Merge pull request #1414 from pikasTech/fix/1403-analysis-tooling

Improve Workbench triad analysis drilldown

scripts/src/hwlab-node-web-observe-analyzer-source.ts

@@ -1132,6 +1132,7 @@ function compactWorkbenchTurnStateTriadForOutput(value) {
   if (!value || typeof value !== "object") return null;
   return {
     summary: value.summary ?? null,
+    drilldown: value.drilldown ?? null,
     invalidFullTriads: Array.isArray(value.invalidFullTriads) ? value.invalidFullTriads.slice(0, 8) : [],
     cardFinalResponseMismatches: Array.isArray(value.cardFinalResponseMismatches) ? value.cardFinalResponseMismatches.slice(0, 8) : [],
     collectorMissingRows: Array.isArray(value.collectorMissingRows) ? value.collectorMissingRows.slice(0, 8) : [],
@@ -2858,6 +2859,8 @@ function buildWorkbenchTurnStateTriadMetrics(samples, timeline = []) {
   const collectorMissingRows = rows.filter((row) => Array.isArray(row.collectorMissing) && row.collectorMissing.length > 0);
   const invalidRowKeys = new Set([...invalidFullTriads, ...cardFinalResponseMismatches].map((row) => row.rowKey));
   const collectorMissingFields = uniqueSorted(collectorMissingRows.flatMap((row) => row.collectorMissing || []));
+  const offendingRows = Array.from(new Map([...invalidFullTriads, ...cardFinalResponseMismatches].map((row) => [row.rowKey, row])).values());
+  const drilldown = buildWorkbenchTurnStateTriadDrilldown(offendingRows);
   return {
     summary: {
       sampleCount: sourceSamples.length,
@@ -2866,10 +2869,14 @@ function buildWorkbenchTurnStateTriadMetrics(samples, timeline = []) {
       invalidRowCount: invalidRowKeys.size,
       invalidFullTriadCount: invalidFullTriads.length,
       cardFinalResponseMismatchCount: cardFinalResponseMismatches.length,
+      invalidGroupCount: drilldown.summary.groupCount,
+      invalidTraceIdCount: drilldown.summary.traceIds.length,
+      invalidMaxObservedSpanMs: drilldown.summary.maxObservedSpanMs,
       collectorMissingRowCount: collectorMissingRows.length,
       collectorMissingFields,
       valuesRedacted: true
     },
+    drilldown,
     invalidFullTriads: invalidFullTriads.slice(0, 120),
     cardFinalResponseMismatches: cardFinalResponseMismatches.slice(0, 120),
     collectorMissingRows: collectorMissingRows.slice(0, 120),
@@ -2878,6 +2885,210 @@ function buildWorkbenchTurnStateTriadMetrics(samples, timeline = []) {
   };
 }
 
+function buildWorkbenchTurnStateTriadDrilldown(rows) {
+  const uniqueRows = Array.from(new Map((Array.isArray(rows) ? rows : [])
+    .filter(Boolean)
+    .map((row) => [row?.rowKey || [row?.seq, row?.traceId, row?.messageId, row?.pageRole].join("|"), row])).values());
+  const groupsByKey = new Map();
+  let firstAt = null;
+  let lastAt = null;
+  for (const row of uniqueRows) {
+    const traceId = firstString(row?.traceId);
+    const mismatchKind = workbenchTriadMismatchKind(row);
+    const tuple = workbenchTriadTuple(row);
+    const key = [
+      traceId || row?.messageId || row?.sessionIdPrefix || "-",
+      row?.sessionIdPrefix || "-",
+      row?.pageRole || "-",
+      mismatchKind,
+      tuple,
+    ].join("|");
+    let group = groupsByKey.get(key);
+    if (!group) {
+      group = {
+        keyHash: sha256(key),
+        mismatchKind,
+        statusTuple: tuple,
+        traceId: traceId || null,
+        messageId: row?.messageId ?? null,
+        sessionIdPrefix: row?.sessionIdPrefix ?? null,
+        pageRole: row?.pageRole ?? null,
+        pageId: row?.pageId ?? null,
+        count: 0,
+        firstSeq: row?.seq ?? null,
+        lastSeq: row?.seq ?? null,
+        firstAt: row?.ts ?? null,
+        lastAt: row?.ts ?? null,
+        promptIndexes: new Set(),
+        commandIds: new Set(),
+        rawRailStatuses: new Set(),
+        rawCardStatuses: new Set(),
+        finalResponseTextBytes: new Set(),
+        examples: [],
+        valuesRedacted: true,
+      };
+      groupsByKey.set(key, group);
+    }
+    group.count += 1;
+    group.firstSeq = minNumberOrValue(group.firstSeq, row?.seq);
+    group.lastSeq = maxNumberOrValue(group.lastSeq, row?.seq);
+    group.firstAt = minIso(group.firstAt, row?.ts ?? null);
+    group.lastAt = maxIso(group.lastAt, row?.ts ?? null);
+    firstAt = minIso(firstAt, row?.ts ?? null);
+    lastAt = maxIso(lastAt, row?.ts ?? null);
+    if (row?.promptIndex !== null && row?.promptIndex !== undefined) group.promptIndexes.add(String(row.promptIndex));
+    if (row?.nearestCommandId) group.commandIds.add(String(row.nearestCommandId));
+    if (row?.railStatusRaw) group.rawRailStatuses.add(String(row.railStatusRaw));
+    if (row?.cardStatusRaw) group.rawCardStatuses.add(String(row.cardStatusRaw));
+    if (row?.finalResponseTextBytes !== null && row?.finalResponseTextBytes !== undefined) group.finalResponseTextBytes.add(String(row.finalResponseTextBytes));
+    if (group.examples.length < 3) {
+      group.examples.push({
+        seq: row?.seq ?? null,
+        ts: row?.ts ?? null,
+        traceId,
+        railStatus: row?.railStatus ?? null,
+        railStatusRaw: row?.railStatusRaw ?? null,
+        cardStatus: row?.cardStatus ?? null,
+        cardStatusRaw: row?.cardStatusRaw ?? null,
+        finalResponsePresent: row?.finalResponsePresent ?? null,
+        finalResponseTextBytes: row?.finalResponseTextBytes ?? null,
+        nearestCommandId: row?.nearestCommandId ?? null,
+        valuesRedacted: true,
+      });
+    }
+  }
+  const groups = Array.from(groupsByKey.values()).map((group) => {
+    const observedSpanMs = isoSpanMs(group.firstAt, group.lastAt);
+    return {
+      keyHash: group.keyHash,
+      mismatchKind: group.mismatchKind,
+      statusTuple: group.statusTuple,
+      traceId: group.traceId,
+      messageId: group.messageId,
+      sessionIdPrefix: group.sessionIdPrefix,
+      pageRole: group.pageRole,
+      pageId: group.pageId,
+      count: group.count,
+      firstSeq: group.firstSeq,
+      lastSeq: group.lastSeq,
+      firstAt: group.firstAt,
+      lastAt: group.lastAt,
+      observedSpanMs,
+      promptIndexes: Array.from(group.promptIndexes).slice(0, 8),
+      commandIds: Array.from(group.commandIds).slice(0, 8),
+      rawRailStatuses: Array.from(group.rawRailStatuses).slice(0, 8),
+      rawCardStatuses: Array.from(group.rawCardStatuses).slice(0, 8),
+      finalResponseTextBytes: Array.from(group.finalResponseTextBytes).slice(0, 8),
+      examples: group.examples,
+      valuesRedacted: true,
+    };
+  }).sort((left, right) => Number(right.observedSpanMs ?? 0) - Number(left.observedSpanMs ?? 0) || Number(right.count ?? 0) - Number(left.count ?? 0));
+  const traceIds = uniqueSorted(uniqueRows.map((row) => row?.traceId).filter(Boolean)).slice(0, 12);
+  const sessionPrefixes = uniqueSorted(uniqueRows.map((row) => row?.sessionIdPrefix).filter(Boolean)).slice(0, 12);
+  const mismatchKinds = uniqueSorted(uniqueRows.map(workbenchTriadMismatchKind).filter(Boolean));
+  return {
+    summary: {
+      invalidUniqueRows: uniqueRows.length,
+      groupCount: groups.length,
+      traceIds,
+      sessionPrefixes,
+      mismatchKinds,
+      firstAt,
+      lastAt,
+      maxObservedSpanMs: groups.reduce((value, item) => Math.max(value, Number(item.observedSpanMs ?? 0)), 0),
+      valuesRedacted: true,
+    },
+    groups: groups.slice(0, 12),
+    otelDrilldown: buildWorkbenchTriadOtelDrilldown(traceIds.slice(0, 4)),
+    staticSourceHints: workbenchTriadStaticSourceHints(),
+    unitTestReproHints: workbenchTriadUnitTestReproHints(),
+    valuesRedacted: true,
+  };
+}
+
+function buildWorkbenchTriadOtelDrilldown(traceIds) {
+  const target = otelTargetForAnalyzer();
+  const ids = (Array.isArray(traceIds) ? traceIds : []).filter(Boolean).slice(0, 4);
+  return {
+    target,
+    commands: ids.flatMap((traceId) => [
+      "bun scripts/cli.ts platform-infra observability diagnose-code-agent --target " + target + " --business-trace-id " + traceId + " --full",
+      "bun scripts/cli.ts platform-infra observability trace --target " + target + " --trace-id <otel-trace-id-from-diagnose> --grep session_ --limit 60 --full",
+      "bun scripts/cli.ts platform-infra observability trace --target " + target + " --trace-id <otel-trace-id-from-diagnose> --grep turn_status_read --limit 40 --full",
+      "bun scripts/cli.ts platform-infra observability trace --target " + target + " --trace-id <otel-trace-id-from-diagnose> --grep projection --limit 120 --full",
+    ]).slice(0, 16),
+    valuesRedacted: true,
+  };
+}
+
+function otelTargetForAnalyzer() {
+  const explicit = firstString(process.env.UNIDESK_WEB_OBSERVE_OTEL_TARGET, process.env.UNIDESK_OBSERVABILITY_TARGET);
+  if (explicit) return explicit;
+  const parts = String(stateDir || "").split(/[\\\/]+/u);
+  const index = parts.lastIndexOf("web-observe");
+  if (index >= 0 && parts[index + 1]) return String(parts[index + 1]).toUpperCase();
+  return "JD01";
+}
+
+function workbenchTriadMismatchKind(row) {
+  if (!row || typeof row !== "object") return "unknown";
+  if (row.railCardMismatch === true) return "rail-card-status-mismatch";
+  if (row.cardFinalResponseMismatch === true) return "completed-card-final-response-absent";
+  if (row.legacyCardFinalResponseMismatch === true) return "completed-card-final-response-uncollected-or-absent";
+  if (row.tupleAllowed === false) return "tuple-not-allowed";
+  return "unknown";
+}
+
+function workbenchTriadTuple(row) {
+  return [
+    "rail=" + (row?.railStatus ?? "-"),
+    "card=" + (row?.cardStatus ?? "-"),
+    "final=" + String(row?.finalResponsePresent ?? "unknown"),
+  ].join(",");
+}
+
+function minNumberOrValue(left, right) {
+  const l = Number(left);
+  const r = Number(right);
+  if (!Number.isFinite(l)) return right ?? left ?? null;
+  if (!Number.isFinite(r)) return left ?? right ?? null;
+  return Math.min(l, r);
+}
+
+function maxNumberOrValue(left, right) {
+  const l = Number(left);
+  const r = Number(right);
+  if (!Number.isFinite(l)) return right ?? left ?? null;
+  if (!Number.isFinite(r)) return left ?? right ?? null;
+  return Math.max(l, r);
+}
+
+function isoSpanMs(firstAt, lastAt) {
+  const start = Date.parse(firstAt || "");
+  const end = Date.parse(lastAt || "");
+  if (!Number.isFinite(start) || !Number.isFinite(end) || end < start) return 0;
+  return end - start;
+}
+
+function workbenchTriadStaticSourceHints() {
+  return [
+    { repo: "pikasTech/HWLAB", path: "web/hwlab-cloud-web/src/stores/workbench-session.ts", reason: "session rail/list status and active session projection source" },
+    { repo: "pikasTech/HWLAB", path: "web/hwlab-cloud-web/src/stores/workbench-server-state.ts", reason: "server snapshot merge path that can overwrite sealed turn state" },
+    { repo: "pikasTech/HWLAB", path: "web/hwlab-cloud-web/src/stores/workbench-message-projection-runtime.ts", reason: "turn card and Final Response projection/runtime merge path" },
+    { repo: "pikasTech/HWLAB", path: "internal/cloud/server-workbench-http*.go", reason: "session list/detail/messages/turn REST read-model contract" },
+    { repo: "pikasTech/HWLAB", path: "internal/cloud/workbench-projection-*.go", reason: "durable projection writer and terminal seal contract" },
+  ];
+}
+
+function workbenchTriadUnitTestReproHints() {
+  return [
+    "backend projector: terminal event must produce a single sealed turn tuple consumed by session list, session detail, messages and turn-status APIs",
+    "backend read model: completed rail status must not coexist with running turn card or missing Final Response for the same trace",
+    "frontend server-state merge: stale running/empty snapshots must not overwrite a sealed completed+Final Response turn",
+    "frontend projection runtime: cross-page hydration must converge from the same durable projection without DOM-only repair",
+  ];
+}
+
 function workbenchTurnStateTriadRow(sample, turn, rail, promptIndex) {
   const collectorMissing = [];
   const railRawStatus = firstString(rail?.status, rail?.dataStatus);
@@ -3168,7 +3379,10 @@ function buildFindings(samples, control, network, errors, sampleMetrics, promptN
       { railStatus: "running", cardStatus: "running", finalResponsePresent: false }
     ],
     samples: turnStateTriadRows.slice(0, 20),
+    drilldown: turnStateTriad.drilldown ?? buildWorkbenchTurnStateTriadDrilldown(turnStateTriadRows),
     collectorMissingSamples: Array.isArray(turnStateTriad.collectorMissingRows) ? turnStateTriad.collectorMissingRows.slice(0, 10) : [],
+    sourceOfTruth: "durable Workbench projection/read model; do not repair via DOM fallback or GET-side state mutation",
+    nextAction: "Use drilldown.otelDrilldown.commands for the listed traceIds, then inspect staticSourceHints and add unit tests from unitTestReproHints before changing UI rendering.",
     rootCause: "workbench_projection_state_triad_not_sealed",
     rootCauseStatus: "confirmed-from-dom-samples",
     rootCauseConfidence: "high",
@@ -3301,7 +3515,21 @@ function buildFindings(samples, control, network, errors, sampleMetrics, promptN
   const evaluatedCrossPageProjectionDiffs = timedCrossPageProjectionDiffs.filter((item) => !controlledNavigationHydrationCrossPageDiff(item, controlledNavigationWindows, sampleBySeq) && !crossPageDiffHasWorkbenchAppShellNotReady(item, sampleBySeq));
   const persistentCrossPageProjectionDiffs = evaluatedCrossPageProjectionDiffs.filter((item) => Number(item.observedSpanMs ?? 0) > crossPageProjectionBudgetMs);
   const transientCrossPageProjectionDiffs = evaluatedCrossPageProjectionDiffs.filter((item) => Number(item.observedSpanMs ?? 0) <= crossPageProjectionBudgetMs);
-  if (persistentCrossPageProjectionDiffs.length > 0) findings.push({ id: "cross-page-projection-divergence", severity: "red", summary: "control and observer pages saw different projection state for the same sampled session beyond the configured budget", count: persistentCrossPageProjectionDiffs.length, budgetMs: crossPageProjectionBudgetMs, samples: persistentCrossPageProjectionDiffs.slice(0, 20) });
+  if (persistentCrossPageProjectionDiffs.length > 0) findings.push({
+    id: "cross-page-projection-divergence",
+    severity: "red",
+    summary: "control and observer pages saw different projection state for the same sampled session beyond the configured budget",
+    count: persistentCrossPageProjectionDiffs.length,
+    budgetMs: crossPageProjectionBudgetMs,
+    samples: persistentCrossPageProjectionDiffs.slice(0, 20),
+    drilldown: buildCrossPageProjectionDrilldown(persistentCrossPageProjectionDiffs),
+    sourceOfTruth: "session list/detail/messages/turn APIs must read from the same durable projection snapshot across pages",
+    rootCause: "workbench_cross_page_projection_not_single_source",
+    rootCauseStatus: "confirmed-from-control-observer-dom-samples",
+    rootCauseConfidence: "high",
+    nextAction: "Use drilldown.traceIds with diagnose-code-agent and compare session_list_read/session_messages_read/turn_status_read rows before changing renderer fallback behavior.",
+    valuesRedacted: true
+  });
   if (transientCrossPageProjectionDiffs.length > 0) findings.push({ id: "cross-page-projection-transient-divergence", severity: "info", summary: "control and observer pages briefly differed near a sampled transition; retained as transient evidence but not treated as persistent projection failure", count: transientCrossPageProjectionDiffs.length, budgetMs: crossPageProjectionBudgetMs, samples: transientCrossPageProjectionDiffs.slice(0, 20) });
   if (controlledNavigationHydrationProjectionDiffs.length > 0) findings.push({ id: "cross-page-projection-controlled-navigation-hydration", severity: "info", summary: "control and observer pages differed while a non-blocking session-invariance navigation command still had an unhydrated blank page; retained as context but not treated as a red projection blocker", count: controlledNavigationHydrationProjectionDiffs.length, budgetMs: crossPageProjectionBudgetMs, samples: controlledNavigationHydrationProjectionDiffs.slice(0, 20) });
   if (appShellNotReadyProjectionDiffs.length > 0) findings.push({ id: "cross-page-projection-app-shell-not-ready", severity: "info", summary: "cross-page projection differences were explained by a page whose Workbench app shell was not mounted; see workbench-app-shell-not-ready for the blocking root cause", count: appShellNotReadyProjectionDiffs.length, budgetMs: crossPageProjectionBudgetMs, samples: appShellNotReadyProjectionDiffs.slice(0, 20) });
@@ -3339,6 +3567,114 @@ function buildFindings(samples, control, network, errors, sampleMetrics, promptN
   return findings;
 }
 
+function buildCrossPageProjectionDrilldown(rows) {
+  const sourceRows = Array.isArray(rows) ? rows.filter(Boolean) : [];
+  const traceIds = uniqueSorted(sourceRows.flatMap((row) => collectIdsFromUnknown(row, "trace")).filter(Boolean)).slice(0, 12);
+  const sessionIds = uniqueSorted(sourceRows.flatMap((row) => collectIdsFromUnknown(row, "session")).filter(Boolean)).slice(0, 12);
+  const diffKinds = uniqueSorted(sourceRows.map((row) => row?.diffKind).filter(Boolean));
+  const groupsByKey = new Map();
+  for (const row of sourceRows) {
+    const rowTraceIds = collectIdsFromUnknown(row, "trace").slice(0, 8);
+    const rowSessionIds = collectIdsFromUnknown(row, "session").slice(0, 8);
+    const key = [
+      row?.diffKind || "projection",
+      rowSessionIds[0] || row?.sessionIdPrefix || "-",
+      rowTraceIds[0] || "-",
+    ].join("|");
+    let group = groupsByKey.get(key);
+    if (!group) {
+      group = {
+        keyHash: sha256(key),
+        diffKind: row?.diffKind ?? null,
+        traceIds: new Set(),
+        sessionIds: new Set(),
+        count: 0,
+        firstSeq: row?.firstSeq ?? row?.controlSeq ?? row?.observerSeq ?? row?.seq ?? null,
+        lastSeq: row?.lastSeq ?? row?.controlSeq ?? row?.observerSeq ?? row?.seq ?? null,
+        firstAt: row?.firstAt ?? row?.controlTs ?? row?.observerTs ?? row?.ts ?? null,
+        lastAt: row?.lastAt ?? row?.controlTs ?? row?.observerTs ?? row?.ts ?? null,
+        maxObservedSpanMs: 0,
+        examples: [],
+      };
+      groupsByKey.set(key, group);
+    }
+    group.count += 1;
+    for (const traceId of rowTraceIds) group.traceIds.add(traceId);
+    for (const sessionId of rowSessionIds) group.sessionIds.add(sessionId);
+    group.firstSeq = minNumberOrValue(group.firstSeq, row?.firstSeq ?? row?.controlSeq ?? row?.observerSeq ?? row?.seq);
+    group.lastSeq = maxNumberOrValue(group.lastSeq, row?.lastSeq ?? row?.controlSeq ?? row?.observerSeq ?? row?.seq);
+    group.firstAt = minIso(group.firstAt, row?.firstAt ?? row?.controlTs ?? row?.observerTs ?? row?.ts ?? null);
+    group.lastAt = maxIso(group.lastAt, row?.lastAt ?? row?.controlTs ?? row?.observerTs ?? row?.ts ?? null);
+    group.maxObservedSpanMs = Math.max(group.maxObservedSpanMs, Number(row?.observedSpanMs ?? 0) || 0);
+    if (group.examples.length < 3) {
+      group.examples.push({
+        diffKind: row?.diffKind ?? null,
+        firstSeq: row?.firstSeq ?? null,
+        lastSeq: row?.lastSeq ?? null,
+        observedSpanMs: row?.observedSpanMs ?? null,
+        controlMessageCount: row?.controlMessageCount ?? null,
+        observerMessageCount: row?.observerMessageCount ?? null,
+        traceIds: rowTraceIds.slice(0, 6),
+        sessionIds: rowSessionIds.slice(0, 4),
+        valuesRedacted: true,
+      });
+    }
+  }
+  const groups = Array.from(groupsByKey.values()).map((group) => ({
+    keyHash: group.keyHash,
+    diffKind: group.diffKind,
+    traceIds: Array.from(group.traceIds).slice(0, 8),
+    sessionIds: Array.from(group.sessionIds).slice(0, 6),
+    count: group.count,
+    firstSeq: group.firstSeq,
+    lastSeq: group.lastSeq,
+    firstAt: group.firstAt,
+    lastAt: group.lastAt,
+    observedSpanMs: Math.max(group.maxObservedSpanMs, isoSpanMs(group.firstAt, group.lastAt)),
+    examples: group.examples,
+    valuesRedacted: true,
+  })).sort((left, right) => Number(right.observedSpanMs ?? 0) - Number(left.observedSpanMs ?? 0) || Number(right.count ?? 0) - Number(left.count ?? 0));
+  return {
+    summary: {
+      rowCount: sourceRows.length,
+      groupCount: groups.length,
+      diffKinds,
+      traceIds,
+      sessionIds,
+      maxObservedSpanMs: groups.reduce((value, item) => Math.max(value, Number(item.observedSpanMs ?? 0)), 0),
+      valuesRedacted: true,
+    },
+    traceIds,
+    sessionIds,
+    groups: groups.slice(0, 12),
+    otelDrilldown: buildWorkbenchTriadOtelDrilldown(traceIds.slice(0, 4)),
+    staticSourceHints: workbenchTriadStaticSourceHints(),
+    unitTestReproHints: [
+      "two independent Workbench pages must converge to identical session list/detail/messages/turn projection after the same durable events",
+      "late session-list or message refresh must not drop a trace that another page already read from durable projection",
+      "read-side APIs must expose enough OTel fields to compare projection version/cursor and trace ids across pages",
+    ],
+    valuesRedacted: true,
+  };
+}
+
+function collectIdsFromUnknown(value, kind, output = new Set(), depth = 0) {
+  if (depth > 4 || value === null || value === undefined) return Array.from(output);
+  if (typeof value === "string") {
+    const pattern = kind === "session" ? /\bses_[A-Za-z0-9_-]+\b/gu : /\btrc_[A-Za-z0-9_-]+\b/gu;
+    for (const match of value.match(pattern) || []) output.add(match);
+    return Array.from(output);
+  }
+  if (Array.isArray(value)) {
+    for (const item of value.slice(0, 40)) collectIdsFromUnknown(item, kind, output, depth + 1);
+    return Array.from(output);
+  }
+  if (typeof value === "object") {
+    for (const item of Object.values(value).slice(0, 80)) collectIdsFromUnknown(item, kind, output, depth + 1);
+  }
+  return Array.from(output);
+}
+
 function buildFrontendFreezeFindings(errors, control) {
   const findings = [];
   const promptTimes = (control || [])

scripts/src/hwlab-node-web-observe-analyzer-timing-source.ts

@@ -2219,6 +2219,9 @@ function renderMarkdown(report) {
   const loadingSummary = loading.summary || {};
   const sessionRailTitles = report.sampleMetrics?.sessionRailTitles || {};
   const sessionRailTitleSummary = sessionRailTitles.summary || {};
+  const workbenchTurnStateTriad = report.sampleMetrics?.workbenchTurnStateTriad || {};
+  const workbenchTurnStateTriadSummary = workbenchTurnStateTriad.summary || {};
+  const workbenchTurnStateTriadDrilldown = workbenchTurnStateTriad.drilldown || {};
   const codeAgentCardTiming = report.sampleMetrics?.codeAgentCardTiming || {};
   const codeAgentCardTimingSummary = codeAgentCardTiming.summary || {};
   const roundCompletion = codeAgentCardTiming.roundCompletion || {};
@@ -2312,6 +2315,18 @@ function renderMarkdown(report) {
   const sessionRailTitleExampleLines = Array.isArray(sessionRailTitles.examples) && sessionRailTitles.examples.length > 0
     ? sessionRailTitles.examples.slice(0, 80).map((item) => "- firstSeq=" + (item.firstSeq ?? "-") + " role=" + (item.pageRole || "-") + " active=" + String(item.active === true) + " sessionPrefix=" + (item.sessionIdPrefix || "-") + " titleHash=" + (item.titleHash || "-") + " preview=" + escapeMarkdownCell(item.titlePreview || "")).join("\n")
     : "- 无 fallback 标题示例。";
+  const workbenchTurnStateTriadGroupLines = Array.isArray(workbenchTurnStateTriadDrilldown.groups) && workbenchTurnStateTriadDrilldown.groups.length > 0
+    ? workbenchTurnStateTriadDrilldown.groups.slice(0, 80).map((item) => "- mismatch=" + (item.mismatchKind || "-") + " tuple=" + (item.statusTuple || "-") + " traceId=" + (item.traceId || "-") + " session=" + (item.sessionIdPrefix || "-") + " role=" + (item.pageRole || "-") + " count=" + (item.count ?? 0) + " spanMs=" + (item.observedSpanMs ?? 0) + " seq=" + (item.firstSeq ?? "-") + ".." + (item.lastSeq ?? "-") + " commands=" + ((Array.isArray(item.commandIds) ? item.commandIds : []).slice(0, 4).join(",") || "-") + " railRaw=" + ((Array.isArray(item.rawRailStatuses) ? item.rawRailStatuses : []).slice(0, 4).join(",") || "-") + " cardRaw=" + ((Array.isArray(item.rawCardStatuses) ? item.rawCardStatuses : []).slice(0, 4).join(",") || "-") + " finalBytes=" + ((Array.isArray(item.finalResponseTextBytes) ? item.finalResponseTextBytes : []).slice(0, 4).join(",") || "-")).join("\n")
+    : "- 无三态不一致聚合分组。";
+  const workbenchTurnStateTriadOtelLines = Array.isArray(workbenchTurnStateTriadDrilldown.otelDrilldown?.commands) && workbenchTurnStateTriadDrilldown.otelDrilldown.commands.length > 0
+    ? workbenchTurnStateTriadDrilldown.otelDrilldown.commands.slice(0, 16).map((item) => "- command=" + String(item).replaceAll(String.fromCharCode(96), "'")).join("\n")
+    : "- 无可用 traceId 生成 OTel drill-down 命令。";
+  const workbenchTurnStateTriadStaticLines = Array.isArray(workbenchTurnStateTriadDrilldown.staticSourceHints) && workbenchTurnStateTriadDrilldown.staticSourceHints.length > 0
+    ? workbenchTurnStateTriadDrilldown.staticSourceHints.map((item) => "- " + (item.repo || "-") + " " + (item.path || "-") + ": " + escapeMarkdownCell(item.reason || "")).join("\n")
+    : "- 无静态源码候选。";
+  const workbenchTurnStateTriadUnitLines = Array.isArray(workbenchTurnStateTriadDrilldown.unitTestReproHints) && workbenchTurnStateTriadDrilldown.unitTestReproHints.length > 0
+    ? workbenchTurnStateTriadDrilldown.unitTestReproHints.map((item) => "- " + escapeMarkdownCell(item)).join("\n")
+    : "- 无单元测试复现提示。";
   const provenanceLines = Array.isArray(report.pageProvenance?.segments) && report.pageProvenance.segments.length > 0
     ? report.pageProvenance.segments.slice(0, 40).map((item) => "- fingerprint=" + (item.assetFingerprint || "-") + " samples=" + item.sampleCount + " seq=" + (item.firstSeq ?? "-") + ".." + (item.lastSeq ?? "-") + " ts=" + (item.firstAt || "-") + ".." + (item.lastAt || "-") + " scripts=" + (item.scriptCount ?? "-") + " styles=" + (item.stylesheetCount ?? "-") + " urlPaths=" + (Array.isArray(item.urlPaths) ? item.urlPaths.slice(0, 4).join(",") : "-")).join("\n")
     : "- 无页面 provenance segment。";
@@ -2471,6 +2486,19 @@ function renderMarkdown(report) {
     + "- policy: 可见 session 列表中 'Session ses_...' fallback 标题超过一半必须报警；修复应让上游 session list projection 直接携带名称，不能靠点击详情后下游修补。\n\n"
     + "#### Session rail fallback samples\n\n" + sessionRailTitleSampleLines + "\n\n"
     + "#### Session rail fallback examples\n\n" + sessionRailTitleExampleLines + "\n\n"
+    + "### Workbench turn-state triad\n\n"
+    + "- rowCount: " + (workbenchTurnStateTriadSummary.rowCount ?? 0) + "\n"
+    + "- invalidRowCount: " + (workbenchTurnStateTriadSummary.invalidRowCount ?? 0) + "\n"
+    + "- invalidFullTriadCount: " + (workbenchTurnStateTriadSummary.invalidFullTriadCount ?? 0) + "\n"
+    + "- cardFinalResponseMismatchCount: " + (workbenchTurnStateTriadSummary.cardFinalResponseMismatchCount ?? 0) + "\n"
+    + "- invalidGroupCount: " + (workbenchTurnStateTriadSummary.invalidGroupCount ?? workbenchTurnStateTriadDrilldown.summary?.groupCount ?? 0) + "\n"
+    + "- invalidTraceIdCount: " + (workbenchTurnStateTriadSummary.invalidTraceIdCount ?? workbenchTurnStateTriadDrilldown.summary?.traceIds?.length ?? 0) + "\n"
+    + "- invalidMaxObservedSpanMs: " + (workbenchTurnStateTriadSummary.invalidMaxObservedSpanMs ?? workbenchTurnStateTriadDrilldown.summary?.maxObservedSpanMs ?? 0) + "\n"
+    + "- allowedTuples: completed/completed/final-present 或 running/running/final-absent；其他组合都应按 projection seal/read-model 合同处理，禁止 UI fallback repair。\n\n"
+    + "#### Triad mismatch groups\n\n" + workbenchTurnStateTriadGroupLines + "\n\n"
+    + "#### Triad OTel drill-down\n\n" + workbenchTurnStateTriadOtelLines + "\n\n"
+    + "#### Triad static source hints\n\n" + workbenchTurnStateTriadStaticLines + "\n\n"
+    + "#### Triad unit-test repro hints\n\n" + workbenchTurnStateTriadUnitLines + "\n\n"
     + "### Page provenance\n\n"
     + "- segmentCount: " + (report.pageProvenance?.summary?.segmentCount ?? 0) + "\n"
     + "- controlSegmentCount: " + (report.pageProvenance?.summary?.controlSegmentCount ?? 0) + "\n\n"

scripts/src/platform-infra-observability/diagnose-code-agent-script.ts

@@ -1621,6 +1621,8 @@ def agentrun_summary(spans, authority=None):
 def hwlab_read_model_summary(spans):
     trace_read_spans = []
     turn_read_spans = []
+    session_list_spans = []
+    session_message_spans = []
     projection_spans = []
     source_event_count = None
     requested_since_seq = None
@@ -1648,6 +1650,10 @@ def hwlab_read_model_summary(spans):
                 status_value = attrs.get("turnStatus") if attrs.get("turnStatus") is not None else attrs.get("status")
                 turn_statuses.append(status_value)
                 turn_status_counts[str(status_value)] += 1
+        if name == "session_list_read":
+            session_list_spans.append(public_span(item))
+        if name == "session_messages_read":
+            session_message_spans.append(public_span(item))
         if name == "projection_sync":
             projection_spans.append(public_span(item))
             after_seq = to_int(attrs.get("afterSeq"))
@@ -1683,6 +1689,10 @@ def hwlab_read_model_summary(spans):
         "turnStatusCounts": [{"status": status, "count": count} for status, count in turn_status_counts.most_common()],
         "traceEventReadSpans": trace_read_spans,
         "turnStatusReadSpans": turn_read_spans,
+        "sessionListReadCount": len(session_list_spans),
+        "sessionMessageReadCount": len(session_message_spans),
+        "sessionListReadSpans": session_list_spans,
+        "sessionMessageReadSpans": session_message_spans,
         "projectionSpans": projection_spans,
     }
 
@@ -2144,6 +2154,8 @@ payload = {
         "traceStatus": read_model.get("traceStatus"),
         "turnStatus": read_model.get("turnStatus"),
         "turnStatusCounts": read_model.get("turnStatusCounts"),
+        "sessionListReadCount": read_model.get("sessionListReadCount"),
+        "sessionMessageReadCount": read_model.get("sessionMessageReadCount"),
     },
     "http": {
         "statusCounts": http_summary.get("statusCounts", [])[:20],
@@ -2159,6 +2171,7 @@ payload = {
         "diagnoseFull": DIAGNOSE_FULL_COMMAND,
         "traceSummary": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --limit 80" % resolved_trace_id,
         "traceReads": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --grep trace_events_read --limit 20 --full" % resolved_trace_id,
+        "sessionReads": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --grep session_ --limit 60 --full" % resolved_trace_id,
         "turnStatus": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --grep turn_status_read --limit 20 --full" % resolved_trace_id,
         "projection": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --grep projection_sync --limit 120 --full" % resolved_trace_id,
         "runnerTerminal": "bun scripts/cli.ts platform-infra observability trace --target ${target.id} --trace-id %s --grep runner_terminal --limit 20 --full" % resolved_trace_id,
@@ -2170,11 +2183,13 @@ if FULL:
     payload["full"] = {
         "traceEventReadSpans": read_model.get("traceEventReadSpans", []),
         "turnStatusReadSpans": read_model.get("turnStatusReadSpans", []),
+        "sessionListReadSpans": read_model.get("sessionListReadSpans", []),
+        "sessionMessageReadSpans": read_model.get("sessionMessageReadSpans", []),
         "projectionSpans": read_model.get("projectionSpans", [])[:LIMIT],
         "terminalSpans": agentrun.get("terminalSpans", [])[:LIMIT],
         "errorSpans": [public_span(item) for item in error_spans[:LIMIT]],
         "httpProblemSpans": [public_span(item) for item in http_summary.get("problemSpans", [])[:LIMIT]],
-        "selectedSpans": [public_span(item) for item in spans if item.get("name") in ("trace_events_read", "turn_status_read", "projection_sync", "command_result")][:LIMIT],
+        "selectedSpans": [public_span(item) for item in spans if item.get("name") in ("trace_events_read", "turn_status_read", "session_list_read", "session_messages_read", "projection_sync", "command_result")][:LIMIT],
     }
 if RAW:
     payload["raw"] = raw_trace_shape(trace_parsed)

scripts/src/platform-infra-observability/render.ts

@@ -749,9 +749,12 @@ export function renderDiagnoseCodeAgentTable(input: {
     textValue(next?.diagnoseFull),
     textValue(next?.traceSummary),
     textValue(next?.traceReads),
+    textValue(next?.sessionReads),
+    textValue(next?.turnStatus),
+    textValue(next?.projection),
   ].filter((item) => item !== "-");
   if (nextCommands.length > 0) {
-    for (const command of nextCommands.slice(0, 3)) lines.push(`  ${command}`);
+    for (const command of nextCommands.slice(0, 6)) lines.push(`  ${command}`);
   } else {
     lines.push(`  ${buildDiagnoseCommand(input.target, input.options, true)}`);
     if (traceId.length > 0 && traceId !== "-") lines.push(`  bun scripts/cli.ts platform-infra observability trace --target ${input.target.id} --trace-id ${traceId}`);