feat(research): read native outcomes and costs from bound evidence

apps/presentation/dashboard/src/features/personal-workspace/goal-research-analysis.tsx

@@ -219,11 +219,12 @@ export function ResearchAnalysisPanel({analysis: a}: {analysis: Controller}) {
           const label = metricCopy[row.metric];
           return <details key={row.metric}><summary><strong>{label?.[zh ? 0 : 1] ?? row.metric}</strong><span>{verdictLabel(row.metric, row.verdict, zh)}</span></summary><p className="analysis-question">{label?.[zh ? 2 : 3]}</p><p>{row.reason}</p>{citationButtons(row.citations)}</details>;
         })}</div></details></section>)}
-        {a.citationBusy ? <p role="status">{zh ? "正在读取引用原文…" : "Reading cited source…"}</p> : null}
-        {a.citationError ? <p className="analysis-error" role="alert">{a.citationError}</p> : null}
-        {a.citation ? <section className="analysis-source" ref={a.sourcePanel} tabIndex={-1} aria-label={zh ? "引用原文" : "Cited original"} onKeyDown={event => {if (event.key === "Escape") {event.stopPropagation(); a.closeCitation();}}}><header><h4>{zh ? "引用原文" : "Cited original"}</h4><button type="button" onClick={a.closeCitation} aria-label={zh ? "关闭原文" : "Close source"}><X size={14} aria-hidden/></button></header><p>{a.citation.artifact} · {zh ? "行" : "lines"} {a.citation.first_line}–{a.citation.last_line}</p><pre>{a.citation.text}</pre></section> : null}
         <details className="analysis-method"><summary>{zh ? "来源、方法、覆盖与分析成本" : "Sources, method, coverage and analysis cost"}</summary><p>{p.objective}</p><dl><dt>{zh ? "来源范围" : "Source scope"}</dt><dd>{p.scope.case_id} · {p.scope.run_id}<br/>{p.scope.window}</dd><dt>{zh ? "方法" : "Method"}</dt><dd>{result.method_version} · {zh ? "模型判断未校准" : "Uncalibrated model judgment"}</dd><dt>{zh ? "模型设置" : "Requested model"}</dt><dd>{p.config.model} / {p.config.effort} / {p.config.service_tier} · {zh ? "实际服务档位" : "Served tier"}: {p.job?.served_service_tier ?? (zh ? "未知" : "unknown")}</dd><dt>{zh ? "分析用量" : "Analysis usage"}</dt><dd>{p.job?.analysis_wall_seconds == null ? "—" : `${p.job.analysis_wall_seconds.toFixed(1)}s`} · {zh ? "输入" : "input"} {p.job?.usage?.input_tokens?.toLocaleString() ?? "—"} / {zh ? "输出" : "output"} {p.job?.usage?.output_tokens?.toLocaleString() ?? "—"} tokens · {zh ? "金额未采集，推理 token 已包含于输出" : "Dollar cost unmeasured; reasoning tokens are included in output"}</dd><dt>{zh ? "覆盖" : "Coverage"}</dt><dd>{p.episode_count} {zh ? "个固定片段" : "fixed episodes"} · {p.source_count} {zh ? "份来源" : "sources"} · {zh ? "总体片段数" : "Population"}: {p.selection.population_episodes ?? (zh ? "未知" : "unknown")}</dd><dt>{zh ? "取样规则" : "Selection"}</dt><dd>{p.selection.rule}</dd><dt>{zh ? "版本" : "Revision"}</dt><dd>{p.scope.revision}</dd></dl><ul>{result.limitations.map((text, i) => <li key={i}>{text}</li>)}</ul><p>{zh ? "引用行存在不等于解释正确；本页不提供独立科学校准结论。" : "A valid citation range does not establish entailment or independent scientific calibration."}</p></details>
       </> : null}
+      {/* Bound native evidence is readable before any model assessment exists. */}
+      {a.citationBusy ? <p role="status">{zh ? "正在读取引用原文…" : "Reading cited source…"}</p> : null}
+      {a.citationError ? <p className="analysis-error" role="alert">{a.citationError}</p> : null}
+      {a.citation ? <section className="analysis-source" ref={a.sourcePanel} tabIndex={-1} aria-label={zh ? "引用原文" : "Cited original"} onKeyDown={event => {if (event.key === "Escape") {event.stopPropagation(); a.closeCitation();}}}><header><h4>{zh ? "引用原文" : "Cited original"}</h4><button type="button" onClick={a.closeCitation} aria-label={zh ? "关闭原文" : "Close source"}><X size={14} aria-hidden/></button></header><p>{a.citation.artifact} · {zh ? "行" : "lines"} {a.citation.first_line}–{a.citation.last_line}</p><pre>{a.citation.text}</pre></section> : null}
     </> : null}
   </div>;
 }

apps/presentation/dashboard/src/features/personal-workspace/goal-research-progress.tsx

@@ -1,13 +1,17 @@
 import {ArrowUpRight, ChevronDown} from "lucide-react";
 import {TRAJECTORY_METRICS} from "../../../../../../loopx/control_plane/capabilities/trajectory_assessment_types.js";
 import type {AssessmentCitation, AssessmentVerdict, TrajectoryAssessment, TrajectoryEpisode, TrajectoryInput, TrajectoryMetric, TrajectorySolverUsage} from "../../../../../../loopx/control_plane/capabilities/trajectory_assessment_types.js";
+import type {TrajectoryNativeOutcomes} from "../../../../../../loopx/control_plane/capabilities/trajectory_native_outcomes.js";
 import "./goal-research-progress.css";
 
 // Read models from benchmark-toolkit. This view formats existing counts and
 // assessments; it never scores, reclassifies an episode or combines runs.
 export type ResearchProgressPack = {pack_id: string; title: string; scope: TrajectoryInput["scope"]; selection: TrajectoryInput["selection"];
   episode_count: number; episodes?: TrajectoryEpisode[]; result?: TrajectoryAssessment | null; applicability?: "current" | "historical";
-  solver_usage?: {artifact: string; artifact_sha256: string; projection: TrajectorySolverUsage; applicability: "current" | "historical"; reason_code: string | null}};
+  job?: {status: string; attempt?: number; analysis_wall_seconds?: number | null;
+    usage?: {input_tokens?: number; output_tokens?: number} | null; config?: {model: string; effort: string; service_tier: string}} | null;
+  solver_usage?: {artifact: string; artifact_sha256: string; projection: TrajectorySolverUsage; applicability: "current" | "historical"; reason_code: string | null};
+  native_outcomes?: {artifact: string; artifact_sha256: string; projection: TrajectoryNativeOutcomes; applicability: "current" | "historical"; reason_code: string | null}};
 export const metricCopy: Record<TrajectoryMetric, [string, string, string, string]> = {
   goal_alignment: ["目标关联", "Goal alignment", "干预是否关联原目标的真实瓶颈", "Does the intervention address a bottleneck of the original objective?"],
   benefit_test: ["收益检验", "Benefit test", "是否真正检验原目标收益，而非只完成前置诊断", "Was original-objective utility tested, beyond diagnostic prerequisites?"],
@@ -25,6 +29,7 @@ export function verdictLabel(metric: TrajectoryMetric, verdict: AssessmentVerdic
 
 export function ResearchProgress({pack: p, packs, selectedId, error, zh, onSelect, onCitation}: {pack: ResearchProgressPack | null; packs: Pick<ResearchProgressPack, "pack_id" | "title">[]; selectedId: string; error: string; zh: boolean; onSelect: (id: string) => void; onCitation: (citation: AssessmentCitation, origin?: HTMLButtonElement) => void}) {
   const measured = p?.solver_usage, usage = measured?.projection, result = p?.result;
+  const native = p?.native_outcomes, outcomes = native?.projection;
   const number = (n: number) => new Intl.NumberFormat(zh ? "zh-CN" : "en", {maximumFractionDigits: 1}).format(n);
   const percent = (n: number) => new Intl.NumberFormat(zh ? "zh-CN" : "en", {style: "percent", maximumFractionDigits: 0}).format(n);
   const duration = (s: number) => `${number(s / 60)} ${zh ? "分钟" : "min"}`;
@@ -67,10 +72,28 @@ export function ResearchProgress({pack: p, packs, selectedId, error, zh, onSelec
         <dt>{zh ? "题目 / 运行" : "Case / run"}</dt><dd>{p.scope.case_id}<br/>{p.scope.run_id}</dd><dt>{zh ? "证据窗口" : "Evidence window"}</dt><dd>{p.scope.window}</dd><dt>{zh ? "取样规则" : "Selection rule"}</dt><dd>{p.selection.rule}</dd>
         <dt>{zh ? "总体研究片段数" : "Population of episodes"}</dt><dd>{p.selection.population_episodes == null ? (zh ? "未知" : "Unknown") : number(p.selection.population_episodes)}</dd>
         {result ? <><dt>{zh ? "分析方法" : "Analysis method"}</dt><dd>{result.method_version} · {zh ? "Agent 判读未校准" : "Uncalibrated Agent judgment"}</dd><dt>{zh ? "结果标识" : "Result key"}</dt><dd><code>{result.analysis_key}</code></dd><dt>{zh ? "分析限制" : "Limitations"}</dt><dd>{result.limitations.map((line, i) => <p key={i}>{line}</p>)}</dd></> : null}
-        <dt>{zh ? "来源版本" : "Source revision"}</dt><dd><code>{p.scope.revision}</code></dd><dt>{zh ? "原生收益" : "Native utility"}</dt><dd>{zh ? "此包尚未接入原生起点、保留最佳、最后有效值、达标时间/成本或增益面积投影。不能从模型判读或成本补算。" : "This pack has no native incumbent, retained-best, final-valid, threshold time/cost or gain-area projection. Model judgments and usage cannot supply them."}</dd>
+        <dt>{zh ? "来源版本" : "Source revision"}</dt><dd><code>{p.scope.revision}</code></dd>
+        <dt>{zh ? "原生结果" : "Native outcomes"}</dt><dd>{outcomes ? <>
+          <p>{outcomes.metric.name} · {outcomes.metric.unit} · {outcomes.metric.higher_is_better ? (zh ? "越高越好" : "higher is better") : (zh ? "越低越好" : "lower is better")}</p>
+          <p>{outcomes.start} — {outcomes.end} · {outcomes.complete ? (zh ? "声明为完整前缀" : "Declared complete prefix") : (zh ? "部分窗口，起点未知" : "Partial window; initial unknown")} · {outcomes.valid_count} / {outcomes.sample_count} {zh ? "项有效结果" : "valid results"}</p>
+          <dl>{([ ["initial", "首次提交", "Initial submission"], ["best_valid", "窗口内最佳有效值", "Best valid in capture"], ["retained", "原生保留结果", "Native retained result"], ["final_valid", "最后有效提交", "Final valid submission"] ] as const).map(([key, cn, en]) => {
+            const point = outcomes[key];
+            return <div key={key}><dt>{zh ? cn : en}</dt><dd>{point ? <>{String(point.value)} · {point.candidate_ref} <button type="button" onClick={event => onCitation(point.citation, event.currentTarget)}>{zh ? "查看原始证据" : "Read original evidence"}</button></> : (zh ? "— 未知，不记零" : "— Unknown, not zero")}</dd></div>;
+          })}</dl>
+          {native?.applicability === "historical" ? <p role="status">{zh ? "原生来源已变化或不可读；仅保留历史捕获，当前结果未核验。" : "Native source changed or is unavailable; retained historical capture, current results unverified."}</p> : null}
+          <p>{zh ? "原生测量与分析器判断分开。有效性和保留选择来自提供方，不认证研究因果收益；用量窗口独立，不补算达标成本或全程账单。" : "Native measurements are separate from analyzer judgments. Validity and retention are provider-recorded, not causal benefit. Usage has its own window; threshold and total-study costs are not inferred."}</p>
+        </> : (zh ? "— 尚无同范围的原生结果捕获；不能从分析器判断或用量补算。" : "— No native capture for this scope; analyzer judgments and usage cannot supply it.")}</dd>
       </dl></details>
       <details className="research-usage"><summary>{zh ? "用量与成本" : "Usage and cost"}{usage ? ` · ${zh ? "捕获跨度" : "Captured elapsed"} ${duration(usage.elapsed_seconds)}` : ` · ${zh ? "尚无完整用量捕获" : "No complete usage capture"}`}{measured?.applicability === "historical" ? (zh ? " · 历史捕获" : " · Historical capture") : ""}</summary>
         <p>{zh ? "只读捕获，窗口可能与以上判读片段及下方 Explore 历史不同。" : "Read-only capture; its window can differ from the judgments above and Explore history below."}</p>
+        <dl className="research-usage-basis">
+          <dt>{zh ? "分析尝试开销" : "Analysis attempt cost"}</dt><dd>{p.job?.usage ? <>
+            {zh ? "第" : "Attempt"} {p.job.attempt ?? "—"} · {zh ? "输入" : "input"} {p.job.usage.input_tokens == null ? "—" : number(p.job.usage.input_tokens)} · {zh ? "输出" : "output"} {p.job.usage.output_tokens == null ? "—" : number(p.job.usage.output_tokens)} tokens
+            {p.job.config ? ` · ${p.job.config.model}` : ""}
+            {p.job.analysis_wall_seconds != null ? ` · ${duration(p.job.analysis_wall_seconds)}` : ""}
+          </> : (zh ? "— 本次分析尝试未测得，不记零" : "— No measurement for this analysis attempt; not zero")}</dd>
+          <dt>{zh ? "合计边界" : "Aggregation boundary"}</dt><dd>{zh ? "分析仅显示当前回执的本次尝试；不含更早失败、恢复、评测或操作者开销。不与求解捕获盲目相加，金额及完整研究成本仍未知。" : "Analysis is this attempt's receipt only; earlier failures, recovery, evaluation and operator work are excluded. It is not blindly added to solver capture. Monetary and complete study costs remain unknown."}</dd>
+        </dl>
         <dl className="research-progress-values">
           <div><dt>{zh ? "捕获窗口跨度" : "Captured elapsed time"}</dt><dd>{usage ? duration(usage.elapsed_seconds) : "—"}</dd><small>{usage ? `${number(usage.completed_turns)} ${zh ? "轮完整记录，不是全程总成本" : "complete Turns, not total study cost"}` : (zh ? "尚无完整用量捕获" : "No complete usage capture")}</small></div>
           <div><dt>{zh ? "未缓存输入" : "Uncached input"}</dt><dd>{usage ? number(usage.uncached_input_tokens) : "—"}</dd><small>{usage ? "tokens" : (zh ? "未测，不记零" : "Unmeasured, not zero")}</small></div>

docs/architecture/rfcs/goal-workspace-surface-v0.md

@@ -732,6 +732,16 @@ independent scientific calibration or completion of V12. Host registration
 remains preparation outside the ordinary reading journey. CLI preparation uses
 the producer's existing interface; no new interactive Lark path is qualified.
 
+The follow-up native-readback proposal extends that same trusted registration
+with an optional provider outcome capture, scoped to the exact case/run/revision.
+The existing source disclosure separates first submission, best valid in capture,
+native-retained and final valid values, with original citations and historical/
+partial/unknown states. Submission order, not feedback arrival, identifies the
+candidate. This is not another scorer or automatic solver feedback. Captured
+solver usage and the current analysis attempt's receipt stay separate; full-study
+cost and causal benefit remain unknown. No first-viewport/hero/CTA change is
+introduced; six uncalibrated process judgments retain their existing role.
+
 ## 12. Open decisions
 
 | # | Decision | Owner | Options | Recommendation | Needed before |

docs/architecture/rfcs/goal-workspace-surface-v0.zh-CN.md

@@ -520,6 +520,12 @@ producer 接入一次显式分析、取消、重试和有引用的结果阅读
 任意 run 自动准备、独立科学校准或 V12 完成。host 注册仍是普通阅读旅程之外的准备，
 CLI 准备复用 producer 既有接口；本切片未验收新增飞书交互入口。
 
+后续原生读回提案在同一可信注册入口增加可选 provider 结果捕获，固定题目、运行和
+来源版本。既有来源展开分别展示首次提交、捕获内最佳有效值、原生保留和最后有效提交，
+含原文引用及历史/部分/未知状态；候选按提交顺序而非反馈到达排序。不新增评分器或
+自动 solver 回灌。求解捕获与本次分析尝试的回执开销分别保留，完整研究成本及因果收益
+仍未知；首屏、hero、CTA 不变，六项未校准的过程判断维持原有角色。
+
 ## 12. 未决事项
 
 | # | 事项 | Owner | 选项 | 建议 | 需在何时前决定 |

docs/reference/trajectory-assessment.md

@@ -192,6 +192,77 @@ journey or qualify the model assessor.
 
 ## Measured solver session cost (optional)
 
+### Native outcome readback (optional, separate from analyzer judgments)
+
+The trusted provider/Agent may add `--native-outcomes-json native-outcomes.json`
+to the existing `benchmark trajectory-pack-register` operation. The file is
+relative to its already authorized `--source-root`; the renderer cannot register
+paths or invent measurements. Reading it does not execute analysis or change a
+solver, evaluator, retained archive or study board. Omitting it preserves the
+previous pack identity and response shape.
+
+The provider-exported JSON has this local diagnostic contract:
+
+```json
+{
+  "schema_version": "trajectory_native_outcomes_v1",
+  "scope": {"goal_id":"solver", "case_id":"case", "run_id":"run", "revision":"fixed", "window":"declared capture"},
+  "start": "2026-01-01T00:00:00Z",
+  "end": "2026-01-01T00:10:00Z",
+  "complete": true,
+  "metric": {"name":"error", "unit":"points", "higher_is_better":false},
+  "retained_result_id": "result-1",
+  "samples": [{
+    "submission_index":1, "result_id":"result-1", "candidate_ref":"candidate-a",
+    "submitted_at":"2026-01-01T00:01:00Z", "observed_at":"2026-01-01T00:02:00Z",
+    "valid":true, "value":2,
+    "citation":{"source_id":"native-source", "first_line":1, "last_line":1}
+  }]
+}
+```
+
+Every scope field must equal the registered assessment input. Source IDs and
+line ranges must belong to that input; measurement/candidate IDs are opaque.
+UTC timestamps must name real calendar instants (including leap-day validity),
+not dates that a JavaScript parser silently normalizes into another window.
+The source must contain provider-recorded released outcomes, not assessor
+ratings or hidden evaluator material. The input curator/provider remains
+responsible for the claim's truth and validity. `complete` declares prefix
+coverage, not official integrity or score countability. A complete prefix has
+contiguous submission indices starting at one. An invalid record uses
+`valid:false,value:null`; zero is a valid measured value, never a missing value.
+`retained_result_id:null` means retention is unknown, not that the mathematically
+best value was adopted.
+
+The existing typed benchmark-toolkit prepare owner returns a separate bounded
+`native_outcomes` projection: first valid submission-one value when the prefix
+is complete, best valid measurement within the capture under the explicit native
+metric direction, provider-selected retained result, and final valid submission.
+These are distinct: delayed feedback does not promote an older submission to
+the final candidate, and a better measured score does not override retention or
+guardrails. No generic score, monetary estimate or causal benefit is derived.
+The original analysis key/model input is unchanged; native capture provenance
+instead versions the pack identity. Changed/unavailable files leave historical
+captured values visible with their warning, rather than borrowing another pack.
+
+The existing Goal research source disclosure renders those facts and opens only
+their bound original citations, even before model analysis exists. Solver
+capture, native outcome and analyzed-episode windows remain separate. The cost
+disclosure also shows the current analysis attempt's observed receipt usage;
+earlier failures, native evaluation, recovery, operator costs and unmeasured
+windows are not silently combined into a full-study bill. This does not complete
+threshold-cost, gain-area, matched-study or scientific qualification.
+
+Synthetic packaged native readback (not benchmark findings):
+
+![Native outcome source disclosure](../development/assets/research-analysis/native-outcomes-desktop.png)
+
+[Narrow-screen source disclosure](../development/assets/research-analysis/native-outcomes-mobile.png).
+The existing six-judgment first viewport is unchanged; these facts live in its
+existing source disclosure, with costs in their separate disclosure.
+
+### Solver capture registration
+
 A trusted pack curator can bind a **complete captured Codex session** separately
 from the model evidence. The existing register command accepts: