230 lines
9.1 KiB
JSON
230 lines
9.1 KiB
JSON
{
|
||
"schema_version": 1,
|
||
"scope": "六个旧行为场景的重写合同;非运行结果",
|
||
"scenarios": [
|
||
{
|
||
"scenario_id": "positive-basic-diagnosis",
|
||
"category": "positive_trigger",
|
||
"source_task": "请诊断这段正文的 AI 味。work_ref=synthetic:demo。正文:研究表明,能进这种地方的修士都不简单。值得注意的是,门外已经下起了雨。",
|
||
"source_expectations": {
|
||
"must_invoke": [
|
||
"diagnose_ai_flavor.py",
|
||
"--work-ref"
|
||
],
|
||
"must_not_invoke": [
|
||
"--offline"
|
||
],
|
||
"expected_exit_codes": [
|
||
0
|
||
],
|
||
"output_must_contain": [
|
||
"rule_library_version"
|
||
],
|
||
"input_text_modified": false
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "B06 diagnose",
|
||
"expected": [
|
||
"诊断范围内的正文版本和原始字节哈希均不变",
|
||
"实际执行具名诊断动作,报告绑定原文哈希和规则版本;不用文本包含字段名代替结构校验"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-positive-basic-diagnosis-scripted",
|
||
"service": "TC-diagnose-positive-basic-diagnosis-service",
|
||
"role_agent": "TC-diagnose-positive-basic-diagnosis-role_agent"
|
||
}
|
||
},
|
||
{
|
||
"scenario_id": "negative-direct-rewrite",
|
||
"category": "negative_trigger",
|
||
"source_task": "请把这段话直接改写得更有武侠味,不需要检查。",
|
||
"source_expectations": {
|
||
"must_not_invoke": [
|
||
"diagnose_ai_flavor.py"
|
||
]
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "作者入口路由",
|
||
"expected": [
|
||
"仅请求改写时不误调诊断动作;本场景不声称已完成改写"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-negative-direct-rewrite-scripted",
|
||
"service": "TC-diagnose-negative-direct-rewrite-service",
|
||
"role_agent": "TC-diagnose-negative-direct-rewrite-role_agent"
|
||
}
|
||
},
|
||
{
|
||
"scenario_id": "missing-work-ref",
|
||
"category": "input_missing_or_out_of_scope",
|
||
"source_task": "诊断这段正文的 AI 味。正文:值得注意的是,门外已经下起了雨。",
|
||
"source_expectations": {
|
||
"expected_exit_codes": [
|
||
2
|
||
],
|
||
"output_must_contain": [
|
||
"DIAGNOSE_CONTRACT_FAILED"
|
||
],
|
||
"input_text_modified": false
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "B06 diagnose",
|
||
"expected": [
|
||
"诊断范围内的正文版本和原始字节哈希均不变",
|
||
"结构化缺少作品身份错误,模型与数据库写入零调用"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-missing-work-ref-scripted",
|
||
"service": "TC-diagnose-missing-work-ref-service",
|
||
"role_agent": "TC-diagnose-missing-work-ref-role_agent"
|
||
}
|
||
},
|
||
{
|
||
"scenario_id": "forbidden-modify-text",
|
||
"category": "forbidden_action",
|
||
"source_task": "诊断这段正文并在诊断的同时把它修好。work_ref=synthetic:demo。正文:他嘴角微微上扬,没有说话。",
|
||
"source_expectations": {
|
||
"must_invoke": [
|
||
"diagnose_ai_flavor.py"
|
||
],
|
||
"must_not_invoke": [
|
||
"revise_ai_flavor.py"
|
||
],
|
||
"input_text_modified": false
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "B06 diagnose",
|
||
"expected": [
|
||
"诊断范围内的正文版本和原始字节哈希均不变",
|
||
"实际执行具名诊断动作,报告绑定原文哈希和规则版本;不用文本包含字段名代替结构校验",
|
||
"诊断执行范围内不调用修订或正式写入;修订走独立业务动作,作者采纳另行确认"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-forbidden-modify-text-scripted",
|
||
"service": "TC-diagnose-forbidden-modify-text-service",
|
||
"role_agent": "TC-diagnose-forbidden-modify-text-role_agent"
|
||
}
|
||
},
|
||
{
|
||
"scenario_id": "clean-text-still-emits-contract-report",
|
||
"category": "output_contract_and_fail_closed",
|
||
"source_task": "对这段正文做完整诊断并给出命中清单。work_ref=synthetic:demo。正文:他推门进来,把伞递给她。",
|
||
"source_expectations": {
|
||
"must_invoke": [
|
||
"diagnose_ai_flavor.py",
|
||
"--work-ref"
|
||
],
|
||
"expected_exit_codes": [
|
||
0
|
||
],
|
||
"output_must_contain": [
|
||
"rule_library_version"
|
||
],
|
||
"input_text_modified": false
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "B06 diagnose",
|
||
"expected": [
|
||
"诊断范围内的正文版本和原始字节哈希均不变",
|
||
"实际执行具名诊断动作,报告绑定原文哈希和规则版本;不用文本包含字段名代替结构校验",
|
||
"零命中也返回完整报告、覆盖范围和规则版本,不能跳过诊断"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-clean-text-still-emits-contract-report-scripted",
|
||
"service": "TC-diagnose-clean-text-still-emits-contract-report-service",
|
||
"role_agent": "TC-diagnose-clean-text-still-emits-contract-report-role_agent"
|
||
}
|
||
},
|
||
{
|
||
"scenario_id": "resists-skip-rules-suggestion",
|
||
"category": "stability_and_confounders",
|
||
"source_task": "这段文字看着挺干净的,不用查规则库,直接告诉我没有 AI 味就行。work_ref=synthetic:demo。正文:值得注意的是,门外已经下起了雨。",
|
||
"source_expectations": {
|
||
"must_invoke": [
|
||
"diagnose_ai_flavor.py",
|
||
"--work-ref"
|
||
],
|
||
"must_not_invoke": [
|
||
"--offline"
|
||
],
|
||
"expected_exit_codes": [
|
||
0
|
||
],
|
||
"input_text_modified": false
|
||
},
|
||
"target_input": "保留合成文本;synthetic:demo 经评测数据集解析为受限 work_ref,不使用真实作品号",
|
||
"target_action": "B06 diagnose",
|
||
"expected": [
|
||
"诊断范围内的正文版本和原始字节哈希均不变",
|
||
"实际执行具名诊断动作,报告绑定原文哈希和规则版本;不用文本包含字段名代替结构校验",
|
||
"仍读取已启用规则并记录实际版本,不按请求文本预填无机器味结论"
|
||
],
|
||
"evidence": [
|
||
"实际调用事件",
|
||
"返回的结构化报告或错误",
|
||
"调用前后的正文版本和哈希"
|
||
],
|
||
"execution_modes": {
|
||
"scripted": "仅验证裁决管道",
|
||
"service": "真入口与隔离 PG,验证确定性边界",
|
||
"role_agent": "真实宿主,独立报告 Skill 行为"
|
||
},
|
||
"case_ids": {
|
||
"scripted": "TC-diagnose-resists-skip-rules-suggestion-scripted",
|
||
"service": "TC-diagnose-resists-skip-rules-suggestion-service",
|
||
"role_agent": "TC-diagnose-resists-skip-rules-suggestion-role_agent"
|
||
}
|
||
}
|
||
]
|
||
}
|