From 903fa94c009436832a928140c9d75c130a1c99b2 Mon Sep 17 00:00:00 2001 From: lyun Date: Mon, 2 Nov 2026 12:03:37 +0800 Subject: [PATCH] =?UTF-8?q?=E4=BF=AE=E5=A4=8D=EF=BC=9A=E5=AE=8C=E6=95=B4?= =?UTF-8?q?=E5=B1=95=E7=A4=BAAI=E5=8E=9F=E6=96=87=E5=B9=B6=E7=A9=BF?= =?UTF-8?q?=E6=8F=92=E5=9B=BE=E8=A1=A8=EF=BC=8C=E7=BB=9F=E4=B8=80=E8=BF=90?= =?UTF-8?q?=E8=A1=8C=E5=8F=8D=E9=A6=88=E4=B8=8E=E5=8F=98=E5=8C=96=E7=AC=A6?= =?UTF-8?q?=E5=8F=B7?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- DESIGN.md | 9 +- ai/nao-quality/RULES.md | 2 + ai/nao-quality/agent/prompts/system-native.md | 2 + ai/nao-quality/agent/prompts/system.md | 2 +- .../analysis/AnalysisProtocolService.java | 24 +- .../quality/analysis/AnalysisTurnService.java | 9 + .../AnalysisProtocolServiceMySqlTest.java | 21 +- .../analysis/AnalysisTurnSnapshotTest.java | 13 + .../analysis/NativeOriginalCoverageTest.java | 21 ++ scripts/analysis_visuals.py | 13 +- scripts/nao_native_run.py | 120 ++++--- verification/test_analysis_visuals.py | 12 +- verification/test_nao_original_body.py | 313 ++++++++++++++++++ verification/test_native_run.py | 128 ++++--- web/src/AnalysisConclusion.tsx | 2 +- web/src/AnalysisNarrative.tsx | 5 +- web/src/AnalysisOriginalContent.tsx | 13 + web/src/AnalysisProcess.tsx | 14 +- web/src/AnalysisReportNarrative.tsx | 5 +- web/src/ManagementPages.tsx | 74 +++-- web/src/analysis-chat.css | 10 +- web/src/analysisConclusionModel.ts | 6 +- web/src/analysisConversationState.ts | 13 + web/src/analysisOriginalBody.ts | 69 ++++ web/src/analysisSnapshot.ts | 13 +- web/src/analysisVisualsModel.ts | 21 +- web/verification/analysis-original.html | 1 + web/verification/analysis-original.tsx | 41 +++ .../analysis_live_status.test.mjs | 62 ++++ .../analysis_original_body.test.mjs | 44 +++ .../analysis_presentation_chain.test.mjs | 10 +- web/verification/analysis_reading.test.mjs | 10 +- web/verification/analysis_round1.test.mjs | 7 +- web/verification/analysis_submission.test.mjs | 20 +- web/verification/analysis_visuals.test.mjs | 59 ++++ 35 files changed, 1016 insertions(+), 172 deletions(-) create mode 100644 backend/src/test/java/cn/yunsheng/quality/analysis/NativeOriginalCoverageTest.java create mode 100644 verification/test_nao_original_body.py create mode 100644 web/src/AnalysisOriginalContent.tsx create mode 100644 web/src/analysisOriginalBody.ts create mode 100644 web/verification/analysis-original.html create mode 100644 web/verification/analysis-original.tsx create mode 100644 web/verification/analysis_live_status.test.mjs create mode 100644 web/verification/analysis_original_body.test.mjs diff --git a/DESIGN.md b/DESIGN.md index bf3049e..4aea7fb 100644 --- a/DESIGN.md +++ b/DESIGN.md @@ -1,6 +1,13 @@ # 质量数据分析平台设计规范 -更新:2026-10-30 +更新:2026-11-02 + +## 智能分析正文与运行反馈(2026-11-02) + +- 原生 AI 会话及报告以模型完整原文为正文,保留段落、表格和顺序,正式查询图表穿插在相关章节后;取消二次改写和统计兜底报告替换。历史兜底报告仅在保存的模型原文可恢复时恢复,原文缺失不伪造补齐。 +- 会话只显示一次最终回答,运行过程保留思考与工具步骤;运行中默认展开,可手动收起。发送立即显示等待状态和计时,收到服务器开始时间后显示运行计时和旋转图标。 +- 不在会话追加范围说明、进度句、错误或“部分完成”提示;真实失败、证据缺口和完成状态仍保存在后台,恢复及重试入口保留。 +- 合格率变化统一为本期减比较期:下降为负 pp,上升为正 pp。用于筛选和排序的降幅指标保持原有语义,转换仅作用于展示值;图表、数据表和导出一致。 ## 已确认方向 - 保持现有导航宽度、模块顺序、筛选承载、指标横条、表格及趋势位置。(用户确认) diff --git a/ai/nao-quality/RULES.md b/ai/nao-quality/RULES.md index adbc24d..5b42ea4 100644 --- a/ai/nao-quality/RULES.md +++ b/ai/nao-quality/RULES.md @@ -10,6 +10,8 @@ 比例转百分数乘100;两比例之差乘100才是“百分点”。例如0.96→0.95是下降1个百分点,不能先对原始比例差保留两位再换算。保留原始精度到最后展示一步。空值表示缺失/不适用,不等于零;新增/退出对象允许描述重量变化,但不补造其缺失期合格率。 +合格率变化统一为“本期-比较期”:下降显示 -X.XX pp,上升显示 +X.XX pp,持平为 0.00 pp;可另外标注方向。rateDrop/rate_drop 是比较期减本期的降幅排序量,不能直接当作有符号变化;展示其变化时取反并按正式单位换算,排序顺序不变。优先引用工具已提供的正式变化值,不能从舍入后的两期率重新计算。 + 重量统一以正式kg展示。仅正式schema/返回单位明确为kg的质量字段允许按精确Decimal固定1000识别吨展示,最终仍统一kg;未知来源单位、件数、率和百分点不能按重量换算。严格比较阈值保留原始精确边界,不能用舍入值改变“低于/超过”等关系。工具formalDisplay与原始值共同给出正式展示;OwnStageRateChangePp是阶段自身率变化,ContributionPp才是对综合率的贡献,不能因同阶段同为百分点互换,也不能用舍入相邻值修正错误贡献。 “汽车交通”是“新能源汽车磁钢”产业部的业务别名,统一用 filters.department=新能源汽车磁钢;不能用国内和海外的 group 多选集合替代产业部范围,两者原始记录覆盖未必相同。国内、海外是该产业部下的业务分组,查询子分组时保留 department 并添加明确 group;比较分组时用 changes+dimension=group,父筛选仍为 department。总体指全公司。规格对应 product。期间、范围、基期、发布版本要一致。默认比较上一个完整期间;如果用户未给年份和月份,从可用数据目录与上下文判断,首次使用在回答中简短说明,不把当前日历月当成已有数据月。 diff --git a/ai/nao-quality/agent/prompts/system-native.md b/ai/nao-quality/agent/prompts/system-native.md index 7032373..128decd 100644 --- a/ai/nao-quality/agent/prompts/system-native.md +++ b/ai/nao-quality/agent/prompts/system-native.md @@ -8,6 +8,8 @@ 优先使用工具formalDisplay的规范数值与单位,重量最终统一kg。阶段自身率变化与对综合率贡献按metricRole明确区分;不能把OwnStageRateChangePp当ContributionPp,也不能从舍入后的阶段变化凑总体等式。只有正式原单位明确kg的质量字段可按固定1000识别吨,数量和未知单位不换算,严格阈值不舍入。 +合格率变化统一为“本期-比较期”:下降显示 -X.XX pp,上升显示 +X.XX pp,持平为 0.00 pp;可另外标注方向。rateDrop/rate_drop 是比较期减本期的降幅排序量,不能直接当作有符号变化;展示其变化时取反并按正式单位换算,排序顺序不变。优先引用工具已提供的正式变化值,不能从舍入后的两期率重新计算。 + 当前已确认 filters、period、comparisonPeriod 是唯一范围权威。历史问答只帮助理解指代,不能恢复已取消条件。每轮使用同一个内部releaseId但不向用户展示内部ID、版本锁定或执行协议。修改用户范围需要下一轮确认;调查内只允许保持父条件、沿授权目录缩小对象。 原因分析须先取得整个已确认范围的正式本期/比较期基线,再决定是否缩小范围;不能用某个子分组或单个规格的率代替整个问题范围。首次compare/query保持_agentTask.filters原样,不额外添加子对象。之后的方向和深度由真实返回的线索决定。 diff --git a/ai/nao-quality/agent/prompts/system.md b/ai/nao-quality/agent/prompts/system.md index 8b48608..d46512f 100644 --- a/ai/nao-quality/agent/prompts/system.md +++ b/ai/nao-quality/agent/prompts/system.md @@ -37,7 +37,7 @@ releaseId、发布版本号、候选发布状态、Story ID/slug/version 仅作 整体变化先用简短摘要说清变化方向、主要阶段和优先核查线索,再展开各节。每节先写一到三句解释已确认的现象及实际含义,必要时附一张紧凑数据表,不只连续抄表、不在“重点关注”重复全部数字。阶段数学影响沿用 stageDecomposition 的服务端结果,不能把阶段自身率变化当作对综合率贡献,也不把数学拆解称现场生产根因。亮片报废下降而投入下降更多时,要明确解释合格率仍可能下降。 常规整体分析的重点规格优先结合报废增量、当前规模、率变化和分阶段有效投入确定核查优先级。单按合格率降幅返回的小批量、0率或阶段缺失规格,另列“极端率变化线索”,不能只因降幅大就称整体主要原因;两期分母未查清时不推断停产、退出或全批报废。用户明确只看降幅排名时保留真实排名与批量背景。不得自设未授权的最低批量阈值或忽略小批量的安全/质量异常。 -即使这些规格报废绝对重量较小,也不能据此断言其对总体综合率的影响“可忽略”或计算贡献;只能说明报废重量及投入规模,并把总体影响留待正式可加总口径核验。0率原因未核实前,不用“更可能是全批报废/覆盖问题”等概率判断,只写待核对的分母、批次和数据覆盖。没有已确认的批量阈值时,不给规格贴“批量有效/无代表性”的确定性标签,可具体写两期投入重量及其变化。合格率变化一律明确比较期→本期,并写“下降/上升X.XX个百分点”,避免只写带符号的pp。 +即使这些规格报废绝对重量较小,也不能据此断言其对总体综合率的影响“可忽略”或计算贡献;只能说明报废重量及投入规模,并把总体影响留待正式可加总口径核验。0率原因未核实前,不用“更可能是全批报废/覆盖问题”等概率判断,只写待核对的分母、批次和数据覆盖。没有已确认的批量阈值时,不给规格贴“批量有效/无代表性”的确定性标签,可具体写两期投入重量及其变化。合格率变化统一为“本期-比较期”:下降显示 -X.XX pp,上升显示 +X.XX pp,持平为 0.00 pp;可另外标注方向。rateDrop/rate_drop 是比较期减本期的降幅排序量,不能直接当作有符号变化;展示其变化时取反并按正式单位换算,排序顺序不变。优先引用工具已提供的正式变化值,不能从舍入后的两期率重新计算。 组织、规格和具体缺陷之间的关系只写实际交叉查询已经核对的联系;没有交叉证据时保持各自范围,不把两个独立排名拼成因果链。最后只列能执行的优先核查问题,分清已确认现象与待验证原因,不编造责任人、完成期限或闭环结果。所有图文继续复用同一轮证据;不要为了报告样式重复查询、重新计算核心指标或自动生成故事。 diff --git a/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisProtocolService.java b/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisProtocolService.java index 48fe0b0..0741f7c 100644 --- a/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisProtocolService.java +++ b/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisProtocolService.java @@ -415,7 +415,8 @@ public class AnalysisProtocolService { jdbc.update("UPDATE analysis_protocol_execution SET status='FAILED' WHERE execution_id=? AND status='RUNNING'",a.executionId()); jdbc.update("UPDATE analysis_protocol_session SET status='FAILED',version=version+1 WHERE session_id=?",a.sessionId()); jdbc.update("UPDATE analysis_protocol_attempt SET status='FAILED' WHERE attempt_id=?",id); - Map answer=Map.of("answer",reason,"results",List.of(),"releaseId",a.releaseId(),"engine",a.input().containsKey("controlledInspection")?"CONTROLLED_INSPECTION_PROTOCOL":"NAO","completion",Map.of("completed",false,"reason",reason)); + boolean nativeConversation=autonomousSession(a.sessionId())&&!a.input().containsKey("controlledInspection"); + Map answer=Map.of("answer",nativeConversation?"":reason,"results",List.of(),"releaseId",a.releaseId(),"engine",a.input().containsKey("controlledInspection")?"CONTROLLED_INSPECTION_PROTOCOL":"NAO","completion",Map.of("completed",false,"reason",reason)); jdbc.update("UPDATE analysis_turn SET status='FAILED',answer_json=CAST(? AS JSON),stop_reason=?,finished_at=CURRENT_TIMESTAMP(6) WHERE id=?",write(answer),reason,a.turnId()); inspection.updateAssociation(a,"FAILED"); event(id,"failed",Map.of("turnId",a.turnId(),"status","FAILED","result",answer,"stopReason",reason)); @@ -628,12 +629,18 @@ public class AnalysisProtocolService { if(Set.of("COMPLETED","PARTIAL").contains(status)&&!"report".equals(task.get("kind")) &&!hasProtocolEvidence(details)) { status="FAILED"; + String original=AnalysisProtocolExploration.nativeAgent(task)?Objects.toString(answer.get("answer"),""):null; answer.clear(); - answer.put("answer",noProofFailureReason(details)); + answer.put("answer",original!=null?original:noProofFailureReason(details)); answer.put("results",List.of()); answer.put("releaseId",a.releaseId()); answer.put("engine","NAO"); answer.put("stopReason","formal_query_evidence_missing"); + if(original!=null) { + answer.put("executionMode","native_agent"); + answer.put("answerQuality",Map.of("version",1,"dataComplete",false,"narrativeFactBinding","not_assessed", + "missingEvidence",List.of(noProofFailureReason(details)))); + } } else if(Set.of("COMPLETED","PARTIAL").contains(status)) { governance.requireBoundAction("QUERY","query.aggregate"); if(AnalysisProtocolInspection.KIND.equals(task.get("kind"))) { @@ -672,7 +679,7 @@ public class AnalysisProtocolService { answer.put("answerQuality",quality); } if(AnalysisProtocolExploration.nativeAgent(task)){for(String key:List.of("scopeCoverage","explorationWarnings","conclusionProtocolQueryIds","attemptProtocolQueryIds"))if(checked.containsKey(key))answer.put(key,checked.get(key));} - if(Boolean.TRUE.equals(task.get("verifySource"))||Boolean.TRUE.equals(task.get("fullDetails"))) { + if(!AnalysisProtocolExploration.nativeAgent(task)&&(Boolean.TRUE.equals(task.get("verifySource"))||Boolean.TRUE.equals(task.get("fullDetails")))) { for(String field:List.of("reportMarkdown","reportHtml","reportTitle"))answer.remove(field); String rawStatus=rawStatusText(checked,map(recordList)); if(planResolution(a,task,actor).scopes().stream().allMatch(AnalysisProtocolService::rawScope))answer.put("answer",rawStatus); @@ -680,7 +687,7 @@ public class AnalysisProtocolService { } if(!Boolean.TRUE.equals(checked.get("complete"))) { status="PARTIAL";answer.put("stopReason","evidence_incomplete"); - if(AnalysisProtocolExploration.nativeAgent(task))appendCoverageNotices(answer,list(checked.get("missingEvidence"))); + if(AnalysisProtocolExploration.nativeAgent(task))appendNativeCoverageDiagnostics(answer,list(checked.get("missingEvidence"))); } } else { validateMetricEvidence(a,task,details); @@ -801,6 +808,15 @@ public class AnalysisProtocolService { } return result; } + /** Native model prose is immutable; coverage gaps belong to the saved execution metadata. */ + static void appendNativeCoverageDiagnostics(Map answer,List missing) { + var gaps=new LinkedHashSet(); + var quality=new LinkedHashMap<>(map(answer.get("answerQuality"))); + for(Object raw:list(quality.get("missingEvidence")))if(raw instanceof String gap&&!gap.isBlank())gaps.add(gap); + for(Object raw:missing)if(raw instanceof String gap&&!gap.isBlank())gaps.add(gap); + if(gaps.isEmpty())return; + quality.putIfAbsent("version",1);quality.put("missingEvidence",List.copyOf(gaps));answer.put("answerQuality",quality); + } /** A server downgrade must remain visible even when the model omitted its own limitations. */ private void appendCoverageNotices(Map answer,List missing) { var gaps=missing.stream().filter(String.class::isInstance).map(String.class::cast).filter(s->!s.isBlank()).distinct().toList(); diff --git a/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisTurnService.java b/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisTurnService.java index 6513b63..4fbd69c 100644 --- a/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisTurnService.java +++ b/backend/src/main/java/cn/yunsheng/quality/analysis/AnalysisTurnService.java @@ -1626,6 +1626,15 @@ public class AnalysisTurnService { if(quality.get(field) instanceof Boolean value)safe.put(field,value); if(quality.get("missingEvidence") instanceof List gaps)safe.put("missingEvidence",gaps.stream() .filter(value->value instanceof String gap&&!gap.isBlank()&&gap.length()<=1000).distinct().limit(20).toList()); + if("not_assessed".equals(quality.get("narrativeFactBinding")))safe.put("narrativeFactBinding","not_assessed"); + for(String field:List.of("citedEvidenceRefs","unresolvedEvidenceRefs"))if(quality.get(field) instanceof List refs) { + var visible=refs.stream().filter(value->value instanceof String ref&&ref.matches("Q[A-F0-9]{8}") + &&(!field.equals("citedEvidenceRefs")||(answer.get("queryEvidenceRefs") instanceof List allowed?allowed:List.of()).contains(ref))).distinct().limit(100).toList(); + safe.put(field,visible); + if(field.equals("unresolvedEvidenceRefs"))safe.put("citationComplete",visible.isEmpty()); + } + if(quality.get("notices") instanceof List notices)safe.put("notices",notices.stream() + .filter(value->value instanceof String notice&&!notice.isBlank()&¬ice.length()<=1000).distinct().limit(20).toList()); answer.put("answerQuality",safe); } Object count = details.get("toolCount"); diff --git a/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisProtocolServiceMySqlTest.java b/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisProtocolServiceMySqlTest.java index b0924a6..0dfe451 100644 --- a/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisProtocolServiceMySqlTest.java +++ b/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisProtocolServiceMySqlTest.java @@ -1134,6 +1134,25 @@ class AnalysisProtocolServiceMySqlTest { assertEquals(40,jdbc.queryForObject("SELECT COUNT(*) FROM analysis_protocol_query",Integer.class)); } finally {pool.shutdownNow();} } + @Test void nativeWithoutFormalEvidenceKeepsOriginalBodyButCannotPublishMetricsOrSucceed() throws Exception { + enableAutonomous();String turn=newTurn("原生回答缺少正式证据");var a=claim(turn,false); + var task=AnalysisProtocolService.map(tx.execute(s->service.confirm(a.attemptId(),nativeTask(turn,Map.of()),actor)).get("task")); + String original="## 结论\n\n模型完整原文,不用平台错误句替换。"; + tx.executeWithoutResult(s->service.finish(a.attemptId(),"PARTIAL",Map.of("answer",original,"visuals",List.of(Map.of("id","unproved"))),Map.of("agentTask",task,"protocolEvidence",List.of()),actor)); + var saved=json.readValue(jdbc.queryForObject("SELECT answer_json FROM analysis_turn WHERE id=?",String.class,turn),Map.class); + assertEquals("FAILED",turnStatus(turn));assertEquals(original,saved.get("answer"));assertFalse(saved.containsKey("visuals")); + assertEquals("native_agent",saved.get("executionMode")); + assertEquals(false,AnalysisProtocolService.map(saved.get("answerQuality")).get("dataComplete")); + assertEquals("formal_query_evidence_missing",saved.get("stopReason")); + } + @Test void nativePlanningFailureStoresReasonWithoutReplacingAnswerWithSystemError() throws Exception { + enableAutonomous();String turn=newTurn("原生规划失败");var a=claim(turn,false); + tx.executeWithoutResult(s->service.fail(a.attemptId(),"本轮分析未完成,请重试")); + var saved=json.readValue(jdbc.queryForObject("SELECT answer_json FROM analysis_turn WHERE id=?",String.class,turn),Map.class); + assertEquals("FAILED",turnStatus(turn));assertEquals("",saved.get("answer")); + assertEquals("本轮分析未完成,请重试",jdbc.queryForObject("SELECT stop_reason FROM analysis_turn WHERE id=?",String.class,turn)); + assertEquals(false,AnalysisProtocolService.map(saved.get("completion")).get("completed")); + } @Test void autonomousReadBudgetAndFinishRequireRealRootEvidence() { enableAutonomous();String turn=newTurn("检查根任务证据");var a=claim(turn,false);var confirmed=AnalysisProtocolService.map(tx.execute(s->service.confirm(a.attemptId(),nativeTask(turn,Map.of()),actor)).get("task")); var registered=step(a,"read",Map.of("kind","base"),List.of());var scope=AnalysisProtocolService.map(registered.get("scope"));var p=scopeParams(scope);var proofs=new ArrayList>(); @@ -1273,7 +1292,7 @@ class AnalysisProtocolServiceMySqlTest { var result=answer();result.put("readingSummary",Map.of("version",1,"lead","已核实总体率","notices",List.of("已有提示"))); tx.executeWithoutResult(s->service.finish(a.attemptId(),"COMPLETED",result,Map.of("agentTask",confirmed,"protocolEvidence",proofs,"factsComplete",true,"evidenceCompletion",Map.of("serverCoverageSufficient",true)),actor)); assertEquals("PARTIAL",turnStatus(turn));var saved=AnalysisProtocolService.map(json.readValue(jdbc.queryForObject("SELECT answer_json FROM analysis_turn WHERE id=?",String.class,turn),Map.class)); - var notices=AnalysisProtocolService.list(AnalysisProtocolService.map(saved.get("readingSummary")).get("notices"));assertTrue(notices.contains("已有提示"));assertTrue(notices.containsAll(gaps));assertTrue(saved.get("answer").toString().contains("尚未解决:"));assertTrue(saved.get("answer").toString().contains("18X15X4+T")); + var notices=AnalysisProtocolService.list(AnalysisProtocolService.map(saved.get("answerQuality")).get("missingEvidence"));assertTrue(notices.containsAll(gaps));assertEquals(result.get("answer"),saved.get("answer"));assertFalse(saved.get("answer").toString().contains("尚未解决:")); } } @Test void rejectedRegistrationDoesNotPolluteLaterSuccessfulScopeCoverage() { diff --git a/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisTurnSnapshotTest.java b/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisTurnSnapshotTest.java index a5024e0..ac21dd1 100644 --- a/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisTurnSnapshotTest.java +++ b/backend/src/test/java/cn/yunsheng/quality/analysis/AnalysisTurnSnapshotTest.java @@ -37,6 +37,19 @@ class AnalysisTurnSnapshotTest { private static final String SESSION_ID = "66b51f0e-0252-4ef1-9aeb-58bd262293db"; private static final String QUERY_FILTERS = "SELECT field_key,black_column,bright_column FROM query_filter_field WHERE enabled=TRUE"; + @Test void originalBodyQualityMetadataSurvivesSavedSnapshotWithoutClaimingFactVerification() { + Harness harness=new Harness("valid"); + var quality=Map.of("version",1,"dataComplete",true,"narrativeComplete",true,"chartsComplete",false, + "narrativeFactBinding","not_assessed","citedEvidenceRefs",List.of("Q1234ABCD","Q00000000","bad"), + "unresolvedEvidenceRefs",List.of("QDEADBEEF","bad"),"notices",List.of("图表尚未齐全"),"missingEvidence",List.of("未知引用")); + var details=Map.of("answerQuality",quality,"queryEvidenceRefs",List.of("Q1234ABCD")); + var answer=harness.service.nativeAnswer(details,"完整AI正文",Map.of(),"release-1","PARTIAL","evidence_incomplete",TURN_ID,null); + var savedQuality=AnalysisProtocolService.map(answer.get("answerQuality")); + assertEquals("not_assessed",savedQuality.get("narrativeFactBinding"));assertEquals(false,savedQuality.get("citationComplete")); + assertEquals(List.of("Q1234ABCD"),savedQuality.get("citedEvidenceRefs"));assertEquals(List.of("QDEADBEEF"),savedQuality.get("unresolvedEvidenceRefs")); + harness.storedOverride=answer;var snapshot=harness.service.getSnapshot(TURN_ID,harness.submittedActor); + assertEquals(savedQuality,snapshot.get("answerQuality"));assertEquals("完整AI正文",snapshot.get("answer")); + } @Test void snapshotReadUsesFreshAuthorizationAndDoesNotRequeryBusinessMetrics() { Harness harness = new Harness("valid"); diff --git a/backend/src/test/java/cn/yunsheng/quality/analysis/NativeOriginalCoverageTest.java b/backend/src/test/java/cn/yunsheng/quality/analysis/NativeOriginalCoverageTest.java new file mode 100644 index 0000000..61d557b --- /dev/null +++ b/backend/src/test/java/cn/yunsheng/quality/analysis/NativeOriginalCoverageTest.java @@ -0,0 +1,21 @@ +package cn.yunsheng.quality.analysis; +import org.junit.jupiter.api.Test; +import java.util.*; +import static org.junit.jupiter.api.Assertions.*; +class NativeOriginalCoverageTest { + @Test void nativeCoverageKeepsModelProseByteForByteAndStoresGapsSeparately() { + String original="## 结论\n\n原始模型正文。\n\n| 变化 |\n| --- |\n| -0.30 pp |"; + var answer=new LinkedHashMap();answer.put("answer",original); + answer.put("answerQuality",Map.of("version",1,"dataComplete",true,"missingEvidence",List.of("已有缺口"))); + AnalysisProtocolService.appendNativeCoverageDiagnostics(answer,List.of("图表未齐","图表未齐", "")); + assertEquals(original,answer.get("answer")); + var quality=AnalysisProtocolService.map(answer.get("answerQuality")); + assertEquals(true,quality.get("dataComplete"));assertEquals(List.of("已有缺口","图表未齐"),quality.get("missingEvidence")); + assertFalse(answer.containsKey("readingSummary")); + } + @Test void noGapsDoNotCreateOrRewriteNarrativeMetadata() { + var answer=new LinkedHashMap();answer.put("answer","原文"); + AnalysisProtocolService.appendNativeCoverageDiagnostics(answer,List.of()); + assertEquals(Map.of("answer","原文"),answer); + } +} diff --git a/scripts/analysis_visuals.py b/scripts/analysis_visuals.py index 0cf4192..98d6206 100644 --- a/scripts/analysis_visuals.py +++ b/scripts/analysis_visuals.py @@ -963,7 +963,7 @@ def _change_chart(document: dict[str, Any], result: dict[str, Any], turn_id: str note = "黑片与亮片两阶段报废重量之和仅作分组描述,不代表独立投入分母或总体变化贡献。" elif order == "rate_drop": title = "合格率变化排名" if stage is None else "分组合格率变化排名" - metric, unit, label, field = "rate_drop_pp", "pp", "合格率下降(pp)", "rateDrop" + metric, unit, label, field = "rate_drop_pp", "pp", "合格率变化(pp)", "rateDrop" required_fields = (["black_good_weight", "black_total_weight", "bright_good_weight", "bright_total_weight"] if stage is None else [f"{stage}_good_weight", f"{stage}_total_weight"]) @@ -971,11 +971,14 @@ def _change_chart(document: dict[str, Any], result: dict[str, Any], turn_id: str raw, ok = _finite_number(row.get(field)) if not ok: return None, False - return (raw * 100 if raw is not None else None), True + points = -raw * 100 if raw is not None else None + if points is not None and not math.isfinite(points): + return None, False + return points, True - note = "正值表示合格率下降,负值表示改善;百分点取自正式分组率差。" + note = "正值表示本期合格率上升,负值表示下降;变化值为本期减比较期,排序仍沿用正式率降排名。" if stage is None: - note = "正值表示综合合格率下降,负值表示改善;按正式综合率比较,不作为生产根因。" + note = "正值表示本期综合合格率上升,负值表示下降;变化值为本期减比较期,排序仍沿用正式率降排名,不代表生产根因。" elif order == 'current_rate': title = '分组本期合格率排名' metric = (stage+'_rate') if stage else 'composite_rate' @@ -1023,7 +1026,7 @@ def _change_chart(document: dict[str, Any], result: dict[str, Any], turn_id: str title = title.replace("分组", "产品") if order == "rate_drop": title = title.replace("合格率变化排名", "合格率降幅线索") - note = "条形末端显示具体变化:正值为下降,负值为改善,单位是百分点。本图只按率降幅排序,未纳入投入规模;少量生产或投入骤变的规格不能仅凭此图认定为重点,需结合分阶段投入与报废增量核查。" + note = "条形末端显示本期减比较期的变化:负值为下降,正值为上升,单位是百分点。本图仍按率降幅排序,未纳入投入规模;少量生产或投入骤变的规格不能仅凭此图认定为重点,需结合分阶段投入与报废增量核查。" elif result["dimension"] == "group": title = title.replace("分组", "业务分组") else: diff --git a/scripts/nao_native_run.py b/scripts/nao_native_run.py index e4563f8..6b01809 100644 --- a/scripts/nao_native_run.py +++ b/scripts/nao_native_run.py @@ -2,6 +2,7 @@ from __future__ import annotations import copy +import re import time @@ -22,12 +23,17 @@ def exploration_notices(scopes, coverage): for kind,labels in grouped.items()] +def _answer_evidence_refs(answer): + """Return only explicit query references present in the upstream answer.""" + return list(dict.fromkeys(re.findall(r'\bQ[A-F0-9]{8}\b', answer))) + + def run_native_exploration(*, request, task, authority, actor_root, delegated, connection, manager, state, cancel_event, emit, prior_evidence_files, phase_timings=None): from nao_quality_runtime import _atomic_json, _authenticate, _sse_request, _read_json from nao_agent_task import bind_task - from nao_protocol_execution import _identity, _platform_visual_context, query_evidence_bindings - from nao_evidence_finalizer import evidence_summaries, synthesize_native + from nao_protocol_execution import _identity, query_evidence_bindings + from nao_evidence_finalizer import evidence_summaries from analysis_visual_evidence import protocol_visual_result from nao_runtime_policy import native_timeout from nao_native_exploration import _proofs @@ -112,6 +118,8 @@ def run_native_exploration(*, request, task, authority, actor_root, delegated, c and not missing and not progress.get('halted')) if native_status != 'COMPLETED' and not corrected_tool_errors and not controlled_no_progress: missing.append('自主调查未正常结束(' + str(native_reason or native_status) + '),已保留核实结果。') + elif native_status == 'COMPLETED' and not natural_finish: + missing.append('自主调查未收到有效的正常结束信号。') if not documents: missing.append('本轮未取得当前执行的可核验查询证据。') context['_protocolScopes'] = {'scopes': scopes, 'pending': snapshot.get('pending', []), 'gaps': missing} @@ -119,24 +127,7 @@ def run_native_exploration(*, request, task, authority, actor_root, delegated, c visuals = protocol_visual_result(documents, scopes, request['question'], request['turnId'], release_id=request['releaseId'], base_context=context) timings['visuals'] = round((time.monotonic() - started) * 1000) - context['_platformVisuals'] = _platform_visual_context(visuals) - emit({'type': 'progress', 'message': '正在核对结论与图表…', 'executionMode': 'native_agent'}) - started = time.monotonic() - final = synthesize_native(base_url=connection[0], api_key=connection[1], model=connection[2], - question=request['question'], evidence=documents, cancel_event=cancel_event, - context=context, native_answer=accumulator.answer) - timings['synthesis'] = round((time.monotonic() - started) * 1000) - _atomic_json(actor_root / 'private' / 'native-composition.json', { - 'turnId': request['turnId'], 'rawModelResponses': final.get('rawModelResponses', []), - 'compositionDiagnostics': final.get('compositionDiagnostics', {}), - 'bindingInput': final.get('_privateBindingInput'), - 'titleMetadata': {key: final.get(key) for key in - ('reportTitle', 'titleGeneratedByModel', 'titleValid', 'titleSource')}, - 'readingSummary': final.get('readingSummary')}) - full_facts = final.get('fullEvidenceFacts', []) - if full_facts: - _atomic_json(actor_root / 'private' / 'native-full-evidence-facts.json', full_facts) - missing.extend(final.get('missingEvidence', [])) + emit({'type': 'progress', 'message': '正在生成正式图表并整理证据状态…', 'executionMode': 'native_agent'}) visual_warnings = list(visuals.get('missingEvidence', [])) if task.get('visualPreference') == 'required': missing.extend(visual_warnings) @@ -144,41 +135,84 @@ def run_native_exploration(*, request, task, authority, actor_root, delegated, c if visual_warnings and not summary_warnings: summary_warnings = ['部分重点图表未能完整展示;已保留核实文字和表格。'] missing = list(dict.fromkeys(str(item) for item in missing if item)) - complete = bool(documents and final.get('evidenceSufficient') and not missing) - used_refs = set(final.get('usedEvidenceRefs', [])) - cited_ids = list(dict.fromkeys(p['queryId'] for doc in documents if doc.get('evidenceRef') in used_refs - for p in _proofs(doc.get('result', {})))) - if final.get('evidenceSufficient') and not cited_ids: - missing.append('解释未关联当前执行的实际查询证明。') - complete = False + answer = accumulator.answer if isinstance(accumulator.answer, str) else '' + original_refs = _answer_evidence_refs(answer) + complete_receipt_ids = {p.get('queryId') for p in receipts if isinstance(p, dict) + and p.get('executionId') == identity['executionId'] and isinstance(p.get('queryId'), str) + and p.get('queryId') in accepted_ids + and complete_scopes is not None and (p.get('nodeId'), p.get('scopeId')) in complete_scopes} + verified_document_ids = {} + for doc in documents: + proofs = _proofs(doc.get('result', {})) if isinstance(doc, dict) else [] + proof_ids = list(dict.fromkeys(p.get('queryId') for p in proofs + if isinstance(p, dict) and isinstance(p.get('queryId'), str))) + if (doc.get('evidenceRef') and proof_ids + and all(query_id in complete_receipt_ids for query_id in proof_ids)): + verified_document_ids[doc['evidenceRef']] = proof_ids + document_refs = set(verified_document_ids) + cited_refs = set(original_refs) & document_refs + cited_ids = list(dict.fromkeys(query_id for doc_ref in original_refs if doc_ref in cited_refs + for query_id in verified_document_ids[doc_ref])) + unresolved_refs = sorted(set(original_refs) - cited_refs) + if unresolved_refs: + missing.extend('原文引用未关联本轮完整scope正式查询证据:' + ref + '。' for ref in unresolved_refs) + background_ids = list(dict.fromkeys(query_id for proof_ids in verified_document_ids.values() + for query_id in proof_ids)) if not original_refs else [] + if not original_refs and not background_ids and documents: + missing.append('本轮没有可关联的完整scope正式查询证明。') + conclusion_ids = cited_ids if original_refs else background_ids + if len(conclusion_ids) > 40: + missing.append('完整查询证据引用超过平台允许的40项上限。') + conclusion_ids = conclusion_ids[:40] + if not answer.strip(): + missing.append('自主分析未生成正文。') + missing = list(dict.fromkeys(str(item) for item in missing if item)) + data_complete = bool(documents and not snapshot.get('rootGoalGaps') + and (complete_scopes is not None or not snapshot.get('gaps')) + and bool(conclusion_ids) and not unresolved_refs + and len(cited_ids if original_refs else background_ids) <= 40) + narrative_complete = bool(answer.strip() and natural_finish + and (native_status == 'COMPLETED' or corrected_tool_errors)) + charts_complete = not visual_warnings + complete = bool(data_complete and narrative_complete and not missing) if warnings or visual_warnings: _atomic_json(actor_root / 'private' / 'native-warning-diagnostics.json', { 'turnId': request['turnId'], 'explorationWarnings': warnings, 'visualWarnings': visual_warnings}) - if summary_warnings: - if isinstance(final.get('readingSummary'), dict): - final['readingSummary']['notices'] = list(dict.fromkeys([ - *final['readingSummary'].get('notices', []), *summary_warnings])) - else: - final['answer'] += '\n\n' + '\n'.join(dict.fromkeys(summary_warnings)) - if missing: - final['answer'] += '\n\n尚未解决:' + ';'.join(missing) - if isinstance(final.get('readingSummary'), dict): - final['readingSummary']['notices'] = list(dict.fromkeys([ - *final['readingSummary'].get('notices', []), *missing])) + _atomic_json(actor_root / 'private' / 'native-original-body-diagnostics.json', { + 'turnId': request['turnId'], 'executionId': identity['executionId'], + 'nativeStatus': native_status, 'nativeReason': native_reason, + 'finishReasons': list(accumulator.finish_reasons), + 'originalBodyCharacters': len(answer), 'originalBodyNonempty': bool(answer.strip()), + 'narrativeComplete': narrative_complete, 'dataComplete': data_complete, + 'narrativeFactBinding': 'not_assessed', 'citedEvidenceRefs': sorted(cited_refs), + 'unresolvedEvidenceRefs': unresolved_refs, 'missingEvidence': missing, + 'visualWarnings': visual_warnings, 'notices': list(dict.fromkeys(summary_warnings))}) result = {'type': 'result', 'engine': 'NAO', 'status': 'COMPLETED' if complete else 'PARTIAL', - 'answer': final['answer'], 'readingSummary': final.get('readingSummary'), - 'reportTitle': final.get('reportTitle'), 'agentTask': task, 'protocol': identity, + 'answer': answer, 'readingSummary': None, + 'reportTitle': None, 'agentTask': task, 'protocol': identity, 'executionMode': 'native_agent', 'completionMethod': 'native_autonomous_analysis', - 'stopReason': 'evidence_finalized' if complete else 'evidence_incomplete', 'finishReason': 'stop', + 'stopReason': 'evidence_finalized' if complete else 'evidence_incomplete', + 'finishReason': ('stop' if narrative_complete else + str(accumulator.finish_reasons[-1]) if natural_finish and accumulator.finish_reasons else + 'incomplete' if native_status == 'COMPLETED' else str(native_reason or native_status)), 'nativeStopReason': native_reason, 'nativeStopSucceeded': stop_succeeded, 'nativeFinishReason': 'stop' if natural_finish else native_reason, - 'conclusionProtocolQueryIds': cited_ids, + 'conclusionProtocolQueryIds': conclusion_ids, + 'answerQuality': {'version': 1, 'dataComplete': data_complete, + 'narrativeComplete': narrative_complete, + 'chartsComplete': charts_complete, + 'narrativeFactBinding': 'not_assessed', + 'citedEvidenceRefs': sorted(cited_refs), + 'backgroundEvidenceQueryIds': background_ids, + 'unresolvedEvidenceRefs': unresolved_refs, + 'missingEvidence': missing, + 'notices': list(dict.fromkeys(summary_warnings))}, 'actualModel': model, 'phaseTimingsMs': timings, 'protocolEvidence': receipts, 'queryEvidenceRefs': [doc['evidenceRef'] for doc in documents], 'queryEvidenceBindings': query_evidence_bindings(documents, receipts), 'toolCalls': accumulator.tool_calls, 'toolCount': len(accumulator.tool_calls), - 'timeline': accumulator.timeline, 'factsComplete': bool(documents and not missing), + 'timeline': accumulator.timeline, 'factsComplete': None, 'evidenceCompletion': {'serverCoverageSufficient': not snapshot.get('rootGoalGaps'), 'missingEvidence': missing}, 'agentContext': {key: copy.deepcopy(task[key]) for key in diff --git a/verification/test_analysis_visuals.py b/verification/test_analysis_visuals.py index 965e85b..3986d16 100644 --- a/verification/test_analysis_visuals.py +++ b/verification/test_analysis_visuals.py @@ -316,14 +316,18 @@ class AnalysisVisualsTest(unittest.TestCase): def test_rate_drop_is_signed_and_converted_to_percentage_points(self): result = changes_result(sort="rate_drop", rows=[ - {"rank": 1, "key": "p-1", "label": "规格一", "rateDrop": -0.025}, - {"rank": 2, "key": "p-2", "label": "规格二", "rateDrop": 0.04}, + {"rank": 1, "key": "p-1", "label": "规格一", "rateDrop": 0.025}, + {"rank": 2, "key": "p-2", "label": "规格二", "rateDrop": -0.04}, ]) chart = build_visuals([evidence(result)], "画图比较产品合格率降幅", "turn-03")["visuals"][0] self.assertEqual(chart["metric"], "rate_drop_pp") self.assertEqual(chart["unit"], "pp") self.assertEqual(chart["series"][0]["values"], [-2.5, 4.0]) + self.assertEqual([row["key"] for row in result["rows"]], ["p-1", "p-2"]) + self.assertEqual([item["key"] for item in chart["categories"]], ["p-1", "p-2"]) self.assertIn("率降幅线索", chart["title"]) + self.assertIn("本期减比较期", chart["note"]) + self.assertIn("仍按率降幅排序", chart["note"]) self.assertIn("未纳入投入规模", chart["note"]) def test_broad_product_focus_omits_rate_only_ranking_chart(self): @@ -377,7 +381,9 @@ class AnalysisVisualsTest(unittest.TestCase): self.assertEqual(chart["requiredFields"], [ "black_good_weight", "black_total_weight", "bright_good_weight", "bright_total_weight" ]) - self.assertEqual(chart["series"][0]["values"], [2.0, -1.0]) + self.assertEqual(chart["series"][0]["values"], [-2.0, 1.0]) + self.assertIn("负值为下降,正值为上升", chart["note"]) + self.assertIn("本期减比较期", chart["note"]) def test_quality_top_uses_ranked_weights_and_never_share(self): result = { diff --git a/verification/test_nao_original_body.py b/verification/test_nao_original_body.py new file mode 100644 index 0000000..65de9f6 --- /dev/null +++ b/verification/test_nao_original_body.py @@ -0,0 +1,313 @@ +"""Regression tests for the isolated native-agent original-body runtime copy.""" +import copy +import importlib.util +import json +import os +import sys +import tempfile +import threading +import unittest +from pathlib import Path +from types import ModuleType, SimpleNamespace +from unittest.mock import Mock, patch + + +PROJECT = Path(__file__).resolve().parents[1] +RUNTIME_COPY = Path(os.environ.get('NAO_ORIGINAL_BODY_PATH', + str(PROJECT / 'scripts/nao_native_run.py'))).expanduser().resolve() +SPEC = importlib.util.spec_from_file_location('nao_native_run_original_body', RUNTIME_COPY) +NATIVE_RUN = importlib.util.module_from_spec(SPEC) +sys.modules[SPEC.name] = NATIVE_RUN +SPEC.loader.exec_module(NATIVE_RUN) + + +IDENTITY = {'attemptId': 'attempt-current', 'sessionId': 'session-current', + 'turnId': 'turn-current', 'revision': 3, 'executionId': 'execution-current'} + + +def proof(query_id='query-current'): + return {**IDENTITY, 'queryId': query_id, 'nodeId': 'root', 'scopeId': 'scope-root', + 'responseHash': 'a' * 64} + + +def document(ref='Q00000001', evidence_proof=None): + return {'evidenceRef': ref, 'result': {'protocolEvidence': evidence_proof or proof()}} + + +def task(): + return {'version': 1, 'goal': '解释下降并根据实际结果调查', 'question': '本期下降原因', + 'executionMode': 'native_agent', 'kind': 'analysis', + 'period': {'year': 2026, 'month': 9}, 'comparisonPeriod': {'year': 2026, 'month': 8}, + 'filters': {'department': '当前产业部'}, 'fields': ['rate'], 'stages': [], + 'dimensions': [], 'dimensionStages': {}, 'visualPreference': 'auto', + 'comparisonRequired': True, 'trendRequired': False, + 'verifySource': False, 'fullDetails': False} + + +class NativeOriginalBodyTests(unittest.TestCase): + def setUp(self): + self.temp = tempfile.TemporaryDirectory() + self.addCleanup(self.temp.cleanup) + self.actor_root = Path(self.temp.name) + (self.actor_root / 'private').mkdir() + (self.actor_root / 'evidence').mkdir() + self.cancel_event = threading.Event() + self.task = task() + self.request = {'sessionId': IDENTITY['sessionId'], 'turnId': IDENTITY['turnId'], + 'actorId': 'actor-current', 'releaseId': 'release-current', + 'question': self.task['question'], + 'context': {'_serverHistory': [], '_serverQuestions': [], '_protocol': {}}} + self.current_proof = proof() + self.documents = [document(evidence_proof=self.current_proof)] + self.snapshot = {'scopes': [{'nodeId': 'root', 'scopeId': 'scope-root', 'task': self.task}], + 'queries': [{'proof': self.current_proof}], + 'scopeCoverage': [{'nodeId': 'root', 'scopeId': 'scope-root', 'complete': True}], + 'rootGoalGaps': [], 'gaps': [], 'pending': [], 'explorationWarnings': []} + self.accumulator = SimpleNamespace(answer='原始回答 [Q00000001]', finish_reasons=['stop'], + tool_calls=[{'tool': 'quality_data', 'status': 'completed', 'input': {'mode': 'compare'}}], + timeline=[{'kind': 'assistant', 'text': '原始回答'}]) + self.native_status, self.native_reason, self.stop_succeeded = 'COMPLETED', 'stop', None + self.events = [] + self.manager = Mock() + self.manager.ensure_instance.return_value = (None, None, 'fixture-model', 'http://native.invalid') + self.delegated = Mock() + self.delegated.call.side_effect = lambda *args, **kwargs: copy.deepcopy(self.snapshot) + self.synthetic_writer = Mock(side_effect=AssertionError('native original-body path must not synthesize text')) + + quality_runtime = ModuleType('nao_quality_runtime') + quality_runtime._atomic_json = self.atomic_json + quality_runtime._authenticate = Mock(return_value=object()) + quality_runtime._sse_request = self.native + quality_runtime._read_json = self.read_json + agent_task = ModuleType('nao_agent_task') + agent_task.bind_task = lambda seed, current_task: {**copy.deepcopy(current_task), **copy.deepcopy(seed)} + protocol_execution = ModuleType('nao_protocol_execution') + protocol_execution._identity = lambda authority: copy.deepcopy(IDENTITY) + protocol_execution._platform_visual_context = lambda visuals: {'visualState': visuals.get('visualState')} + protocol_execution.query_evidence_bindings = lambda docs, receipts: [ + {'evidenceRef': item['evidenceRef'], 'proofs': item['result'].get('protocolEvidence')} + for item in docs] + evidence_finalizer = ModuleType('nao_evidence_finalizer') + evidence_finalizer.evidence_summaries = lambda *args, **kwargs: copy.deepcopy(self.documents) + evidence_finalizer.synthesize_native = self.synthetic_writer + visual_evidence = ModuleType('analysis_visual_evidence') + visual_evidence.protocol_visual_result = self.visual_result + runtime_policy = ModuleType('nao_runtime_policy') + runtime_policy.native_timeout = lambda: 30 + native_exploration = ModuleType('nao_native_exploration') + native_exploration._proofs = lambda result: ([result['protocolEvidence']] + if isinstance(result, dict) and isinstance(result.get('protocolEvidence'), dict) else []) + self.modules = {'nao_quality_runtime': quality_runtime, 'nao_agent_task': agent_task, + 'nao_protocol_execution': protocol_execution, + 'nao_evidence_finalizer': evidence_finalizer, + 'analysis_visual_evidence': visual_evidence, + 'nao_runtime_policy': runtime_policy, + 'nao_native_exploration': native_exploration} + self.module_patch = patch.dict(sys.modules, self.modules) + self.module_patch.start() + self.addCleanup(self.module_patch.stop) + + def atomic_json(self, path, payload): + path = Path(path) + path.parent.mkdir(parents=True, exist_ok=True) + path.write_text(json.dumps(payload, ensure_ascii=False), encoding='utf-8') + + @staticmethod + def read_json(path): + return json.loads(Path(path).read_text(encoding='utf-8')) + + def native(self, **kwargs): + return self.accumulator, self.native_status, self.native_reason, self.stop_succeeded + + def visual_result(self, documents, scopes, question, turn_id, **kwargs): + return {'visualState': 'ready', 'visuals': [{'id': 'formal-query-chart', + 'evidenceRefs': [item['evidenceRef'] for item in documents]}], 'missingEvidence': []} + + def invoke(self): + return NATIVE_RUN.run_native_exploration(request=copy.deepcopy(self.request), + task=copy.deepcopy(self.task), authority=copy.deepcopy(IDENTITY), + actor_root=self.actor_root, delegated=self.delegated, + connection=('https://model.invalid/v1', 'synthetic-fixture-key', 'fixture-model'), + manager=self.manager, state={}, cancel_event=self.cancel_event, + emit=self.events.append, prior_evidence_files={'prior-evidence.json'}, + phase_timings={'planningAndReview': 10, 'taskConfirmation': 20}) + + def test_persists_the_complete_original_body_and_keeps_formal_charts(self): + body = '原生回答第一行\n' + ('原生正文' * 1800) + '\n[Q00000001]' + self.accumulator.answer = body + self.accumulator.timeline = [{'kind': 'assistant', 'text': 'x' * 8000}] + + result = self.invoke() + + self.assertEqual(result['status'], 'COMPLETED') + self.assertEqual(result['answer'], body) + self.assertIsNone(result['readingSummary']) + self.assertEqual(result['visuals'][0]['id'], 'formal-query-chart') + self.assertEqual(result['queryEvidenceRefs'], ['Q00000001']) + self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current']) + self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed') + self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], []) + self.assertIsNone(result['factsComplete']) + self.assertEqual(self.read_json(self.actor_root / 'private/native-draft.json')['answer'], body) + self.assertEqual(self.read_json(self.actor_root / 'private/native-completion.json')['answer'], body) + self.assertEqual(self.read_json(self.actor_root / 'private/native-original-body-diagnostics.json') + ['originalBodyCharacters'], len(body)) + self.synthetic_writer.assert_not_called() + + def test_empty_native_body_stays_empty_and_cannot_be_completed(self): + self.accumulator.answer = '' + + result = self.invoke() + + self.assertEqual(result['answer'], '') + self.assertEqual(result['status'], 'PARTIAL') + self.assertFalse(result['answerQuality']['narrativeComplete']) + self.assertIn('自主分析未生成正文。', result['answerQuality']['missingEvidence']) + self.synthetic_writer.assert_not_called() + + def test_missing_formal_evidence_is_metadata_and_does_not_rewrite_the_body(self): + self.documents = [] + body = '原始正文照常保留。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['answer'], body) + self.assertEqual(result['status'], 'PARTIAL') + self.assertFalse(result['answerQuality']['dataComplete']) + self.assertIn('本轮未取得当前执行的可核验查询证据。', result['answerQuality']['missingEvidence']) + self.assertNotIn('尚未解决', result['answer']) + self.synthetic_writer.assert_not_called() + + def test_partial_or_interrupted_generation_is_not_rewritten_or_marked_complete(self): + self.native_status, self.native_reason = 'PARTIAL', 'length' + self.accumulator.finish_reasons = ['length'] + body = '模型已输出的局限与正文。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['answer'], body) + self.assertEqual(result['status'], 'PARTIAL') + self.assertFalse(result['answerQuality']['narrativeComplete']) + self.assertEqual(result['finishReason'], 'length') + self.synthetic_writer.assert_not_called() + + def test_completed_status_without_natural_stop_remains_partial(self): + body = '原生回答保留,结束信号不足。' + self.accumulator.answer = body + cases = ( + ('missing_stop_finish_reason', [], None), + ('stop_control_failed', ['stop'], False), + ) + for label, finish_reasons, stop_succeeded in cases: + with self.subTest(label=label): + self.accumulator.finish_reasons = finish_reasons + self.stop_succeeded = stop_succeeded + + result = self.invoke() + + self.assertEqual(result['answer'], body) + self.assertEqual(result['status'], 'PARTIAL') + self.assertFalse(result['answerQuality']['narrativeComplete']) + self.assertEqual(result['finishReason'], 'incomplete') + self.assertIn('自主调查未收到有效的正常结束信号。', + result['answerQuality']['missingEvidence']) + self.synthetic_writer.assert_not_called() + + def test_recovered_tool_error_keeps_natural_stop_completion_without_fact_binding(self): + self.native_status, self.native_reason = 'PARTIAL', 'mcp_error' + body = '原生回答保留自己的局限。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['status'], 'COMPLETED') + self.assertEqual(result['answer'], body) + self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current']) + self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], ['query-current']) + self.assertTrue(result['answerQuality']['narrativeComplete']) + self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed') + self.assertIsNone(result['factsComplete']) + self.synthetic_writer.assert_not_called() + + def test_no_explicit_refs_cannot_use_docs_without_verified_complete_scope_receipts(self): + self.snapshot['scopeCoverage'] = None + body = '正常原生正文不写证据编号。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['answer'], body) + self.assertEqual(result['status'], 'PARTIAL') + self.assertEqual(result['conclusionProtocolQueryIds'], []) + self.assertEqual(result['answerQuality']['backgroundEvidenceQueryIds'], []) + self.assertFalse(result['answerQuality']['dataComplete']) + self.assertIn('本轮没有可关联的完整scope正式查询证明。', + result['answerQuality']['missingEvidence']) + self.synthetic_writer.assert_not_called() + + def test_unbound_original_citation_is_a_gap_without_rewriting_the_body(self): + body = '原生回答里保留未绑定的引用。[QABCDEF99]' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['status'], 'PARTIAL') + self.assertEqual(result['answer'], body) + self.assertFalse(result['answerQuality']['dataComplete']) + self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['QABCDEF99']) + self.assertIn('原文引用未关联本轮完整scope正式查询证据:QABCDEF99。', + result['answerQuality']['missingEvidence']) + self.assertNotIn('QABCDEF99', result['conclusionProtocolQueryIds']) + self.synthetic_writer.assert_not_called() + + def test_explicit_hexadecimal_query_reference_maps_only_its_receipt(self): + hex_proof = proof('query-hex') + self.snapshot['queries'].append({'proof': hex_proof}) + self.documents.append(document('QABCDEF12', hex_proof)) + body = '原文引用正式查询[QABCDEF12]。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['status'], 'COMPLETED') + self.assertEqual(result['answer'], body) + self.assertEqual(result['answerQuality']['citedEvidenceRefs'], ['QABCDEF12']) + self.assertEqual(result['conclusionProtocolQueryIds'], ['query-hex']) + self.synthetic_writer.assert_not_called() + + def test_cancelled_turn_remains_cancelled_and_does_not_read_or_synthesize(self): + for cancellation in ('event', 'native_status'): + with self.subTest(cancellation=cancellation): + self.cancel_event.clear() + self.native_status, self.native_reason = 'COMPLETED', 'stop' + if cancellation == 'event': + self.cancel_event.set() + else: + self.native_status = 'CANCELLED' + self.accumulator.answer = '中断前的草稿' + + result = self.invoke() + + self.assertEqual(result['status'], 'CANCELLED') + self.assertEqual(result['answer'], '') + self.delegated.call.assert_not_called() + self.synthetic_writer.assert_not_called() + + def test_server_coverage_gap_keeps_body_and_prevents_completed_status(self): + self.snapshot['rootGoalGaps'] = ['亮片必答阶段尚无完整两期依据。'] + body = '原生结论及其局限。' + self.accumulator.answer = body + + result = self.invoke() + + self.assertEqual(result['answer'], body) + self.assertEqual(result['status'], 'PARTIAL') + self.assertFalse(result['answerQuality']['dataComplete']) + self.assertIn('亮片必答阶段尚无完整两期依据。', result['answerQuality']['missingEvidence']) + self.synthetic_writer.assert_not_called() + + +if __name__ == '__main__': + unittest.main() diff --git a/verification/test_native_run.py b/verification/test_native_run.py index fabb6a6..06203cb 100644 --- a/verification/test_native_run.py +++ b/verification/test_native_run.py @@ -1,6 +1,8 @@ """Exercise the native runner with isolated files and mocked upstream/bridge calls.""" import copy +import importlib.util import json +import os import sys import tempfile import threading @@ -10,8 +12,15 @@ from pathlib import Path from types import SimpleNamespace from unittest.mock import Mock, patch -sys.path[:0] = [str(Path(__file__).resolve().parents[1] / 'scripts')] -from nao_native_run import run_native_exploration +PROJECT = Path(__file__).resolve().parents[1] +sys.path[:0] = [str(PROJECT / 'scripts')] +RUNTIME_PATH = Path(os.environ.get('NAO_ORIGINAL_BODY_PATH', + str(PROJECT / 'scripts/nao_native_run.py'))).expanduser().resolve() +RUNTIME_SPEC = importlib.util.spec_from_file_location('nao_native_run_integration', RUNTIME_PATH) +RUNTIME_MODULE = importlib.util.module_from_spec(RUNTIME_SPEC) +sys.modules[RUNTIME_SPEC.name] = RUNTIME_MODULE +RUNTIME_SPEC.loader.exec_module(RUNTIME_MODULE) +run_native_exploration = RUNTIME_MODULE.run_native_exploration from nao_quality_runtime import _atomic_json @@ -54,31 +63,25 @@ class NativeRunTests(unittest.TestCase): 'missingEvidence': ['节点bright-detail范围基础范围缺少bright来源具体缺陷重量图。']} result = self.invoke() self.assertEqual(result['status'], 'COMPLETED') - self.assertEqual(result['readingSummary']['notices'], + self.assertEqual(result['answerQuality']['notices'], ['亮片细分缺陷尚未完整核实:当前范围;未用于已核实结论。']) self.assertNotIn('调查局限', result['answer']) self.assertNotIn('bright', result['answer']) stored = json.loads((self.actor_root / 'private/native-warning-diagnostics.json').read_text()) self.assertIn('bright', stored['visualWarnings'][0]) - def test_full_facts_remain_private_instead_of_being_appended_to_the_business_answer(self): - self.final['fullEvidenceFacts'] = [{'text': '后台保留的完整统计事实', 'evidenceRef': 'Q00000001'}] + def test_original_answer_is_not_replaced_by_a_generated_report(self): + self.accumulator.answer = '原生回答原样保留,局限也照原文保留。' result = self.invoke() - self.assertNotIn('完整统计依据', result['answer']) - self.assertNotIn('后台保留的完整统计事实', result['answer']) - stored = json.loads((self.actor_root / 'private/native-full-evidence-facts.json').read_text()) - self.assertEqual(stored, self.final['fullEvidenceFacts']) + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertIsNone(result['readingSummary']) + self.synthesis.assert_not_called() - def test_issued_binding_catalog_remains_private_for_exact_replay(self): - issued = {'bindingInventory': {'N17': {'factId': 'F2', 'value': '493.9966', 'unit': 'kg'}}, - 'facts': [{'id': 'F2', 'rawValue': '493.99660000'}]} - self.final['_privateBindingInput'] = issued + def test_no_second_composition_artifact_is_written(self): result = self.invoke() - stored = json.loads((self.actor_root / 'private/native-composition.json').read_text()) - self.assertEqual(stored['bindingInput'], issued) - self.assertNotIn('_privateBindingInput', result) - self.assertNotIn('bindingInput', result) - self.assertNotIn('bindingInventory', json.dumps(self.events, ensure_ascii=False)) + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertFalse((self.actor_root / 'private/native-composition.json').exists()) + self.synthesis.assert_not_called() def setUp(self): self.folder = tempfile.TemporaryDirectory() @@ -105,11 +108,6 @@ class NativeRunTests(unittest.TestCase): 'scopeCoverage': [{'nodeId': 'root', 'scopeId': 'scope-root', 'complete': True}, {'nodeId': 'group', 'scopeId': 'scope-group', 'complete': True}], 'rootGoalGaps': [], 'gaps': [], 'pending': [], 'explorationWarnings': []} - self.final = {'answer': '按已核实结果解释[Q00000001]', 'reportTitle': '质量数据分析', - 'readingSummary': {'version': 1, 'lead': '已核实解释[Q00000001]', - 'evidence': [], 'notices': [], 'evidenceRefs': ['Q00000001']}, - 'usedEvidenceRefs': ['Q00000001'], 'evidenceSufficient': True, - 'missingEvidence': [], 'rawModelResponses': [], 'compositionDiagnostics': {}} self.accumulator = SimpleNamespace(answer='原生草稿', finish_reasons=['stop'], tool_calls=[{'tool': 'quality_data', 'status': 'failed', 'input': {'mode': 'compare'}}, {'tool': 'quality_data', 'status': 'completed', 'input': {'mode': 'compare'}}], @@ -127,7 +125,7 @@ class NativeRunTests(unittest.TestCase): self.read_evidence = stack.enter_context(patch('nao_evidence_finalizer.evidence_summaries', side_effect=lambda *args, **kwargs: copy.deepcopy(self.documents))) self.synthesis = stack.enter_context(patch('nao_evidence_finalizer.synthesize_native', - side_effect=lambda **kwargs: copy.deepcopy(self.final))) + side_effect=AssertionError('native-agent original body must not be synthesized'))) self.visuals = stack.enter_context(patch('analysis_visual_evidence.protocol_visual_result', return_value={'visualState': 'not_requested', 'visuals': [], 'missingEvidence': []})) stack.enter_context(patch('nao_runtime_policy.native_timeout', return_value=30)) @@ -161,8 +159,12 @@ class NativeRunTests(unittest.TestCase): self.assertTrue(result['evidenceCompletion']['serverCoverageSufficient']) self.assertEqual(result['timeline'], self.accumulator.timeline) - self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current', 'query-compare']) - self.assertIn(self.snapshot['explorationWarnings'][0], result['readingSummary']['notices']) + self.assertEqual(result['conclusionProtocolQueryIds'], + ['query-current', 'query-compare', 'query-group']) + self.assertEqual(result['answerQuality']['citedEvidenceRefs'], []) + self.assertEqual(result['answerQuality']['narrativeFactBinding'], 'not_assessed') + self.assertIn(self.snapshot['explorationWarnings'][0], result['answerQuality']['notices']) + self.assertEqual(result['answer'], self.accumulator.answer) self.assertIn('firstUsefulMs', result['phaseTimingsMs']) self.assertGreaterEqual(result['phaseTimingsMs']['firstUsefulMs'], 30) self.delegated.call.assert_called_once_with('/api/analysis/protocol/queries', payload={}, method='POST') @@ -172,35 +174,33 @@ class NativeRunTests(unittest.TestCase): self.assertIsNone(self.sse.call_args.kwargs['chat_id']) stored = json.loads((self.actor_root / 'private/native-completion.json').read_text()) self.assertEqual(stored['status'], 'COMPLETED') + self.assertEqual(stored['answer'], self.accumulator.answer) self.assertEqual(self.events[-1], result) - def test_generated_chart_contract_is_available_before_the_single_synthesis(self): + def test_generated_chart_contract_uses_formal_query_evidence_without_rewriting_answer(self): self.task['visualPreference']='required' self.visuals.return_value={'visualState':'ready','missingEvidence':[], 'visuals':[{ 'id':'actual-chart','type':'line','metric':'composite_rate','title':'正式趋势', 'scope':{'department':'当前产业部'},'period':{'year':2026,'month':9}, 'categories':[{'key':'2026-09'}],'series':[{'values':[.95]}], 'evidenceRefs':['Q00000001']}]} - def synthesize(**kwargs): - preview=kwargs['context']['_platformVisuals'] - self.assertEqual(preview['visualState'],'ready') - self.assertEqual(preview['charts'][0]['id'],'actual-chart') - self.assertEqual(preview['charts'][0]['evidenceRefs'],['Q00000001']) - self.assertEqual(preview['charts'][0]['pointCount'],1) - self.assertNotIn('series',preview['charts'][0]) - return copy.deepcopy(self.final) - self.synthesis.side_effect=synthesize result=self.invoke() self.assertEqual(result['status'],'COMPLETED') - self.visuals.assert_called_once();self.synthesis.assert_called_once() + self.visuals.assert_called_once() + self.assertEqual(self.visuals.call_args.args[0], self.documents) + self.assertEqual(result['visuals'][0]['id'], 'actual-chart') + self.assertEqual(result['visuals'][0]['evidenceRefs'], ['Q00000001']) + self.assertEqual(result['answer'], self.accumulator.answer) + self.synthesis.assert_not_called() def test_root_gap_remains_partial_even_with_stop_and_available_evidence(self): self.snapshot['rootGoalGaps'] = ['亮片必答阶段尚无完整两期依据。'] result = self.invoke() self.assertEqual(result['status'], 'PARTIAL') self.assertFalse(result['evidenceCompletion']['serverCoverageSufficient']) - self.assertIn(self.snapshot['rootGoalGaps'][0], result['answer']) - self.assertIn(self.snapshot['rootGoalGaps'][0], result['readingSummary']['notices']) + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertIn(self.snapshot['rootGoalGaps'][0], result['evidenceCompletion']['missingEvidence']) + self.assertIn(self.snapshot['rootGoalGaps'][0], result['answerQuality']['missingEvidence']) def test_three_no_progress_attempts_remain_partial_after_natural_stop(self): _atomic_json(self.actor_root / 'private/native-exploration-state.json', @@ -210,7 +210,8 @@ class NativeRunTests(unittest.TestCase): self.assertEqual(result['status'], 'PARTIAL') self.assertTrue(any('连续三次' in gap for gap in result['evidenceCompletion']['missingEvidence'])) self.assertEqual(result['phaseTimingsMs']['formalQuery'], 1234) - self.assertTrue(any('连续三次' in notice for notice in result['readingSummary']['notices'])) + self.assertTrue(any('连续三次' in gap for gap in result['answerQuality']['missingEvidence'])) + self.assertEqual(result['answer'], self.accumulator.answer) def test_controlled_no_progress_stop_does_not_add_generic_tool_failure(self): _atomic_json(self.actor_root / 'private/native-exploration-state.json', @@ -233,7 +234,7 @@ class NativeRunTests(unittest.TestCase): self.assertEqual(result['status'], 'PARTIAL') self.assertTrue(any('未正常结束' in gap for gap in result['evidenceCompletion']['missingEvidence'])) - def test_only_complete_current_scope_documents_reach_synthesis(self): + def test_only_complete_current_scope_documents_reach_visual_generation_and_result(self): self.snapshot['queries'].extend([{'proof': self.p_failed}, {'proof': self.p_old}]) self.snapshot['scopeCoverage'].append( {'nodeId': 'failed', 'scopeId': 'scope-failed', 'complete': False}) @@ -245,27 +246,52 @@ class NativeRunTests(unittest.TestCase): {'evidenceRef': 'Q00000007', 'result': {'summary': {'unproved': True}}}, ]) result = self.invoke() - admitted = self.synthesis.call_args.kwargs['evidence'] + admitted = self.visuals.call_args.args[0] self.assertEqual([doc['evidenceRef'] for doc in admitted], ['Q00000001', 'Q00000002']) self.assertEqual(result['queryEvidenceRefs'], ['Q00000001', 'Q00000002']) + self.assertEqual(result['conclusionProtocolQueryIds'], + ['query-current', 'query-compare', 'query-group']) + self.assertNotIn('query-failed', result['conclusionProtocolQueryIds']) + self.assertNotIn('query-old', result['conclusionProtocolQueryIds']) + self.assertNotIn('query-not-in-current-receipts', result['conclusionProtocolQueryIds']) + self.assertEqual(result['answer'], self.accumulator.answer) self.read_evidence.assert_called_once_with(self.actor_root / 'evidence', RELEASE, exclude_files={'prior-evidence.json'}, allow_empty=True) + self.synthesis.assert_not_called() - def test_used_refs_map_only_to_actual_current_proofs_not_unused_or_foreign_documents(self): + def test_original_citations_map_only_to_actual_current_query_proofs(self): self.snapshot['queries'].append({'proof': self.p_old}) self.documents.append(document('Q00000003', self.p_old)) - self.final['usedEvidenceRefs'] = ['Q00000001', 'Q00000003', 'QDEADBEEF', 'Q00000001'] + self.accumulator.answer = '原文引用当前证据[Q00000001],以及外部证据[Q00000003]。' result = self.invoke() self.assertEqual(result['conclusionProtocolQueryIds'], ['query-current', 'query-compare']) + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['Q00000003']) + self.assertEqual(result['status'], 'PARTIAL') self.assertNotIn('query-group', result['conclusionProtocolQueryIds']) self.assertNotIn('query-old', result['conclusionProtocolQueryIds']) - def test_claimed_sufficient_answer_without_current_cited_proof_remains_partial(self): - self.final['usedEvidenceRefs'] = ['QDEADBEEF'] + def test_uppercase_hexadecimal_reference_maps_to_its_actual_receipt(self): + hexadecimal = proof('query-hexadecimal') + self.snapshot['queries'].append({'proof': hexadecimal}) + self.documents.append(document('QABCDEF12', hexadecimal)) + self.accumulator.answer = '原文显式引用[QABCDEF12]。' + + result = self.invoke() + + self.assertEqual(result['status'], 'COMPLETED') + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertEqual(result['conclusionProtocolQueryIds'], ['query-hexadecimal']) + self.assertEqual(result['answerQuality']['citedEvidenceRefs'], ['QABCDEF12']) + + def test_unbound_original_citation_remains_a_metadata_gap(self): + self.accumulator.answer = '原生回答的引用[QABCDEF99]仍按原文保留。' result = self.invoke() self.assertEqual(result['status'], 'PARTIAL') self.assertEqual(result['conclusionProtocolQueryIds'], []) - self.assertTrue(any('实际查询证明' in gap for gap in result['evidenceCompletion']['missingEvidence'])) + self.assertEqual(result['answer'], self.accumulator.answer) + self.assertEqual(result['answerQuality']['unresolvedEvidenceRefs'], ['QABCDEF99']) + self.assertTrue(any('QABCDEF99' in gap for gap in result['evidenceCompletion']['missingEvidence'])) def test_cancelled_native_turn_never_reads_query_snapshot_or_synthesizes(self): for cancellation in ('event', 'native_status'): @@ -295,12 +321,12 @@ class NativeRunTests(unittest.TestCase): self.assertNotIn('图表尚未完整生成,可独立恢复。', result['answer']) self.assertNotIn('结果为部分完成', result['answer']) self.assertNotIn('共有15张图表', result['answer']) - self.assertEqual(result['readingSummary']['notices'], + self.assertEqual(result['answerQuality']['notices'], ['部分重点图表未能完整展示;已保留核实文字和表格。']) else: - self.assertIn('图表尚未完整生成,可独立恢复。', result['answer']) - self.assertIn('图表尚未完整生成,可独立恢复。', result['readingSummary']['notices']) - self.assertIn('结果为部分完成', result['answer']) + self.assertNotIn('图表尚未完整生成,可独立恢复。', result['answer']) + self.assertNotIn('结果为部分完成', result['answer']) + self.assertIn('图表尚未完整生成,可独立恢复。', result['answerQuality']['missingEvidence']) if __name__ == '__main__': diff --git a/web/src/AnalysisConclusion.tsx b/web/src/AnalysisConclusion.tsx index f1b5748..24d1a4a 100644 --- a/web/src/AnalysisConclusion.tsx +++ b/web/src/AnalysisConclusion.tsx @@ -21,7 +21,7 @@ export function AnalysisConclusion({ reading, visuals, renderVisuals }: {
{comparison.previousLabel}{comparison.previous}
{comparison.currentLabel}{comparison.current}
-
{comparison.direction} {comparison.change}{comparison.unit}
+
{comparison.change}{comparison.unit} · {comparison.direction}
} {section.markdown && } diff --git a/web/src/AnalysisNarrative.tsx b/web/src/AnalysisNarrative.tsx index d3a7ff3..53f597c 100644 --- a/web/src/AnalysisNarrative.tsx +++ b/web/src/AnalysisNarrative.tsx @@ -1,3 +1,5 @@ +import { originalAnalysisBody } from './analysisOriginalBody'; +import { AnalysisOriginalContent } from './AnalysisOriginalContent'; import { useEffect, useMemo, useRef, useState } from 'react'; import { Button } from 'antd'; import { AiMarkdown } from './AiMarkdown'; @@ -103,6 +105,7 @@ export function AnalysisNarrative({ answer, visuals, visualState, question, turn reveal(); window.addEventListener('hashchange', reveal); return () => window.removeEventListener('hashchange', reveal); }, [model.evidence, prefix]); + const original = originalAnalysisBody(turn); const primaryIds = new Set(primaryVisuals.map(visual => visual.id)); const compact = (visual: AnalysisVisual) => !reading.hasAppendix && useCompactRateComparison(visual, question); const comparisons = model.sections.flatMap(section => section.visuals).filter(compact); @@ -130,7 +133,7 @@ export function AnalysisNarrative({ answer, visuals, visualState, question, turn ; return
- {reading.hasAppendix ? } /> : fullContent} + {original ? section.visuals)} renderVisuals={figures => } /> : reading.hasAppendix ? } /> : fullContent} {turn && drilldown && setDrilldown(null)} />}
; } diff --git a/web/src/AnalysisOriginalContent.tsx b/web/src/AnalysisOriginalContent.tsx new file mode 100644 index 0000000..64052b6 --- /dev/null +++ b/web/src/AnalysisOriginalContent.tsx @@ -0,0 +1,13 @@ +import type { ReactNode } from 'react'; +import { AiMarkdown } from './AiMarkdown'; +import { originalAnalysisSections } from './analysisOriginalBody'; +import type { AnalysisVisual } from './analysisVisualsModel'; +export function AnalysisOriginalContent({ answer, visuals, renderVisuals }: { + answer: string; visuals: AnalysisVisual[]; renderVisuals: (visuals: AnalysisVisual[]) => ReactNode; +}) { + return
{originalAnalysisSections(answer, visuals).map(section => +
+ {section.markdown.trim() && } + {!!section.visuals.length && renderVisuals(section.visuals)} +
)}
; +} diff --git a/web/src/AnalysisProcess.tsx b/web/src/AnalysisProcess.tsx index 696ac3f..afe52d0 100644 --- a/web/src/AnalysisProcess.tsx +++ b/web/src/AnalysisProcess.tsx @@ -1,7 +1,7 @@ import { Spin } from 'antd'; import { useEffect, useRef, useState, type ReactNode } from 'react'; import { CheckCircleOutlined, CloseCircleOutlined, ClockCircleOutlined, RightOutlined } from '@ant-design/icons'; -import { analysisIsRunning, compactAnalysisTimeline, duplicatesFinalAnalysisAnswer } from './analysisConversationState'; +import { analysisIsRunning, compactAnalysisTimeline } from './analysisConversationState'; import { AiMarkdown } from './AiMarkdown'; import { sanitizeAnalysisDisplayText } from './analysisDisplayText'; import { AnalysisThinkingBlock } from './ModelThinkingPanel'; @@ -45,32 +45,32 @@ export function AnalysisProcess({ turn, metadata }: { turn: Record ({ ...step, kind: 'tool', sequence: index, at: step.startedAt, hasResult: ['completed', 'failed'].includes(String(step.status)) })); const events = groupAdjacentThinking(orderedAnalysisProcess(turn, timeline.length ? timeline : fallback)); const turnId = String(turn.id || turn.turnId || ''); - const [open, setOpen] = useState(false); + const [open, setOpen] = useState(running); const previous = useRef({ turnId, running }); - const visibleOpen = previous.current.turnId === turnId && open; + const visibleOpen = previous.current.turnId !== turnId ? running : previous.current.running && !running ? false : open; useEffect(() => { const changedTurn = previous.current.turnId !== turnId; const completed = previous.current.running && !running; - if (changedTurn || completed) setOpen(false); + if (changedTurn || !previous.current.running && running) setOpen(running); + else if (completed) setOpen(false); previous.current = { turnId, running }; }, [running, turnId]); const content = events.map(entry => { if (entry.kind === 'thinking') return ; const text = sanitizeAnalysisDisplayText(String(entry.text || '')); if (entry.kind === 'assistant') { - if (!text.trim() || !running && entry.sequence === timeline.at(-1)?.sequence && timeline.at(-1)?.kind === 'assistant' && duplicatesFinalAnalysisAnswer(text, answer)) return null; + if (!running || !text.trim()) return null; return
; } const step = steps.find(item => item.toolCallId && item.toolCallId === entry.toolCallId); return ; }); - if (!events.length) return <>{metadata}{running &&

{sanitizeAnalysisDisplayText(String(turn.progressMessage || plan.progressMessage || '正在准备分析…'))}

}; + if (!events.length) return <>{metadata}; return
setOpen(event.currentTarget.open)}> {metadata || {running ? '分析中' : '已完成'}}
{content}
diff --git a/web/src/AnalysisReportNarrative.tsx b/web/src/AnalysisReportNarrative.tsx index 5721d8d..b5bc4a6 100644 --- a/web/src/AnalysisReportNarrative.tsx +++ b/web/src/AnalysisReportNarrative.tsx @@ -1,3 +1,5 @@ +import { originalAnalysisBody } from './analysisOriginalBody'; +import { AnalysisOriginalContent } from './AnalysisOriginalContent'; import { useMemo } from 'react'; import { AiMarkdown } from './AiMarkdown'; import { AnalysisConclusion } from './AnalysisConclusion'; @@ -15,6 +17,7 @@ export function AnalysisReportNarrative({ turn, answer, images }: { turn: Row; a const { model, reading, primaryVisuals } = useMemo(() => buildAnalysisReadingModel(answer, visuals, String(turn.question || ''), turn), [answer, signature, turn]); const prefix = `report-${String(turn.id || turn.turnId || 'answer').replace(/[^a-zA-Z0-9_-]/g, '')}-`; const text = (content: string) => ; + const original = originalAnalysisBody(turn); const primaryIds = new Set(primaryVisuals.map(visual => visual.id)); const complete = <> {model.sections.filter(section => !primaryIds.size || section.narrative || section.tables.length || section.focusProducts?.length || section.visuals.some(visual => !primaryIds.has(visual.id))).map(section =>
@@ -30,7 +33,7 @@ export function AnalysisReportNarrative({ turn, answer, images }: { turn: Row; a ; return
- {reading.hasAppendix ? figures.map(visual =>
+ {original ? figures.map(visual =>
image.id === visual.id)} hideEvidence />
)} /> : reading.hasAppendix ? figures.map(visual =>
image.id === visual.id)} hideEvidence />
)} /> : complete}
; diff --git a/web/src/ManagementPages.tsx b/web/src/ManagementPages.tsx index e913003..e3b2eb4 100644 --- a/web/src/ManagementPages.tsx +++ b/web/src/ManagementPages.tsx @@ -38,7 +38,7 @@ import { } from 'antd'; import type { ColumnsType } from 'antd/es/table'; import { CompactManagementTables, QualityTable as Table } from './QualityTable'; -import { DeleteOutlined, EditOutlined, MoreOutlined, ReloadOutlined, RobotOutlined } from '@ant-design/icons'; +import { DeleteOutlined, EditOutlined, MoreOutlined, ReloadOutlined, RobotOutlined, LoadingOutlined } from '@ant-design/icons'; import { ApiError, api, collectionItems, decodeQueryValueList } from './api'; import type { AuthUser, Collection, OptionsResponse, QueryParams, SourceItem, ViewKind } from './types'; import type { DataWorkflowStatus } from './NoCodeDataTypes'; @@ -50,11 +50,11 @@ import { AnalysisReportExport, NativeAnalysisReport } from './AnalysisReportExpo import { AiMarkdown } from './AiMarkdown'; import { authorizeAnalysisSnapshot } from './analysisSnapshot'; import { AnalysisApiSettings } from './AnalysisApiSettings'; -import { acceptAnalysisEvent, analysisScrollFollows, partialAnalysisNotice, partialAnswerIsOnlyProgress, type AnalysisScrollIntent } from './analysisConversationState'; +import { acceptAnalysisEvent, analysisScrollFollows, analysisStreamStatus, partialAnswerIsOnlyProgress, type AnalysisScrollIntent } from './analysisConversationState'; import { AnalysisProcess } from './AnalysisProcess'; import { AnalysisMemberListPanel } from './AnalysisMemberList'; import { AnalysisRecordListPanel } from './AnalysisRecordList'; -import { analysisScopeText, restoredAnalysisDraft, analysisComposerState, analysisFollowRequestReady } from './analysisScope'; +import { restoredAnalysisDraft, analysisComposerState, analysisFollowRequestReady } from './analysisScope'; import { isValidAnalysisPeriodContext } from './analysisPeriodModel'; import { beginAnalysisSubmission, requestSubmissionCancellation, submissionOutcomeUnknown, type AnalysisSubmission } from './analysisSubmission'; import { activeAnalysisTurn, analysisIsRunning, beginAnalysisSelection, defaultAnalysisSession, rememberAnalysisSession, rememberPendingAnalysisSession, settledAnalysisTurn } from './analysisConversationState'; @@ -1330,6 +1330,28 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow let streamedResult: Record = {}; let terminalEventStatus = ''; setLastTurn(previous => ({ ...(previous || {}), id, question, status: previous?.status || 'QUEUED' })); + let timingReadInFlight = false, timingReadAgain = false, startConfirmed = false; + const refreshTiming = async (): Promise => { + if (!linkedSessionId || !current() || startConfirmed) return; + if (timingReadInFlight) { timingReadAgain = true; return; } + timingReadInFlight = true; + try { + const history = await api.analysisSessionTurns(linkedSessionId); + if (!current()) return; + const timing = collectionItems(history).find(item => item.id === id); + if (!timing) return; + startConfirmed = typeof timing.startedAt === 'string' && Number.isFinite(Date.parse(timing.startedAt)); + setLastTurn(previous => previous?.id !== id ? previous : { ...previous, + ...(typeof timing.createdAt === 'string' ? { createdAt: timing.createdAt } : {}), + ...(startConfirmed ? { startedAt: timing.startedAt } : {}), + }); + } catch (value) { if (current()) onUnauthorized(value); } + finally { + timingReadInFlight = false; + if (timingReadAgain && !startConfirmed && current()) { timingReadAgain = false; void refreshTiming(); } + } + }; + void refreshTiming(); try { let detail: Record = {}; let reconnectAttempt = 0; @@ -1350,15 +1372,13 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow if (eventResult) streamedResult = { ...streamedResult, ...eventResult }; const eventName = event.event.toUpperCase().replaceAll('.', '_').replaceAll('-', '_'); const eventStatus = String(data.status || '').toUpperCase(); - if (ANALYSIS_FINAL_STATUSES.has(eventStatus)) terminalEventStatus = eventStatus; - else if (/^(?:TURN_)?(COMPLETED|PARTIAL|FAILED|CANCELLED|NEEDS_CLARIFICATION)$/.test(eventName)) { - terminalEventStatus = eventName.replace(/^TURN_/, ''); - } + const turnEventStatus = analysisStreamStatus(undefined, event.event, data.status); + if (ANALYSIS_FINAL_STATUSES.has(turnEventStatus)) terminalEventStatus = turnEventStatus; setLastTurn(previous => { const prior = previous || {}; if (prior.id === id && ANALYSIS_FINAL_STATUSES.has(String(prior.status || '').toUpperCase()) && !ANALYSIS_FINAL_STATUSES.has(eventStatus)) return prior; const priorResult = isObjectRecord(prior.result) ? prior.result : {}; - const status = data.status || (['progress', 'message', 'heartbeat'].includes(event.event.toLowerCase()) ? prior.status || 'RUNNING' : event.event.toUpperCase()); + const status = analysisStreamStatus(prior.status, event.event, data.status); return { ...prior, id, @@ -1379,6 +1399,7 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow // The SSE request can otherwise remain open after its terminal event. // Closing this subscription does not cancel the analysis itself. if (terminalEventStatus) controller.abort(); + else if (eventStatus === 'RUNNING' || eventName.startsWith('TOOL_')) void refreshTiming(); }, }); } catch (streamError) { @@ -1530,7 +1551,8 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow followRequest.current = attempt.idempotencyKey; followOutput.current = true; explicitlyFollowing.current = false; - setLastTurn({ ...turn, status: 'QUEUED', _uiFollowKey: attempt.idempotencyKey }); + const submittedAt = new Date().toISOString(); + setLastTurn({ id, sessionId: targetSessionId, question, status: 'SUBMITTING', _uiSubmittedAt: submittedAt, _uiFollowKey: attempt.idempotencyKey }); const current = () => mounted.current && generation === navigationGeneration.current && submissionAttempt.current === attempt; setError(''); setBusy(true); setActiveTurnId(id); try { @@ -1548,7 +1570,7 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow if (!current()) return; const retriedDetail = await api.analysisTurn(retryId); if (!current()) return; - setLastTurn({ ...retriedDetail, ...retried, id: retryId, sessionId: retrySessionId, question, _uiFollowKey: attempt.idempotencyKey }); + setLastTurn({ ...retriedDetail, ...retried, id: retryId, sessionId: retrySessionId, question, _uiSubmittedAt: submittedAt, _uiFollowKey: attempt.idempotencyKey }); writeAnalysisLocation(retrySessionId, retryId); void loadSessions(); if (!current()) return; @@ -1590,7 +1612,8 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow setError(''); setRetryQuestion(question.trim()); if (typeof lastTurn?.id === 'string' && lastTurn.id) setTurns(previous => previous.some(item => item.id === lastTurn.id) ? previous : [...previous, lastTurn]); - setLastTurn({ question: question.trim(), status: 'QUEUED', _uiFollowKey: attempt.idempotencyKey }); + const submittedAt = new Date().toISOString(); + setLastTurn({ question: question.trim(), status: 'SUBMITTING', _uiSubmittedAt: submittedAt, _uiFollowKey: attempt.idempotencyKey }); setQuery(''); try { let selectedSessionId = sessionId; @@ -1619,7 +1642,7 @@ function AnalysisPage({ onUnauthorized, user }: { onUnauthorized: (error: unknow attempt.uncertain = false; attempt.acceptedTurnId = turnId; setActiveTurnId(turnId); - setLastTurn({ ...turn, id: turnId, question: question.trim(), sessionId: selectedSessionId, _uiFollowKey: attempt.idempotencyKey }); + setLastTurn({ ...turn, status: turn.status || 'QUEUED', id: turnId, question: question.trim(), sessionId: selectedSessionId, _uiSubmittedAt: submittedAt, _uiFollowKey: attempt.idempotencyKey }); writeAnalysisLocation(selectedSessionId, turnId); if (attempt.cancelRequested && await cancelAcceptedTurn(turnId, generation, question.trim())) return; void loadSessions(); @@ -1744,7 +1767,6 @@ function AnalysisTurn({ turn: sourceTurn, onResume, onRetry, onClarify, onUnauth const question = typeof turn.question === 'string' ? turn.question : ''; if (!turn.id && typeof turn.submissionNotice === 'string') return
{question &&

{question}

} -

{turn.submissionNotice}

; const result = isObjectRecord(turn.result) ? turn.result : turn; const memberList = result.memberList ?? turn.memberList; @@ -1754,24 +1776,15 @@ function AnalysisTurn({ turn: sourceTurn, onResume, onRetry, onClarify, onUnauth const visuals = result.visualSchemaVersion === 1 ? result.visuals : undefined; const reportMarkdown = typeof result.reportMarkdown === 'string' ? result.reportMarkdown : typeof turn.reportMarkdown === 'string' ? turn.reportMarkdown : ''; const reportHtml = typeof result.reportHtml === 'string' ? result.reportHtml : typeof turn.reportHtml === 'string' ? turn.reportHtml : undefined; - const resumable = ['QUEUED', 'RUNNING', 'PROCESSING', 'PLANNING', 'TOOL_RUNNING'].includes(status); - const errorValue = result.error || turn.error; - const failure = typeof errorValue === 'string' ? errorValue : '分析暂时中断,可以重试或继续提问。'; - const emptyStatus = status === 'CANCELLED' ? '本次分析已取消,未返回回答内容。' - : status === 'PARTIAL' ? '本轮已保存查询过程,尚未形成最终答复。' - : status === 'COMPLETED' ? '分析任务已完成,但未返回回答内容。' - : failure; - const scope = analysisScopeText(turn); + const resumable = ['SUBMITTING', 'QUEUED', 'RUNNING', 'PROCESSING', 'PLANNING', 'TOOL_RUNNING'].includes(status); return
{question &&

{question}

} - {scope &&

{scope}

} } /> {(!resumable || !Array.isArray(turn.timeline || (isObjectRecord(turn.plan) ? turn.plan.timeline : undefined))) && (answer || !resumable || !retryDisabled) &&
{status === 'NEEDS_CLARIFICATION' && onClarify ? : answer ? <> {resumable ? : authorizeAnalysisSnapshot(turn)} turn={turn} />} - :

{resumable ? '分析任务仍在运行,尚未返回完整回答。' : emptyStatus}

} - {status === 'PARTIAL' && answer &&

{partialAnalysisNotice(result.stopReason || turn.stopReason || (isObjectRecord(turn.plan) ? turn.plan.stopReason : ''), result)}

} + : null}
{(reportMarkdown.trim() || reportHtml?.trim()) && } {!reportMarkdown.trim() && !reportHtml?.trim() && answer && ['COMPLETED', 'PARTIAL'].includes(status) && } @@ -1779,7 +1792,6 @@ function AnalysisTurn({ turn: sourceTurn, onResume, onRetry, onClarify, onUnauth {['FAILED', 'CANCELLED', 'PARTIAL'].includes(status) && typeof turn.id === 'string' && onRetry && } {resumable && !retryDisabled && typeof turn.id === 'string' && onResume && }
- {recovery.error&&

{recovery.error}

}
} {memberList !== undefined && memberList !== null && } {recordList !== undefined && recordList !== null && } @@ -1788,21 +1800,23 @@ function AnalysisTurn({ turn: sourceTurn, onResume, onRetry, onClarify, onUnauth function AnalysisTurnMetadata({ turn, status, compact = false }: { turn: Record; status: string; compact?: boolean }) { const [now, setNow] = useState(Date.now()); - const running = analysisIsRunning(status); + const running = analysisIsRunning(status) || status === 'SUBMITTING'; useEffect(() => { if (!running) return; setNow(Date.now()); const timer = window.setInterval(() => setNow(Date.now()), 1000); return () => window.clearInterval(timer); }, [running]); const plan = isObjectRecord(turn.plan) ? turn.plan : {}; const date = (value: unknown) => typeof value === 'string' && Number.isFinite(Date.parse(value)) ? Date.parse(value) : undefined; const start = date(turn.startedAt) ?? date(plan.startedAt); + const waitingSince = date(turn._uiSubmittedAt) ?? date(turn.createdAt); + const clockStart = start ?? (running ? waitingSince : undefined); const finish = date(turn.finishedAt); - const duration = typeof turn.durationMs === 'number' ? turn.durationMs : start !== undefined && (running || finish !== undefined) ? Math.max(0, (running ? now : finish!) - start) : undefined; + const duration = !running && typeof turn.durationMs === 'number' ? turn.durationMs : clockStart !== undefined && (running || finish !== undefined) ? Math.max(0, (running ? now : finish!) - clockStart) : undefined; const seconds = duration === undefined ? undefined : Math.floor(duration / 1000); const durationText = seconds === undefined ? '' : seconds >= 3600 ? `${Math.floor(seconds / 3600)}小时${Math.floor(seconds % 3600 / 60)}分${seconds % 60}秒` : seconds >= 60 ? `${Math.floor(seconds / 60)}分${seconds % 60}秒` : `${seconds}秒`; const stamp = (value: number) => new Intl.DateTimeFormat('zh-CN', { timeZone: 'Asia/Shanghai', month: '2-digit', day: '2-digit', hour: '2-digit', minute: '2-digit', second: '2-digit', hour12: false }).format(value); const steps = Array.isArray(turn.steps) ? turn.steps : Array.isArray(plan.steps) ? plan.steps : []; - const labels: Record = { QUEUED: '排队中', RUNNING: '运行中', COMPLETED: '已完成', PARTIAL: '部分完成', FAILED: '未完成', CANCELLED: '已停止', NEEDS_CLARIFICATION: '待补充信息' }; + const labels: Record = { SUBMITTING: '提交中', QUEUED: '排队中', RUNNING: '运行中', COMPLETED: '已完成', PARTIAL: '', FAILED: '未完成', CANCELLED: '已停止', NEEDS_CLARIFICATION: '待补充信息' }; return - {labels[status] || (running ? '运行中' : '分析记录')} - {durationText && {running ? '已运行' : '运行时长'} {durationText}} + {status !== 'PARTIAL' && {labels[status] || (running ? '运行中' : '分析记录')}} + {durationText && {running && } {!compact && start !== undefined && 开始 {stamp(start)}} {!compact && !running && finish !== undefined && {status === 'COMPLETED' ? '完成时间' : '结束时间'} {stamp(finish)}} {!compact && steps.length > 0 && 工具调用 {steps.length} 次} diff --git a/web/src/analysis-chat.css b/web/src/analysis-chat.css index fcf6a41..61d81e6 100644 --- a/web/src/analysis-chat.css +++ b/web/src/analysis-chat.css @@ -141,15 +141,15 @@ .analysis-process-history{margin:0;min-width:0;color:var(--muted-ink,#63758b)} .analysis-process-history>summary{display:flex;align-items:center;gap:8px;width:fit-content;min-height:24px;cursor:pointer;list-style:none;font-size:var(--font-control)} .analysis-process-history>summary::-webkit-details-marker{display:none} -.analysis-process-history>summary .anticon{font-size:var(--font-note);transition:transform .16s ease} -.analysis-process-history[open]>summary .anticon{transform:rotate(90deg)} +.analysis-process-history>summary>.anticon{font-size:var(--font-note);transition:transform .16s ease} +.analysis-process-history[open]>summary>.anticon{transform:rotate(90deg)} .analysis-process-history>summary:focus-visible{outline:2px solid var(--primary-color,#607ae1);outline-offset:3px} .analysis-process-history-body{padding:8px 0;max-height:360px;overflow:auto;overscroll-behavior:contain} .analysis-queued-followup{display:flex;align-items:center;gap:8px;color:var(--muted-ink,#63758b);font-size:var(--font-control);min-width:0} .analysis-queued-followup>span{min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap} .analysis-queued-followup .ant-btn{flex:none;font-size:var(--font-control)} @media(max-width:700px){.analysis-simple-page .ai-actions{align-items:flex-start;gap:8px;flex-wrap:wrap}.analysis-composer-buttons{margin-left:auto;gap:5px}} -@media(prefers-reduced-motion:reduce){.analysis-process-history>summary .anticon{transition:none}} +@media(prefers-reduced-motion:reduce){.analysis-process-history>summary>.anticon{transition:none}} @keyframes analysis-event-appear{from{opacity:.4;transform:translateY(2px)}to{opacity:1;transform:none}} .analysis-agent-stream>*{animation:analysis-event-appear .14s ease-out} @@ -171,3 +171,7 @@ .analysis-clarification-actions{display:flex;justify-content:flex-end;margin-top:10px} .analysis-clarification fieldset:disabled .analysis-clarification-option{cursor:default;opacity:.65} .analysis-clarification-option:focus-within{outline:2px solid var(--primary-color,#607ae1);outline-offset:2px} + +.analysis-runtime-clock{display:inline-flex;align-items:center;gap:5px} +.analysis-turn-metadata .analysis-runtime-spinner{font-size:12px} +@media(prefers-reduced-motion:reduce){.analysis-runtime-spinner,.analysis-runtime-spinner svg{animation:none!important}} diff --git a/web/src/analysisConclusionModel.ts b/web/src/analysisConclusionModel.ts index 170d6ee..243ab4a 100644 --- a/web/src/analysisConclusionModel.ts +++ b/web/src/analysisConclusionModel.ts @@ -16,7 +16,11 @@ export function checkedConclusionComparison(sentence: string): ConclusionCompari const value = sentence.replace(/\*\*/g, '').trim(); const match = value.match(/^([^。;\n]{1,50}?)(综合合格率|黑片合格率|亮片合格率|合格率)(?:为|是|[::])?\s*(\d+(?:\.\d+)?%)\s*[,,]\s*(?:较|相比|对比|比)\s*([^,,。;\n]{1,30}?)的?\s*(\d+(?:\.\d+)?%)\s*(下降|上升|提高|降低)\s*(\d+(?:\.\d+)?)(个百分点|pp)[。.]?$/); if (!match || [match[3], match[5]].some(rate => Number.parseFloat(rate) > 100)) return null; - return { metric: match[2], currentLabel: match[1], current: match[3], previousLabel: match[4].replace(/的$/, ''), previous: match[5], direction: match[6], change: match[7], unit: match[8], original: value }; + const currentRate = Number.parseFloat(match[3]), previousRate = Number.parseFloat(match[5]); + const isDecrease = /下降|降低/.test(match[6]); + if (currentRate !== previousRate && (currentRate < previousRate) !== isDecrease) return null; + const change = `${isDecrease ? '-' : '+'}${match[7]}`; + return { metric: match[2], currentLabel: match[1], current: match[3], previousLabel: match[4].replace(/的$/, ''), previous: match[5], direction: match[6], change, unit: match[8], original: value }; } function topicMatches(title: string, visual: AnalysisVisual): boolean { diff --git a/web/src/analysisConversationState.ts b/web/src/analysisConversationState.ts index 49b052e..65c780e 100644 --- a/web/src/analysisConversationState.ts +++ b/web/src/analysisConversationState.ts @@ -136,3 +136,16 @@ export function partialAnswerIsOnlyProgress(turn: Record): bool if (lastAssistant < 0 || String(record(entries[lastAssistant]).text || '').trim() !== answer) return false; return entries.slice(lastAssistant + 1).some(item => ['tool_start', 'tool_result'].includes(String(record(item).kind))); } + +/** Tool event names describe a step, never the lifecycle of the whole turn. */ +export function analysisStreamStatus(previous: unknown, event: string, reported: unknown): string { + const prior = String(previous || '').toUpperCase(); + const terminal = new Set(['COMPLETED', 'PARTIAL', 'FAILED', 'CANCELLED', 'NEEDS_CLARIFICATION']); + if (terminal.has(prior)) return prior; + const name = event.toUpperCase().replaceAll('.', '_').replaceAll('-', '_'); + const status = String(reported || '').toUpperCase(); + if (/^(?:TURN_)?(COMPLETED|PARTIAL|FAILED|CANCELLED|NEEDS_CLARIFICATION)$/.test(name)) return name.replace(/^TURN_/, ''); + if (!name.startsWith('TOOL_') && (analysisIsRunning(status) || terminal.has(status))) return status; + if (name.startsWith('TOOL_') || name === 'ANSWER') return 'RUNNING'; + return analysisIsRunning(prior) ? prior : 'RUNNING'; +} diff --git a/web/src/analysisOriginalBody.ts b/web/src/analysisOriginalBody.ts new file mode 100644 index 0000000..66c1e48 --- /dev/null +++ b/web/src/analysisOriginalBody.ts @@ -0,0 +1,69 @@ +import { analysisTextEvidenceRefs } from './analysisEvidenceRefs.ts'; +import type { AnalysisVisual } from './analysisVisualsModel.ts'; +type Row = Record; +const object = (value: unknown): Row => value && typeof value === 'object' && !Array.isArray(value) ? value as Row : {}; +const text = (value: unknown) => typeof value === 'string' ? value : ''; +export interface OriginalAnalysisBody { answer: string; source: 'saved' | 'timeline' | 'unavailable' } + +/** Older native turns saved a replacement report; their final assistant remains in the authorized timeline. */ +export function originalAnalysisBody(turn?: Row): OriginalAnalysisBody | null { + if (!turn) return null; + const result = turn.result && typeof turn.result === 'object' ? object(turn.result) : turn; + const plan = object(turn.plan || result.plan); + if (![turn.executionMode, result.executionMode, plan.executionMode].includes('native_agent')) return null; + const saved = text(result.answer); + // Explicitly empty/redacted answers must not be replenished from an older transcript. + if (!saved.trim()) return { answer: '', source: 'unavailable' }; + const lead = text(object(result.readingSummary).lead); + const replaced = /已取得本轮统计结果,答案整理尚未完整完成|已取得本轮统计结果,部分解释尚未完成|现有证据不足以支持本轮完整分析,已保留可核验统计结果/.test(lead || saved.slice(0, 160)); + if (!replaced) return { answer: saved, source: 'saved' }; + const raw = turn.timeline || result.timeline || plan.timeline; + const timeline = Array.isArray(raw) ? raw.map(object) : []; + let lastTool = -1; + timeline.forEach((entry, index) => { if (/^tool(?:_|$)/.test(text(entry.kind))) lastTool = index; }); + const parts = timeline.slice(lastTool + 1).filter(entry => entry.kind === 'assistant' && text(entry.text).trim()).map(entry => text(entry.text)); + // Historical public timeline entries were capped at 8000 characters without a truncation flag. + if (parts.some(part => part.length >= 8000)) return { answer: '', source: 'unavailable' }; + return parts.length ? { answer: parts.join('\n\n'), source: 'timeline' } : { answer: '', source: 'unavailable' }; +} + +export interface OriginalAnalysisSection { id: string; markdown: string; visuals: AnalysisVisual[] } +/** Keep every Markdown line in order; images are additions after the associated chapter. */ +export function originalAnalysisSections(answer: string, visuals: AnalysisVisual[]): OriginalAnalysisSection[] { + const sections: Array = []; + let current = { id: 'original-1', markdown: '', visuals: [] as AnalysisVisual[], title: '', lines: [] as string[] }; + sections.push(current); + let fenced = false; + for (const line of answer.split('\n')) { + if (/^\s*```/.test(line)) fenced = !fenced; + const heading = !fenced ? line.match(/^#{1,6}\s+(.+?)\s*#*\s*$/) : null; + if (heading && current.lines.length) { + current = { id: `original-${sections.length + 1}`, markdown: '', visuals: [], title: heading[1], lines: [] }; + sections.push(current); + } else if (heading) current.title = heading[1]; + current.lines.push(line); + } + sections.forEach(section => { section.markdown = section.lines.join('\n'); }); + const relevant = (title: string, visual: AnalysisVisual): number => { + const stage = visual.stage || (visual.metric.startsWith('black_') ? 'black' : visual.metric.startsWith('bright_') ? 'bright' : undefined); + if (stage === 'black' && /亮片/.test(title) && !/黑片/.test(title) + || stage === 'bright' && /黑片/.test(title) && !/亮片/.test(title)) return 0; + + if (visual.dimension === 'product') return /规格|产品/.test(title) ? 4 : 0; + if (visual.dimension === 'department' || visual.dimension === 'group') return /组织|产业部|分组|区域/.test(title) ? 4 : 0; + if (visual.dimension === 'defect' || visual.metric.startsWith('source_defect')) return /缺陷|缺角|隐裂|源字段/.test(title) ? 4 : 0; + if (/category/.test(visual.sectionKey || '')) return /报废大类|报废类别|正式报废/.test(title) ? 4 : 0; + if (visual.type === 'line' || visual.type === 'area') return /趋势|逐月|月度/.test(title) ? 4 : 0; + return /总体|阶段/.test(title) || stage === 'black' && /黑片/.test(title) || stage === 'bright' && /亮片/.test(title) ? 4 : /结论/.test(title) ? 2 : 0; + }; + for (const visual of visuals) { + const candidates = sections.map(section => { + const refs = analysisTextEvidenceRefs(section.markdown); + const topic = relevant(section.title, visual); + return { section, score: topic ? topic + (visual.evidenceRefs.some(ref => refs.includes(ref)) ? 10 : 0) : 0 }; + }).filter(item => item.score > 0).sort((a, b) => b.score - a.score); + if (candidates.length) candidates[0].section.visuals.push(visual); + else sections.push({ id: `original-chart-${sections.length + 1}`, title: '', lines: [], markdown: '', visuals: [visual] }); + } + return sections.filter(section => section.markdown.trim() || section.visuals.length).map(({ id, markdown, visuals }) => ({ id, markdown, visuals })); +} diff --git a/web/src/analysisSnapshot.ts b/web/src/analysisSnapshot.ts index 2156ad1..2c261d5 100644 --- a/web/src/analysisSnapshot.ts +++ b/web/src/analysisSnapshot.ts @@ -7,6 +7,12 @@ export function analysisSnapshotResult(turn: Row): Row { return turn.result && typeof turn.result === 'object' && !Array.isArray(turn.result) ? object(turn.result) : turn; } +function originalTranscript(turn: Row): unknown { + const result = analysisSnapshotResult(turn), plan = object(turn.plan || result.plan); + if (![turn.executionMode, result.executionMode, plan.executionMode].includes('native_agent')) return null; + return turn.timeline || result.timeline || plan.timeline || []; +} + function canonical(value: unknown): string { if (Array.isArray(value)) return `[${value.map(canonical).join(',')}]`; if (value && typeof value === 'object') { @@ -19,16 +25,17 @@ export function reportSnapshotIdentity(turn: Row): string { const result = analysisSnapshotResult(turn); const value = (key: string, fallback: unknown) => ['readingSummary', 'queryEvidenceRefs'].includes(key) ? result[key] ?? fallback : result[key] ?? turn[key] ?? fallback; - return canonical({ answer: value('answer', ''), visuals: value('visuals', []), + return canonical({ originalTranscript: originalTranscript(turn), answer: value('answer', ''), visuals: value('visuals', []), results: value('results', []), reportMarkdown: value('reportMarkdown', ''), reportHtml: value('reportHtml', ''), memberList: value('memberList', null), recordList: value('recordList', null), readingSummary: value('readingSummary', null), queryEvidenceRefs: value('queryEvidenceRefs', []), question: turn.question || '' }); } export function reportSnapshotDifferences(left: Row, right: Row): string[] { - const fields = ['answer', 'visuals', 'results', 'reportMarkdown', 'reportHtml', 'memberList', 'recordList', 'readingSummary', 'queryEvidenceRefs', 'question']; + const fields = ['answer', 'visuals', 'results', 'reportMarkdown', 'reportHtml', 'memberList', 'recordList', 'readingSummary', 'queryEvidenceRefs', 'question', 'originalTranscript']; return fields.filter(key => { const value = (turn: Row) => { + if (key === 'originalTranscript') return originalTranscript(turn); const result = analysisSnapshotResult(turn); return key === 'question' ? turn.question || '' : ['readingSummary', 'queryEvidenceRefs'].includes(key) ? result[key] ?? (key === 'readingSummary' ? null : []) : result[key] ?? turn[key] ?? (key === 'memberList' || key === 'recordList' || key === 'readingSummary' ? null : key === 'visuals' || key === 'results' || key === 'queryEvidenceRefs' ? [] : ''); @@ -47,7 +54,7 @@ export async function authorizeAnalysisSnapshot(turn: Row): Promise { } const differences = reportSnapshotDifferences(current, turn); if (differences.length) { - const labels: Record = { answer: '回答', visuals: '图表', results: '查询结果', reportMarkdown: '报告正文', reportHtml: '报告 HTML', memberList: '成员清单', recordList: '原始记录', readingSummary: '核心结论', queryEvidenceRefs: '数据依据', question: '问题' }; + const labels: Record = { answer: '回答', visuals: '图表', results: '查询结果', reportMarkdown: '报告正文', reportHtml: '报告 HTML', memberList: '成员清单', recordList: '原始记录', readingSummary: '核心结论', queryEvidenceRefs: '数据依据', question: '问题', originalTranscript: '原始正文' }; throw new Error(`分析结果或访问权限已变化(${differences.map(key => labels[key] || key).join('、')}),请重新打开这条分析后再导出。`); } } diff --git a/web/src/analysisVisualsModel.ts b/web/src/analysisVisualsModel.ts index 65e65c8..776d18c 100644 --- a/web/src/analysisVisualsModel.ts +++ b/web/src/analysisVisualsModel.ts @@ -121,6 +121,15 @@ function hasOnlyKeys(value: Record, allowed: Set): bool return Object.keys(value).every(key => allowed.has(key)); } +function canonicalLegacyRateDropNote(note: string | undefined): string { + const details = (note || '') + .replace(/^(?:条形末端显示具体变化:)?[^;。]*?(?:正值|负值)[^;。]*[;。]\s*/, '') + .replace(/^[;;,,。.\s]+/, '') + .trim(); + const basis = '正值表示本期合格率上升,负值表示下降;变化值为本期减比较期。'; + return details ? `${basis} ${details}` : basis; +} + /** 解析前仅复制普通 JSON 数据,阻止 getter、toJSON、原型和函数进入渲染链路。 */ function copyJsonData(value: unknown, seen = new Set(), budget = { nodes: 0 }, depth = 0): unknown { if (depth > 32 || ++budget.nodes > 100_000) throw new Error('unsafe JSON data'); @@ -272,7 +281,7 @@ function parseVisual(value: unknown, index: number, issues: string[]): AnalysisV const comparisonPeriod = value.comparisonPeriod === undefined ? undefined : safeContext(value.comparisonPeriod); const evidenceRefs = safeTextList(value.evidenceRefs, 100); const requiredFields = safeTextList(value.requiredFields, 100); - const note = value.note === undefined ? undefined : boundedText(value.note, 500); + const rawNote = value.note === undefined ? undefined : boundedText(value.note, 500); const stage = value.stage === undefined ? undefined : value.stage; const sectionKey = value.sectionKey === undefined ? undefined : value.sectionKey; const dimension = value.dimension === undefined ? undefined : value.dimension; @@ -292,6 +301,12 @@ function parseVisual(value: unknown, index: number, issues: string[]): AnalysisV ...(coverage.denominator === undefined ? {} : { denominator: Number(coverage.denominator) }), } : null; + // v1 saved rate-drop charts used positive values for declines. New charts + // carry "本期减比较期" in the existing bounded note field. This survives the + // backend's strict v1 field allowlist; unmarked snapshots are legacy values. + const normalizeLegacyRateDrop = metric === 'rate_drop_pp' && unit === 'pp' + && !(rawNote || '').includes('本期减比较期'); + const note = normalizeLegacyRateDrop ? canonicalLegacyRateDropNote(rawNote || undefined) : rawNote; const categories = Array.isArray(value.categories) && value.categories.length > 0 && value.categories.length <= MAX_VISUAL_CATEGORIES ? value.categories.map(category => { @@ -325,7 +340,7 @@ function parseVisual(value: unknown, index: number, issues: string[]): AnalysisV return null; } if (unit === 'ratio' && (item < 0 || (!allowsAboveOneRatio(metric || '') && item > 1))) invalidValueCells.push(`${seriesIndex}:${valueIndex}`); - return item; + return normalizeLegacyRateDrop ? -item : item; }); series.push({ key, label, values }); } @@ -335,7 +350,7 @@ function parseVisual(value: unknown, index: number, issues: string[]): AnalysisV const valid = !!id && !!type && !!title && subtitle !== null && UNIT_VALUES.has(unit as AnalysisVisualUnit) && !!metric && !!originTurnId && !!scope && isValidAnalysisPeriodContext(scope) && !!period && isValidAnalysisPeriodContext(period) && (value.comparisonPeriod === undefined || !!comparisonPeriod && isValidAnalysisPeriodContext(comparisonPeriod)) - && !!evidenceRefs && !!requiredFields && (value.note === undefined || note !== null) + && !!evidenceRefs && !!requiredFields && (value.note === undefined || rawNote !== null) && (stage === undefined || stage === 'black' || stage === 'bright') && (sectionKey === undefined || SECTION_KEYS.has(sectionKey as AnalysisVisualSectionKey)) && (dimension === undefined || DIMENSIONS.has(dimension as AnalysisVisualDimension)) diff --git a/web/verification/analysis-original.html b/web/verification/analysis-original.html new file mode 100644 index 0000000..3083a32 --- /dev/null +++ b/web/verification/analysis-original.html @@ -0,0 +1 @@ +AI原文与运行状态验收
diff --git a/web/verification/analysis-original.tsx b/web/verification/analysis-original.tsx new file mode 100644 index 0000000..6cdab7a --- /dev/null +++ b/web/verification/analysis-original.tsx @@ -0,0 +1,41 @@ +// Local acceptance fixture: all requests are intercepted; no model, API or database writes. +import React from 'react'; +import {createRoot} from 'react-dom/client'; +import {App,ConfigProvider} from 'antd'; +import ManagementPages from '../src/ManagementPages'; +import '../src/styles.css'; +import '../src/workspace-finish.css'; +const user={id:'original-fixture',username:'fixture',displayName:'本地验收',roles:['SYSTEM_ADMIN'],departments:[],groups:[]}; +const modelAnswer='## 结论\n\n本期合格率下降,两个阶段的变化机制不同。完整模型正文必须保留。\n\n### 总体与阶段\n\n黑片合格率从97.58%降至97.27%,变化为 -0.30 pp。\n\n| 阶段 | 变化 |\n| --- | --- |\n| 黑片 | -0.30 pp |\n\n表格之后继续解释投入规模与报废规模,不能因为增加图表就丢掉这段。[Q11111111]\n\n### 下一步\n\n上述是统计线索,仍需现场核实生产原因。'; +let question='',status='QUEUED',createdAt='',startedAt='',finishedAt='';let timeline:Record[]=[]; +const visual={id:'stage-change',type:'column',title:'黑片合格率变化',subtitle:'',unit:'pp',metric:'rate_drop_pp',stage:'black',categories:[{key:'black',label:'黑片'}],series:[{key:'rateDrop',label:'合格率降幅',values:[.30]}],evidenceRefs:['Q11111111'],requiredFields:['black_good_weight','black_total_weight'],scope:{year:2026,month:9,releaseId:'fixture'},period:{year:2026,month:9},comparisonPeriod:{year:2026,month:8},originTurnId:'fixture-turn',coverage:{totalRows:1,returnedRows:1,truncated:false}}; +const saved=()=>({id:'fixture-turn',sessionId:'fixture-session',question,status,createdAt,startedAt,finishedAt,executionMode:'native_agent',plan:{timeline},answer:'已取得本轮统计结果,答案整理尚未完整完成。',readingSummary:{version:1,lead:'已取得本轮统计结果,答案整理尚未完整完成。',evidence:[],notices:[],evidenceRefs:['Q11111111']},queryEvidenceRefs:['Q11111111'],visualSchemaVersion:1,visualState:'ready',visuals:[visual],stopReason:'evidence_incomplete'}); +const json=(value:unknown)=>new Response(JSON.stringify(value),{headers:{'Content-Type':'application/json'}}); +window.fetch=async(input,init)=>{ + const path=new URL(String(input),location.origin).pathname; + if(path==='/api/auth/csrf')return json({token:'fixture'}); + if(path==='/api/options')return json({years:[2026],departments:[],groups:[],products:[],projectGroups:[]}); + if(path==='/api/analysis/sessions'&&init?.method==='POST'){await new Promise(r=>setTimeout(r,1500));return json({session:{id:'fixture-session'}});} + if(path==='/api/analysis/sessions')return json({items:question?[{id:'fixture-session',title:question,activeTurnStatus:status}]:[]}); + if(path==='/api/analysis/sessions/fixture-session/turns'&&init?.method==='POST'){ + question=JSON.parse(String(init.body)).question;createdAt=new Date().toISOString();await new Promise(r=>setTimeout(r,300));return json({turn:{id:'fixture-turn',sessionId:'fixture-session',status,createdAt}}); + } + if(path==='/api/analysis/sessions/fixture-session/turns')return json({items:[{id:'fixture-turn',status,createdAt,startedAt,finishedAt}]}); + if(path==='/api/analysis/sessions/fixture-session')return json({id:'fixture-session',context:{year:2026},turns:status==='PARTIAL'?[saved()]:[]}); + if(path==='/api/analysis/turns/fixture-turn'||path.endsWith('/snapshot'))return json(saved()); + if(path.endsWith('/events')){ + const encoder=new TextEncoder();let closed=false;const timers:ReturnType[]=[]; + const stream=new ReadableStream({start(controller){ + let eventId=0;const send=(event:string,data:unknown)=>{if(!closed)controller.enqueue(encoder.encode(`id: ${++eventId}\nevent: ${event}\ndata: ${JSON.stringify(data)}\n\n`));}; + send('progress',{status:'QUEUED'}); + timers.push(setTimeout(()=>{status='RUNNING';startedAt=new Date().toISOString();send('progress',{status,message:'正在理解本轮目标和连续追问…'});},300)); + timers.push(setTimeout(()=>{timeline=[{kind:'tool_start',sequence:1,tool:'read',toolCallId:'read-1',label:'核对正式指标',status:'running'}];send('tool_started',{timeline,step:timeline[0]});},800)); + timers.push(setTimeout(()=>{timeline.push({kind:'tool_result',sequence:2,tool:'read',toolCallId:'read-1',label:'核对正式指标',status:'completed'});send('tool_completed',{timeline,step:timeline[1]});},6000)); + timers.push(setTimeout(()=>{status='PARTIAL';finishedAt=new Date().toISOString();timeline.push({kind:'assistant',sequence:3,text:modelAnswer});send('partial',{status,result:saved()});if(!closed){closed=true;controller.close();}},30000)); + init?.signal?.addEventListener('abort',()=>{if(!closed){closed=true;controller.close();}timers.forEach(clearTimeout);},{once:true}); + },cancel(){closed=true;timers.forEach(clearTimeout);}}); + return new Response(stream,{headers:{'Content-Type':'text/event-stream'}}); + } + return json({items:[]}); +}; +createRoot(document.getElementById('root')!).render(
false} onNavigate={()=>{}}/>
); diff --git a/web/verification/analysis_live_status.test.mjs b/web/verification/analysis_live_status.test.mjs new file mode 100644 index 0000000..fed95b1 --- /dev/null +++ b/web/verification/analysis_live_status.test.mjs @@ -0,0 +1,62 @@ +import test from 'node:test'; +import assert from 'node:assert/strict'; +import fs from 'node:fs'; +import ts from 'typescript'; +import React from 'react'; +import {renderToStaticMarkup} from 'react-dom/server'; +import {analysisStreamStatus,analysisIsRunning,acceptAnalysisEvent,settledAnalysisTurn} from '../src/analysisConversationState.ts'; +const source=fs.readFileSync(new URL('../src/ManagementPages.tsx',import.meta.url),'utf8'); +const compile=code=>ts.transpileModule(code,{fileName:'status.tsx',compilerOptions:{target:ts.ScriptTarget.ES2022,jsx:ts.JsxEmit.React}}).outputText; +const tick=()=>new Promise(resolve=>setImmediate(resolve)); +const deferred=()=>{let resolve;const promise=new Promise(r=>resolve=r);return{promise,resolve};}; +const isObjectRecord=v=>!!v&&typeof v==='object'&&!Array.isArray(v); +const finalStatuses=new Set(['COMPLETED','PARTIAL','FAILED','CANCELLED','NEEDS_CLARIFICATION']); +function fixture(){ + let lastTurn={id:'t1',question:'q',status:'QUEUED',createdAt:'2026-11-02T01:00:00Z'},onEvent,reads=0; + const stream=deferred(),firstRead=deferred(),secondRead=deferred(); + const env={mounted:{current:true},streamController:{current:null},subscriptionGeneration:{current:0},sessionId:'s1',eventCursors:{current:{}},ANALYSIS_FINAL_STATUSES:finalStatuses, + setLastTurn:next=>lastTurn=typeof next==='function'?next(lastTurn):next,writeAnalysisLocation(){},setActiveTurnId(){},setBusy(){},setRetryQuestion(){},setTurns(){},setError(){},loadSessions:async()=>{},onUnauthorized:()=>false,apiMessage:e=>e.message, + isObjectRecord,analysisStreamStatus,analysisIsRunning,acceptAnalysisEvent,settledAnalysisTurn,collectionItems:value=>value.items||[],modelThinkingEntries:()=>[],mergeModelThinking:()=>[]}; + const api={analysisTurnEvents:async(_id,options)=>{onEvent=options.onEvent;return stream.promise;},analysisSessionTurns:()=>{reads++;return reads===1?firstRead.promise:secondRead.promise;}}; + const code=source.slice(source.indexOf(' const followTurn = async'),source.indexOf('\n useEffect(() => {\n if (!active || restoreStarted.current)')); + const follow=new Function('api',...Object.keys(env),compile(code+'\nreturn followTurn;'))(api,...Object.values(env)); + const work=follow('t1','q','s1'); + return{get:()=>lastTurn,reads:()=>reads,switchTurn(){env.subscriptionGeneration.current++;lastTurn={id:'other',status:'RUNNING'};},emit(event,data){onEvent({id:String(Math.random()),event,data});},firstRead,secondRead,async close(){env.streamController.current?.abort();stream.resolve();await work;}}; +} +test('tool step events keep the whole turn running and terminal status does not regress',()=>{ + assert.equal(analysisStreamStatus('RUNNING','tool_started',undefined),'RUNNING'); + assert.equal(analysisStreamStatus('RUNNING','tool_completed','completed'),'RUNNING'); + assert.equal(analysisStreamStatus('RUNNING','completed','COMPLETED'),'COMPLETED'); + assert.equal(analysisStreamStatus('COMPLETED','progress','RUNNING'),'COMPLETED'); +}); +test('actual event subscriber keeps status live and fills the server clock without re-entering',async()=>{ + const ui=fixture();assert.equal(ui.reads(),1); + ui.emit('progress',{status:'RUNNING'}); + ui.firstRead.resolve({items:[{id:'t1',createdAt:'2026-11-02T01:00:00Z',startedAt:''}]});await tick(); + assert.equal(ui.reads(),2,'a running event during an initial queued timing read schedules one non-overlapping follow-up'); + ui.secondRead.resolve({items:[{id:'t1',startedAt:'2026-11-02T01:00:05Z'}]});await tick(); + ui.emit('tool_started',{step:{status:'running'}}); + assert.equal(ui.get().status,'RUNNING');assert.equal(ui.get().startedAt,'2026-11-02T01:00:05Z'); + ui.emit('tool_completed',{step:{status:'completed'}});assert.equal(ui.get().status,'RUNNING'); + assert.equal(ui.reads(),2);await ui.close(); +}); +test('late timing metadata cannot update a different selected turn',async()=>{ + const ui=fixture();ui.switchTurn();await ui.close();ui.firstRead.resolve({items:[{id:'t1',startedAt:'2026-11-02T01:00:05Z'}]});await tick(); + assert.equal(ui.get().startedAt,undefined); +}); +test('metadata renders a waiting clock immediately and a spinning running clock after server start',()=>{ + const code=source.slice(source.indexOf('function AnalysisTurnMetadata('),source.indexOf('\nfunction isObjectRecord(')); + const Metadata=new Function('React','useState','useEffect','analysisIsRunning','isObjectRecord','LoadingOutlined',compile(code+'\nreturn AnalysisTurnMetadata;'))(React,initial=>[initial,()=>{}],()=>{},analysisIsRunning,isObjectRecord,props=>React.createElement('i',props)); + const now=Date.now(); + const waiting=renderToStaticMarkup(React.createElement(Metadata,{turn:{_uiSubmittedAt:new Date(now-30000).toISOString()},status:'SUBMITTING',compact:true})); + assert.match(waiting,/提交中/);assert.match(waiting,/已等待 30秒/); + const running=renderToStaticMarkup(React.createElement(Metadata,{turn:{startedAt:new Date(now-10000).toISOString()},status:'RUNNING',compact:true})); + assert.match(running,/已运行 10秒/);assert.match(running,/analysis-runtime-spinner/); +}); + +test('partial remains a backend state without an extra partial label in conversation metadata',()=>{ + const code=source.slice(source.indexOf('function AnalysisTurnMetadata('),source.indexOf('\nfunction isObjectRecord(')); + const Metadata=new Function('React','useState','useEffect','analysisIsRunning','isObjectRecord','LoadingOutlined',compile(code+'\nreturn AnalysisTurnMetadata;'))(React,initial=>[initial,()=>{}],()=>{},analysisIsRunning,isObjectRecord,()=>null); + const html=renderToStaticMarkup(React.createElement(Metadata,{turn:{durationMs:6000},status:'PARTIAL',compact:true})); + assert.doesNotMatch(html,/部分完成/);assert.match(html,/运行时长 6秒/); +}); diff --git a/web/verification/analysis_original_body.test.mjs b/web/verification/analysis_original_body.test.mjs new file mode 100644 index 0000000..8996fbb --- /dev/null +++ b/web/verification/analysis_original_body.test.mjs @@ -0,0 +1,44 @@ +import test from 'node:test'; +import assert from 'node:assert/strict'; +import { originalAnalysisBody, originalAnalysisSections } from '../src/analysisOriginalBody.ts'; +const original='## 结论\n\n合格率下降来自两个不同的阶段。\n\n### 总体与阶段\n\n先说明投入规模。\n\n| 阶段 | 变化 |\n| --- | --- |\n| 黑片 | -0.30 pp |\n\n表格之后的完整解释不能挪走。[Q11111111]\n\n### 组织维度\n\n国内与海外结构不同。\n\n### 局限\n\n不能据此认定生产根因。'; +const legacy=()=>({status:'PARTIAL',result:{executionMode:'native_agent',answer:'已取得本轮统计结果,答案整理尚未完整完成。\n\n正式统计表',readingSummary:{lead:'已取得本轮统计结果,答案整理尚未完整完成。'}},plan:{timeline:[{kind:'assistant',text:'先进行查询。'},{kind:'tool_start',sequence:1},{kind:'tool_result',sequence:2},{kind:'assistant',sequence:3,text:original}]}}); +test('legacy native fallback restores the entire final model body instead of another summary',()=>{ + const body=originalAnalysisBody(legacy());assert.equal(body.source,'timeline');assert.equal(body.answer,original); + assert.doesNotMatch(body.answer,/已取得本轮统计结果/); +}); +test('new saved original body wins over an 8000-character transcript and remains complete',()=>{ + const data=legacy();const answer='正文'.repeat(6000);data.result.answer=answer;delete data.result.readingSummary; + data.plan.timeline.at(-1).text=answer.slice(0,8000)+'…';assert.equal(originalAnalysisBody(data).answer,answer); +}); +test('a capped historical transcript cannot masquerade as the complete original body',()=>{ + const data=legacy();data.plan.timeline.at(-1).text='原'.repeat(8000); + assert.deepEqual(originalAnalysisBody(data),{answer:'',source:'unavailable'}); +}); +test('a failed native result with no formal evidence still renders its saved original body',()=>{ + assert.deepEqual(originalAnalysisBody({status:'FAILED',result:{executionMode:'native_agent',answer:original,results:[],answerQuality:{dataComplete:false}}}),{answer:original,source:'saved'}); +}); +test('earlier assistant progress, redacted answers, and unavailable originals are never promoted',()=>{ + const data=legacy();data.plan.timeline.pop();assert.deepEqual(originalAnalysisBody(data),{answer:'',source:'unavailable'}); + data.result.answer='';data.plan.timeline.push({kind:'assistant',text:original});assert.equal(originalAnalysisBody(data).answer,''); + delete data.result.executionMode;assert.equal(originalAnalysisBody(data),null); +}); +test('original Markdown lines, table placement and scope caveats stay in their exact order',()=>{ + const parts=originalAnalysisSections(original,[]);assert.equal(parts.map(p=>p.markdown).join('\n'),original); + assert.ok(parts[1].markdown.indexOf('| 黑片'){ + const visual={id:'stage',title:'阶段合格率',metric:'black_rate',type:'column',evidenceRefs:['Q11111111']}; + const sections=originalAnalysisSections(original,[visual]);assert.equal(sections[1].visuals[0],visual); + assert.equal(sections.flatMap(s=>s.visuals).length,1);assert.equal(sections.map(s=>s.markdown).join('\n'),original); +}); + +test('alternate historical evidence fallback also recovers native prose',()=>{ + const data=legacy();data.result.readingSummary.lead='现有证据不足以支持本轮完整分析,已保留可核验统计结果。';assert.equal(originalAnalysisBody(data).answer,original); +}); +test('black and bright figures cannot be attached to the opposite stage chapter',()=>{ + const body='## 黑片阶段\n\n黑片完整解释。\n\n## 亮片阶段\n\n亮片完整解释。'; + const figures=[{id:'black',type:'column',stage:'black',metric:'black_rate',evidenceRefs:[]},{id:'bright',type:'column',stage:'bright',metric:'bright_rate',evidenceRefs:[]}]; + const sections=originalAnalysisSections(body,figures);assert.equal(sections[0].visuals[0].id,'black');assert.equal(sections[1].visuals[0].id,'bright'); +}); diff --git a/web/verification/analysis_presentation_chain.test.mjs b/web/verification/analysis_presentation_chain.test.mjs index 4f9a1ea..5a99fba 100644 --- a/web/verification/analysis_presentation_chain.test.mjs +++ b/web/verification/analysis_presentation_chain.test.mjs @@ -274,17 +274,17 @@ test('history proof and lineage in model-controlled task fields cannot authorize assert.deepEqual(selectedHistoryIds(turn), []); }); -test('running process starts folded, stays open after a user opens it, closes on completion, and ignores late running status', () => { +test('running process starts open, honors manual collapse, and closes on completion', () => { const harness = hookHarness(); const turn = { id: 'process-turn', status: 'RUNNING', steps: [{ tool: 'read', label: '读取数据', status: 'running' }] }; let element = harness.render(turn); assert.equal(element.type, 'details'); - assert.equal(element.props.open, false); + assert.equal(element.props.open, true); - element.props.onToggle({ currentTarget: { open: true } }); + element.props.onToggle({ currentTarget: { open: false } }); turn.steps = [...turn.steps, { tool: 'mcp_call', label: '查询质量数据', status: 'running' }]; element = harness.render(turn); - assert.equal(element.props.open, true); + assert.equal(element.props.open, false); turn.status = 'COMPLETED'; turn.steps = [...turn.steps, { tool: 'mcp_call', label: '查询质量数据', status: 'completed' }]; @@ -293,5 +293,5 @@ test('running process starts folded, stays open after a user opens it, closes on turn.status = 'RUNNING'; element = harness.render(turn); - assert.equal(element.props.open, false); + assert.equal(element.props.open, true); }); diff --git a/web/verification/analysis_reading.test.mjs b/web/verification/analysis_reading.test.mjs index 657a64f..017fadb 100644 --- a/web/verification/analysis_reading.test.mjs +++ b/web/verification/analysis_reading.test.mjs @@ -1,3 +1,4 @@ +import { originalAnalysisBody, originalAnalysisSections } from '../src/analysisOriginalBody.ts'; import test from 'node:test'; import assert from 'node:assert/strict'; import { readFileSync } from 'node:fs'; @@ -22,7 +23,9 @@ const turn = (readingSummary = summary()) => ({ id: 'turn', question: '9月合 releaseId: 'release', visuals: [chart], queryEvidenceRefs: ['Q11111111', 'Q22222222'], readingSummary, } }); +let OriginalContent=()=>null; function component(source, name, bindings) { + bindings={originalAnalysisBody,AnalysisOriginalContent:props=>React.createElement(OriginalContent,props),...bindings}; const cleaned = source.replace(/^import .*;\n/gm, '').replace(/^export /gm, ''); return new Function('React', ...Object.keys(bindings), ts.transpileModule(`${cleaned}\nreturn ${name};`, { fileName: `${name}.tsx`, compilerOptions: { target: ts.ScriptTarget.ES2022, jsx: ts.JsxEmit.React }, @@ -30,6 +33,7 @@ function component(source, name, bindings) { } const markdownSource = readFileSync(new URL('../src/AiMarkdown.tsx', import.meta.url), 'utf8'); const AiMarkdown = component(markdownSource, 'AiMarkdown', { sanitizeAnalysisDisplayText, analysisEvidenceParts }); +OriginalContent=component(readFileSync(new URL('../src/AnalysisOriginalContent.tsx',import.meta.url),'utf8'),'AnalysisOriginalContent',{AiMarkdown,originalAnalysisSections}); test('conclusion layout separates the lead, findings and table titles without rewriting checked facts', () => { const content = '9月综合合格率**95.13%**,较8月的**95.57%**下降**0.44个百分点**。主要来自**黑片阶段**,正式影响为**-0.30个百分点**;亮片影响**-0.15个百分点**。\n\n' @@ -81,7 +85,7 @@ const visibleSummary = html => html.match(/
([\s\S] test('comparison readout uses the checked change, preserving rounded stage values and explicit scope', () => { const sentence='汽车交通2026年9月黑片合格率**97.27%**,较8月的**97.58%**下降**0.30个百分点**。'; const value=checkedConclusionComparison(sentence); - assert.equal(value.change,'0.30'); assert.equal(value.currentLabel,'汽车交通2026年9月'); + assert.equal(value.change,'-0.30'); assert.equal(value.currentLabel,'汽车交通2026年9月'); assert.equal(value.previousLabel,'8月'); assert.equal(value.current,'97.27%'); assert.equal(value.previous,'97.58%'); for (const invalid of ['9月合格率95.13%。','9月合格率95.13%,较8月95.57%。','9月合格率101.00%,较8月95.57%下降0.44个百分点。']) assert.equal(checkedConclusionComparison(invalid),null); assert.equal(checkedConclusionComparison('9月合格率0.00%,较8月95.57%下降95.57个百分点。').current,'0.00%'); @@ -93,7 +97,7 @@ test('conclusion chapters put each authorized figure beside its cited topic, wit '**重点规格**:甲规格报废增加128.45 kg。[Q11111111]']}); const figures=[{...chart,id:'department',metric:'rate_drop_pp',dimension:'department',evidenceRefs:['Q22222222']},{...chart,id:'product',metric:'scrap_delta',dimension:'product'}]; const document=buildConclusionDocument(buildAnalysisReadingModel('',[],data.question,data).reading,figures); - assert.equal(document.comparison.change,'0.44'); + assert.equal(document.comparison.change,'-0.44'); assert.deepEqual(document.sections.find(x=>x.title==='产业部变化').visuals.map(v=>v.id),['department']); assert.deepEqual(document.sections.find(x=>x.title==='重点规格').visuals.map(v=>v.id),['product']); assert.deepEqual(document.sections.flatMap(x=>x.visuals.map(v=>v.id)),['department','product']); @@ -109,7 +113,7 @@ test('reference-style cards and chapters keep chat/report values, figures and or data.result.answer=data.result.readingSummary.lead+'\n\n'+data.result.readingSummary.evidence[0]; for(const html of [renderChat(data),renderReport(data)]) { assert.match(html,/analysis-comparison-current[^>]*>2026年9月<\/span>95.13%<\/strong>/); - assert.match(html,/下降 0.44/); assert.match(html,/95.57%/); + assert.match(html,/-0.44/); assert.match(html,/95.57%/); assert.equal((html.match(//g)||[]).length,1); assert.doesNotMatch(html,/Q11111111|Q22222222|关键依据|数据依据|analysis-full-appendix/); } diff --git a/web/verification/analysis_round1.test.mjs b/web/verification/analysis_round1.test.mjs index 970340d..6bcac17 100644 --- a/web/verification/analysis_round1.test.mjs +++ b/web/verification/analysis_round1.test.mjs @@ -158,10 +158,10 @@ test('settled process is collapsed, with the original records retained inside',( assert.doesNotMatch(html,/
{ +test('running process is expanded by default while retaining its current records inside',()=>{ const html=renderToStaticMarkup(React.createElement(Process,{turn:processTurn('RUNNING')})); assert.match(html,/读取数据/);assert.match(html,/
{ @@ -329,7 +329,8 @@ test('agent process interleaves thoughts, tool calls and text with one runtime h ]}; const html=renderToStaticMarkup(React.createElement(Process,{turn,metadata:React.createElement('span',null,'运行时长 25秒')})); assert.equal((html.match(/运行时长/g)||[]).length,1); - const positions=['先核对范围','我来查询两个期间。','读取数据','再整理结果'].map(value=>html.indexOf(value)); + assert.doesNotMatch(html,/我来查询两个期间。/); + const positions=['先核对范围','读取数据','再整理结果'].map(value=>html.indexOf(value)); assert.ok(positions.every((p,i)=>p>=0&&(!i||p>positions[i-1]))); assert.equal((html.match(/class="analysis-thinking-block"/g)||[]).length,2); for(const hidden of ['model-hidden','任务理解','整理结论','analysis-thinking-heading','展示接口实际返回'])assert.ok(!html.includes(hidden),hidden); diff --git a/web/verification/analysis_submission.test.mjs b/web/verification/analysis_submission.test.mjs index 79561e9..b6adb8f 100644 --- a/web/verification/analysis_submission.test.mjs +++ b/web/verification/analysis_submission.test.mjs @@ -182,7 +182,7 @@ test('unknown 503 submission renders recovery notice without claiming a queued o await ui.submit('question with unknown outcome'); assert.equal(ui.get().attempt.uncertain,true); const html=renderDraft(ui.get().lastTurn); - assert.match(html,/提交结果暂未确认,请重试上一问恢复原任务/); + assert.doesNotMatch(html,/提交结果暂未确认/); assert.doesNotMatch(html,/排队中|正在准备分析|仍在运行/); assert.equal(ui.get().lastTurn.status,undefined); }); @@ -192,7 +192,23 @@ test('an explicit 400 rejection renders an unsuccessful draft without a running await ui.submit('rejected question'); assert.equal(ui.get().attempt,null); const html=renderDraft(ui.get().lastTurn); - assert.match(html,/本次提交未成功,请检查错误提示后重试/); + assert.doesNotMatch(html,/本次提交未成功/); assert.doesNotMatch(html,/排队中|正在准备分析|仍在运行|恢复原任务/); assert.equal(ui.get().lastTurn.status,undefined); }); + + +test('a fresh submission immediately carries waiting status and a clock before acceptance',async()=>{ + const pending=deferred();const ui=fixture({startAnalysisTurn:()=>pending.promise}); + const work=ui.submit('即时状态');assert.equal(ui.get().lastTurn.status,'SUBMITTING'); + assert.ok(Number.isFinite(Date.parse(ui.get().lastTurn._uiSubmittedAt))); + pending.resolve({turn:{id:'accepted',status:'RUNNING'}});await work; + assert.ok(Number.isFinite(Date.parse(ui.get().lastTurn._uiSubmittedAt))); +}); + +test('retry starts a fresh waiting clock without displaying the old body or plan clock',async()=>{ + const response=deferred();const ui=fixture({retryAnalysisTurn:()=>response.promise,analysisTurn:async()=>({id:'old',status:'QUEUED'})}); + const work=ui.retryTurn({id:'old',status:'PARTIAL',answer:'old text',createdAt:'2020-01-01T00:00:00Z',startedAt:'2020-01-01T00:00:01Z',plan:{startedAt:'2020-01-01T00:00:01Z'}}); + assert.equal(ui.get().lastTurn.status,'SUBMITTING');assert.equal(ui.get().lastTurn.answer,undefined);assert.equal(ui.get().lastTurn.plan,undefined); + response.resolve({turn:{id:'old',sessionId:'session-1',status:'QUEUED'}});await work; +}); diff --git a/web/verification/analysis_visuals.test.mjs b/web/verification/analysis_visuals.test.mjs index e4806aa..1f03ca8 100644 --- a/web/verification/analysis_visuals.test.mjs +++ b/web/verification/analysis_visuals.test.mjs @@ -15,6 +15,7 @@ import { visualUnitLabel, } from '../src/analysisVisualsModel.ts'; import { buildAnalysisNarrative } from '../src/analysisReportModel.ts'; +import { buildConclusionDocument, checkedConclusionComparison } from '../src/analysisConclusionModel.ts'; function visual(overrides = {}) { return { @@ -305,6 +306,64 @@ test('negative percentage point changes remain signed and keep a zero baseline', assert.equal(formatVisualValue(0.5, 'pp'), '+0.5 pp'); }); +test('legacy rate-drop charts normalize once and chat tables and exports use current minus prior', () => { + const legacy = visual({ + id: 'legacy-rate-drop', type: 'bar', title: '业务分组合格率降幅线索', subtitle: '8月 → 9月', + unit: 'pp', metric: 'rate_drop_pp', dimension: 'group', sectionKey: 'group', + categories: [{ key: 'g1', label: '分组一' }, { key: 'g2', label: '分组二' }], + series: [{ key: 'rate_drop_pp', label: '合格率变化(pp)', values: [2.5, -4] }], + evidenceRefs: ['evidence-1'], requiredFields: ['black_good_weight', 'black_total_weight'], + scope: { department: '工业电机' }, period: { label: '2026-09' }, comparisonPeriod: { label: '2026-08' }, + coverage: { totalRows: 2, returnedRows: 2, truncated: false }, + note: '条形末端显示具体变化:正值为下降,负值为改善,单位是百分点。本图只按率降幅排序,未纳入投入规模。', + }); + const model = parseAnalysisVisuals([legacy], 1).visuals[0]; + assert.deepEqual(model.series[0].values, [-2.5, 4]); + assert.match(model.note, /本期减比较期/); + assert.match(model.note, /仍按率降幅排序|只按率降幅排序/); + assert.deepEqual(parseAnalysisVisuals([model], 1).visuals[0].series[0].values, [-2.5, 4]); + + const legacyNotes = [ + '正值表示合格率下降,负值表示改善;百分点取自正式分组率差。', + '正值表示综合合格率下降,负值表示改善;按正式综合率比较,不作为生产根因。', + undefined, + ]; + for (const [index, note] of legacyNotes.entries()) { + const snapshot = { ...legacy, id: `legacy-rate-drop-${index}`, note }; + const normalized = parseAnalysisVisuals([snapshot], 1).visuals[0]; + assert.deepEqual(normalized.series[0].values, [-2.5, 4]); + assert.match(normalized.note, /本期减比较期/); + assert.deepEqual(parseAnalysisVisuals([normalized], 1).visuals[0].series[0].values, [-2.5, 4]); + } + + const chart = buildAnalysisVisualOption(model); + assert.deepEqual(chart.series[0].data, [-2.5, 4]); + const table = renderVisualTableHtml(legacy); + assert.match(table, /-2\.5 pp/); + assert.match(table, /\+4 pp/); + const exported = buildAnalysisVisualExportOption(model, 1280, 'report'); + assert.deepEqual(exported.option.series[0].data, [-2.5, 4]); + + const modern = { ...legacy, id: 'current-rate-drop', note: '正值表示本期合格率上升,负值表示下降;变化值为本期减比较期。', + series: [{ ...legacy.series[0], values: [-2.5, 4] }] }; + assert.deepEqual(parseAnalysisVisuals([modern], 1).visuals[0].series[0].values, [-2.5, 4]); + + const reading = { lead: '分组变化', evidence: [], notices: [], evidenceRefs: ['evidence-1'], + hasAppendix: false, fullAnswer: '### 业务分组变化\n\n合格率有变化。[evidence-1]', visualCount: 1 }; + const conclusion = buildConclusionDocument(reading, [model]); + assert.deepEqual(conclusion.sections.find(section => section.title === '业务分组变化').visuals[0].series[0].values, [-2.5, 4]); +}); + +test('checked conclusion comparisons give declines negative pp and increases positive pp', () => { + const down = checkedConclusionComparison('9月综合合格率95.13%,较8月的95.57%下降0.44个百分点。'); + const up = checkedConclusionComparison('9月综合合格率95.57%,较8月的95.13%上升0.44个百分点。'); + assert.equal(down.change, '-0.44'); + assert.equal(up.change, '+0.44'); + assert.equal(down.direction, '下降'); + assert.equal(up.direction, '上升'); + assert.equal(checkedConclusionComparison('9月综合合格率95.13%,较8月的95.57%上升0.44个百分点。'), null); +}); + test('non-finite and out-of-range ratio values degrade to a table while retaining zero and null', () => { const parsed = parseAnalysisVisuals([visual({ series: [{ key: 'r', label: '率', values: [0, Number.NaN] }] })]); assert.equal(parsed.visuals.length, 1);