_index":0}]}}}`,
|
|
`{"fields":{"coverage":{"value":"是","basis":"semantic_only","reason":"fixture","quotes":[]}}}`,
|
|
`{"fields":{"product_category":{"value":"MG400","basis":"semantic_only","reason":"fixture","quotes":[]},"product_category":{"value":"未知","basis":"semantic_only","reason":"fixture","quotes":[]}}}`,
|
|
} {
|
|
if _, err := DecodeAssessment(bad, input); err == nil {
|
|
t.Fatal("invalid taxonomy, invented evidence, coverage output or duplicate key accepted")
|
|
}
|
|
}
|
|
}
|
|
|
|
func TestQuestionAnswerCorrectnessDoesNotRequireReferenceExcerpt(t *testing.T) {
|
|
input := AssessmentInput{Question: "2加2等于几?", Answer: "2加2等于4。", NeededFields: []string{"is_correct"}}
|
|
raw := `{"fields":{"is_correct":{"value":"是","basis":"semantic_only","reason":"实际回答给出的计算结果正确且回答了问题","quotes":[{"source":"answer","text":"2加2等于4。","reference_index":0}]}}}`
|
|
if _, err := DecodeAssessment(raw, input); err != nil {
|
|
t.Fatalf("question/answer assessment wrongly requires reference text: %v", err)
|
|
}
|
|
}
|
|
|
|
func TestOnlyUncoveredAssessmentMayRemainPending(t *testing.T) {
|
|
for _, coverage := range []string{"covered", "uncovered", ""} {
|
|
input := AssessmentInput{Question: "问题", Answer: "已有回答", Coverage: coverage, NeededFields: []string{"is_correct"}}
|
|
raw := `{"fields":{"is_correct":{"value":"待确认","basis":"insufficient_evidence","reason":"没有引用原文","quotes":[]}}}`
|
|
_, err := DecodeAssessment(raw, input)
|
|
if (err == nil) != (coverage == "uncovered") {
|
|
t.Fatalf("pending allowed for coverage=%q: %v", coverage, err)
|
|
}
|
|
}
|
|
}
|