diff --git a/apps/presentation/dashboard/smoke/research-analysis-read-smoke.mjs b/apps/presentation/dashboard/smoke/research-analysis-read-smoke.mjs index edc67d2be8..601948ebba 100644 --- a/apps/presentation/dashboard/smoke/research-analysis-read-smoke.mjs +++ b/apps/presentation/dashboard/smoke/research-analysis-read-smoke.mjs @@ -9,6 +9,33 @@ import {createElement} from "react"; import {renderToStaticMarkup} from "react-dom/server"; import {resolveTestPython} from "../../../../scripts/test-python.mjs"; import {readAnalysisSnapshot, ResearchAnalysisPanel} from "../src/features/personal-workspace/goal-research-analysis.tsx"; +import {ResearchCaseStudy} from "../src/features/personal-workspace/research-case-study.tsx"; +import {ResearchGoalNavigation} from "../src/features/personal-workspace/goal-research-workbench.tsx"; + +// A case belongs to a source-scoped child Goal. Parent navigation must never +// combine private-source siblings or turn organization into execution. +const caseFixture = JSON.parse(await readFile(resolve('src/data/fixtures/case-study-child.json'), 'utf8')); +const study = caseFixture.analyses[0].case_study; +const renderedStudy = renderToStaticMarkup(createElement(ResearchCaseStudy, { + study, zh: false, historical: true, scope: caseFixture.analyses[0].input.scope, onCitation: () => {throw Error('read executed a citation');}, +})); +assert.ok(renderedStudy.includes(study.title)); +assert.ok(renderedStudy.includes('Showing the saved Chinese original.')); +assert.ok(renderedStudy.includes('not current results')); +assert.ok(renderedStudy.includes('unknown'), 'Missing values do not become zero'); +for (const section of study.sections) assert.ok(renderedStudy.includes(section.title)); +const goals = [ + {goalId:'parent',title:'Benchmark'}, + {goalId:'child',parentGoalId:'parent',title:'Local case'}, + {goalId:'foreign',parentGoalId:'parent',title:'Foreign case',sourceAccess:{namespace:'other',label:'Other',readOnly:true}}, +]; +const navigate = goalId => renderToStaticMarkup(createElement(ResearchGoalNavigation, { + goalId, goals, zh:false, onSelect: () => {throw Error('read navigated');}, +})); +assert.ok(navigate('parent').includes('Local case')); +assert.ok(!navigate('parent').includes('Foreign case')); +assert.ok(navigate('child').includes('Benchmark')); +assert.equal(navigate('foreign'), '', 'An absent same-source parent is not a navigation grant'); // A language-specific cache miss must not hide the same registered pack's // older, unspecified-language result or turn a read into model execution. diff --git a/apps/presentation/dashboard/src/data/fixtures/case-study-child.json b/apps/presentation/dashboard/src/data/fixtures/case-study-child.json new file mode 100644 index 0000000000..d736ab7f91 --- /dev/null +++ b/apps/presentation/dashboard/src/data/fixtures/case-study-child.json @@ -0,0 +1,165 @@ +{ + "goal_id": "synthetic-case-search", + "parent_goal_id": "synthetic-benchmark", + "display_name": "Search · Case 研究", + "events": [], + "analyses": [ + { + "title": "初版解读", + "input": { + "schema_version": "trajectory_assessment_input_v1", + "scope": { + "goal_id": "synthetic-solver", + "case_id": "Search", + "run_id": "synthetic-run", + "revision": "fixed-synthetic-evidence", + "window": "Three synthetic observations" + }, + "objective": "Compare valid error measurements without inventing retention or causality.", + "selection": { + "rule": "Complete synthetic illustration", + "population_episodes": 1, + "exhaustive": true + }, + "sources": [ + { + "id": "facts", + "artifact": "facts.txt", + "artifact_sha256": "2dec56af0cdca4e603227b1a42b56d15c2ccec3ad4973bd2b34d2d94122aeb91", + "first_line": 1, + "text": "Synthetic candidate A measured error 7 points.\nSynthetic candidate B was invalid and has no score.\nSynthetic candidate C measured error 3 points.\nRetention and matched treatment controls were not captured." + } + ], + "episodes": [ + { + "id": "search", + "question": "What changed and what is still unknown?", + "source_ids": [ + "facts" + ], + "eligible_metrics": [ + "benefit_test" + ] + } + ] + }, + "artifacts": { + "facts.txt": "Synthetic candidate A measured error 7 points.\nSynthetic candidate B was invalid and has no score.\nSynthetic candidate C measured error 3 points.\nRetention and matched treatment controls were not captured." + }, + "case_study": { + "title": "搜索示例：误差下降了，是否真正采用仍未知", + "language": "zh", + "summary": "合成观测中，后续候选把误差从 **7 分降至 3 分**，中间还有一次无效候选。这个变化值得研究，但不能直接称作最终交付，更不能当成 LoopX 的因果收益。\n\n理解本案需要分开三个问题：怎样算更好，测到了什么，以及改善是否真的留在最终成果里。下面按这个顺序解释，原始事实可在每一节查阅。", + "citations": [ + { + "source_id": "facts", + "first_line": 1, + "last_line": 4 + } + ], + "sections": [ + { + "id": "problem", + "title": "先理解题目：什么才算一次有效改善", + "text": "本例用误差衡量候选结果与目标的距离，因此数值越低越好。有效性先于数值比较：一次未通过有效性检查的尝试没有可比较的分数，不能把它补成零。\n\n**本页全部数据都是合成的展示样本。** 它用于验证负例、未知状态和证据阅读，不代表真实 benchmark 的结果。后面的最佳值也只描述这个固定窗口，不代表窗口外的历史。", + "citations": [ + { + "source_id": "facts", + "first_line": 1, + "last_line": 4 + } + ], + "table": { + "columns": [ + "概念", + "怎样理解", + "本例中可说什么" + ], + "rows": [ + [ + "误差", + "偏离目标的程度；更低更好", + "候选 C 的 3 分低于 A 的 7 分" + ], + [ + "无效结果", + "没有通过资格检查", + "候选 B 没有分数，不能当作零" + ], + [ + "最终采用", + "改善被保留在实际成果中", + "未捕获保留证据，仍未知" + ] + ] + } + }, + { + "id": "trajectory", + "title": "过程与结果：看到进展，也保留空白", + "text": "只看第一轮和最后一轮，会漏掉中间的无效尝试。把三次观测放回顺序中，可以确认后续出现了更低误差，但看不出它为何出现。\n\n这里不连接缺失点，也不从提交序号推导时间成本。没有耗时记录，就不能把一次更好的结果进一步解释成效率提高。", + "citations": [ + { + "source_id": "facts", + "first_line": 1, + "last_line": 3 + } + ], + "series": { + "label": "固定窗口内的观测", + "unit": "误差分 · 更低更好", + "points": [ + { + "label": "候选 A", + "value": 7 + }, + { + "label": "候选 B", + "value": null + }, + { + "label": "候选 C", + "value": 3 + } + ] + }, + "timeline": [ + { + "label": "第一次", + "title": "有了可比较的起点", + "text": "候选 A 的误差为 7 分。" + }, + { + "label": "第二次", + "title": "一次失败不能从故事中消失", + "text": "候选 B 无效，没有分数。失败仍是过程的一部分。" + }, + { + "label": "第三次", + "title": "更低误差出现，但保留未明", + "text": "候选 C 的误差为 3 分；最终采用仍需另一份证据。" + } + ] + }, + { + "id": "limits", + "title": "关于系统强弱，证据支持到哪一步", + "text": "这条记录支持“窗口内出现过更好的有效候选”。它还不能回答是哪种机制促成改善：模型本身、额外搜索、工具反馈或结构化状态，都可能影响结果。\n\n如果没有相同条件下的对照，便无法把全部变化归于某一个系统。类似地，没有最终成果的保留记录，也无法把最佳观测值称为交付值。\n\n这些信息共同界定了判断的范围：**有进展的观测，有无效尝试，也有尚未回答的问题。**", + "citations": [ + { + "source_id": "facts", + "first_line": 4, + "last_line": 4 + } + ] + } + ], + "limitations": [ + "本页是合成示例；不作为科研结论或产品效果证据。", + "未记录保留状态、耗时或匹配对照，不能补造这些事实。" + ] + }, + "outputs": {} + } + ] +} diff --git a/apps/presentation/dashboard/src/data/fixtures/case-study-empty-child.json b/apps/presentation/dashboard/src/data/fixtures/case-study-empty-child.json new file mode 100644 index 0000000000..0ec25a1191 --- /dev/null +++ b/apps/presentation/dashboard/src/data/fixtures/case-study-empty-child.json @@ -0,0 +1,6 @@ +{ + "goal_id": "synthetic-case-unread", + "parent_goal_id": "synthetic-benchmark", + "display_name": "另一 Case · 尚无解读", + "events": [] +} diff --git a/apps/presentation/dashboard/src/data/fixtures/case-study-parent.json b/apps/presentation/dashboard/src/data/fixtures/case-study-parent.json new file mode 100644 index 0000000000..9182485333 --- /dev/null +++ b/apps/presentation/dashboard/src/data/fixtures/case-study-parent.json @@ -0,0 +1,5 @@ +{ + "goal_id": "synthetic-benchmark", + "display_name": "Benchmark · 研究示例", + "events": [] +} diff --git a/apps/presentation/dashboard/src/features/personal-workspace/goal-research-analysis.tsx b/apps/presentation/dashboard/src/features/personal-workspace/goal-research-analysis.tsx index 2b47843827..211cb001b7 100644 --- a/apps/presentation/dashboard/src/features/personal-workspace/goal-research-analysis.tsx +++ b/apps/presentation/dashboard/src/features/personal-workspace/goal-research-analysis.tsx @@ -7,6 +7,8 @@ import {requestJson} from "../../data/chat"; import {sourceReadApiUrl} from "../../data/local-status-query"; import {ContextHelp} from "./context-help"; import {ResearchSynthesis} from "./goal-research-synthesis"; +import {ResearchCaseStudy} from "./research-case-study"; +import type {TrajectoryCaseStudy} from "../../../../../../loopx/control_plane/capabilities/trajectory_case_study.js"; import "./goal-research-analysis.css"; // Transport views of the existing toolkit producer, not an assessment, score, @@ -16,6 +18,8 @@ type Job = {status: string; attempt?: number; analysis_wall_seconds?: number | n config?: {model: string; effort: string; service_tier: string}; served_service_tier?: string | null; usage?: {input_tokens?: number; output_tokens?: number; reasoning_output_tokens?: number} | null}; type Pack = ResearchProgressPack & {source_count: number; analysis_key: string; method_version: string; + case_study?: TrajectoryCaseStudy; + case_study_summary?: {title: string; language: TrajectoryOutputLanguage}; display_request?: {schema_version: 'research_display_v1'; snapshot_key: string; source_id: string; source_namespace: string; eligible: number}; result_attempt?: number; result_job?: Job; @@ -346,6 +350,26 @@ export function ResearchReportSummary({analysis: a}: {analysis: Controller}) { onExpand={origin => {a.trigger.current = origin; a.show();}} onCitation={a.readCitation}/> : null; } +export function ResearchCaseReading({analysis: a, onHistory}: {analysis: Controller; onHistory: () => void}) { + const packs = (a.snapshot?.catalog.packs ?? []).filter(p => p.case_study_summary); + const p = a.pack; + const firstStudy = packs[0]?.pack_id; + useEffect(() => { + if (!a.loading && !a.readError && p && !p.case_study && firstStudy) a.select(firstStudy); + }, [a.loading, a.readError, p?.pack_id, Boolean(p?.case_study), firstStudy]); + return
+
+ + {p?.case_study ? : !a.loading && !a.readError ?

{a.zh ? "选择一份解读，阅读问题背景、策略差异、关键转折与证据。" : "Choose a study to read its background, approaches, turning points and evidence."}

: null} + +
; +} + function CitedToolText({text, zh}: {text: string; zh: boolean}) { // Decode only the saved direct shell-result envelope, never recursively // reinterpret arbitrary JSON or infer successful work from its text. diff --git a/apps/presentation/dashboard/src/features/personal-workspace/goal-research-history.tsx b/apps/presentation/dashboard/src/features/personal-workspace/goal-research-history.tsx index 3ba8c18b2b..b69b0e27ca 100644 --- a/apps/presentation/dashboard/src/features/personal-workspace/goal-research-history.tsx +++ b/apps/presentation/dashboard/src/features/personal-workspace/goal-research-history.tsx @@ -4,7 +4,7 @@ import {AlertCircle, ArrowLeft, ArrowRight, Circle, CircleHelp, FlaskConical, Gi import {Fragment, useEffect, useId, useLayoutEffect, useMemo, useRef, useState} from "react"; import {defaultParseSearch} from "@tanstack/react-router"; import {ResearchRecordSearch} from "./research-record-browser"; -import {ResearchAnalysisButton, ResearchAnalysisPanel, ResearchAnalysisStatus, ResearchCitationPanel, ResearchProgressSummary, ResearchReportSummary, useResearchAnalysis} from "./goal-research-analysis"; +import {ResearchAnalysisButton, ResearchAnalysisPanel, ResearchAnalysisStatus, ResearchCitationPanel, ResearchProgressSummary, ResearchReportSummary, ResearchCaseReading, useResearchAnalysis} from "./goal-research-analysis"; import {buildExploreTimeline, parseExploreFindingSections, type TimelineFinding} from "../../../../../../loopx/control_plane/presentation/explore_timeline.js"; import {GoalResearchTimeline} from "./goal-research-timeline"; import {requestJson} from "../../data/chat"; @@ -54,6 +54,22 @@ const viewFromUrl = (goal: string) => { * a judge. All recorded updates survive; unsupported analysis stays unavailable. */ export function GoalResearchHistory({goalId, active, zh, refreshRevision = 0, readSourceUrl}: {goalId: string; active: boolean; zh: boolean; refreshRevision?: number; readSourceUrl?: string}) { const analysis = useResearchAnalysis(goalId, active, zh, readSourceUrl); + const readCaseMode = () => { + const query = new URLSearchParams(window.location.search); + return query.get("research_goal") !== goalId || query.get("research_reading") !== "history"; + }; + const [caseReading, setCaseReading] = useState(readCaseMode); + useEffect(() => { + const changed = () => setCaseReading(readCaseMode()); + changed(); window.addEventListener("popstate", changed); + return () => window.removeEventListener("popstate", changed); + }, [goalId, readSourceUrl]); + const setCaseMode = (reading: boolean) => { + const url = new URL(window.location.href); + url.searchParams.set("research_goal", goalId); + url.searchParams.set("research_reading", reading ? "case" : "history"); + window.history.pushState({}, "", url); setCaseReading(reading); + }; const [read, setRead] = useState({events: [], total: null, complete: false}); const [originalDisplay, setOriginalDisplay] = useState(false); const [displayIdentity, setDisplayIdentity] = useState<{events: ExploreHistoryEvent[]; namespace: string; key: string} | null>(null); @@ -449,7 +465,10 @@ export function GoalResearchHistory({goalId, active, zh, refreshRevision = 0, re const emptyHistory = !busy && !readFailure && read.total === 0; const emptyWithAnalysis = emptyHistory && Boolean(analysis.pack?.result || analysis.pack?.native_outcomes); const emptyHistoryMessage = zh ? "这个 Goal 尚无研究历程" : "No research history for this Goal yet"; + const hasCases = (analysis.snapshot?.catalog.packs ?? []).some(p => p.case_study_summary); + if (hasCases && caseReading) return setCaseMode(false)}/>; return
+ {hasCases ? : null} {progressiveOverview ? <>