feat(web): critique notes, run drift metrics, benchmark outlier filter, hub shortcuts

apps/web/app/api/outliers/route.ts

@@ -11,6 +11,7 @@ function readFilterParams(url: URL) {
         fast: url.searchParams.get("fast") ?? undefined,
         from: url.searchParams.get("from") ?? undefined,
         to: url.searchParams.get("to") ?? undefined,
+        benchmark: url.searchParams.get("benchmark") ?? undefined,
     };
 }
 
@@ -28,7 +29,10 @@ export async function GET(request: Request) {
     }
 
     const filteredIndex = applyIndexFilters(index, filters);
-    const rows = await buildOutlierExplorerRows(filteredIndex);
+    const benchmarkId = (filters.benchmark ?? "").trim();
+    const rows = await buildOutlierExplorerRows(filteredIndex, {
+        benchmarkId: benchmarkId || undefined,
+    });
 
     if (format === "csv") {
         const csv = outliersToCsv(rows);

apps/web/app/benchmarks/[id]/page.tsx

@@ -22,7 +22,10 @@ import {
     buildBenchmarkIndexLookup,
     resolveIndexedModeLabel,
 } from "../../../lib/benchmarkIndexLookup";
-import { buildBenchmarkOutlierLookup } from "../../../lib/outlierLookup";
+import {
+    buildBenchmarkOutlierLookup,
+    outliersBenchmarkHref,
+} from "../../../lib/outlierLookup";
 import { questionHubHref } from "../../../lib/questionGroups";
 import {
     extractBenchmarkSummaryDisplay,
@@ -167,7 +170,7 @@ export default async function BenchmarkDetailPage({
                     </Link>
                     {outlierCount > 0 ? (
                         <Link
-                            href={`/outliers?q=${encodeURIComponent(benchmark.question)}`}
+                            href={outliersBenchmarkHref(benchmark.id)}
                             className="button secondary"
                         >
                             Outlier runs ({outlierCount})

apps/web/app/outliers/page.tsx

@@ -19,6 +19,7 @@ type OutliersSearchParams = {
     fast?: string;
     from?: string;
     to?: string;
+    benchmark?: string;
 };
 
 export default async function OutliersPage({
@@ -47,7 +48,10 @@ export default async function OutliersPage({
 
     const { models, presets } = collectIndexFacets(rawIndex);
     const index = applyIndexFilters(rawIndex, params);
-    const rows = await buildOutlierExplorerRows(index);
+    const benchmarkId = (params.benchmark ?? "").trim();
+    const rows = await buildOutlierExplorerRows(index, {
+        benchmarkId: benchmarkId || undefined,
+    });
 
     return (
         <section className="stack">
@@ -78,15 +82,33 @@ export default async function OutliersPage({
                 params={params}
                 totalRuns={rawIndex.runs.length}
                 filteredRuns={index.runs.length}
+                preserveKeys={benchmarkId ? ["benchmark"] : []}
             />
 
+            {benchmarkId ? (
+                <div className="card">
+                    <p className="muted" style={{ margin: 0 }}>
+                        Showing outliers for benchmark{" "}
+                        <Link href={`/benchmarks/${benchmarkId}`}>
+                            <code>{benchmarkId}</code>
+                        </Link>
+                        .{" "}
+                        <Link href="/outliers" className="button secondary">
+                            Clear benchmark filter
+                        </Link>
+                    </p>
+                </div>
+            ) : null}
+
             {rows.length > 0 ? (
                 <div
                     className="page-actions"
                     style={{ display: "flex", gap: 10, flexWrap: "wrap" }}
                 >
                     <a
-                        href={`/api/outliers${buildQueryString(params, {})}`}
+                        href={`/api/outliers${buildQueryString(params, {
+                            benchmark: benchmarkId || undefined,
+                        })}`}
                         className="button secondary"
                         target="_blank"
                         rel="noopener noreferrer"
@@ -95,7 +117,9 @@ export default async function OutliersPage({
                         Export JSON
                     </a>
                     <a
-                        href={`/api/outliers${buildQueryString(params, {})}&format=csv`}
+                        href={`/api/outliers${buildQueryString(params, {
+                            benchmark: benchmarkId || undefined,
+                        })}&format=csv`}
                         className="button secondary"
                         download="outlier-runs.csv"
                     >

apps/web/app/runs/[id]/page.tsx

@@ -16,15 +16,18 @@ import { RunMetricsSummary } from "../../../components/RunMetricsSummary";
 import { ConsensusSummary } from "../../../components/ConsensusSummary";
 import { CritiqueBreakdown } from "../../../components/CritiqueBreakdown";
 import { CritiqueByAgent } from "../../../components/CritiqueByAgent";
+import { MetricCard } from "../../../components/MetricCard";
 import { extractConsensusSummary } from "../../../lib/consensusSummary";
 import { extractCritiqueByType } from "../../../lib/critiqueBreakdown";
 import { extractCritiqueByAgent } from "../../../lib/critiqueByAgent";
+import { extractCritiqueNotesFromRun } from "../../../lib/critiqueNotes";
+import { buildDriftCompareHref } from "../../../lib/confidenceDrift";
 import { DownloadArtifactLink } from "../../../components/DownloadArtifactLink";
 import { CopyPageLink } from "../../../components/CopyPageLink";
 import { CopyTextButton } from "../../../components/CopyTextButton";
 import { summarizeRun } from "../../../lib/runCompare";
 import { buildCompareSuggestions } from "../../../lib/compareSuggestions";
-import { questionHubHref } from "../../../lib/questionGroups";
+import { questionHubHref, listQuestionInsightLinks } from "../../../lib/questionGroups";
 import {
     buildRunStepTiming,
     formatDurationMs,
@@ -76,6 +79,10 @@ export default async function RunTracePage({
     const consensusSummary = extractConsensusSummary(run);
     const critiqueByType = extractCritiqueByType(run);
     const critiqueByAgent = extractCritiqueByAgent(run);
+    const critiqueNotes = extractCritiqueNotesFromRun(run);
+    const indexedRun = index?.runs.find((entry) => entry.id === run.id);
+    const insightLinks = listQuestionInsightLinks(run.question);
+    const driftCompareHref = buildDriftCompareHref(allRuns, run.id);
     const { schemaVersion, pipelineVersion } = run.metadata;
 
     return (
@@ -209,6 +216,108 @@ export default async function RunTracePage({
                 <RunMetricsSummary summary={metricsSummary} />
             </div>
 
+            {indexedRun ? (
+                <div className="card">
+                    <h2 style={{ marginTop: 0 }}>Confidence drift</h2>
+                    <p className="small muted" style={{ marginBottom: "1rem" }}>
+                        Stage-to-stage confidence deltas from the analysis index
+                        — compare with peers on the same question.
+                    </p>
+                    <div className="grid-4">
+                        <MetricCard
+                            label="Solver→revision Δ"
+                            value={
+                                indexedRun.confidence.solverToRevisionDelta !=
+                                null
+                                    ? indexedRun.confidence.solverToRevisionDelta.toFixed(
+                                          3,
+                                      )
+                                    : "—"
+                            }
+                            helpKey="solverToRevisionDelta"
+                        />
+                        <MetricCard
+                            label="Revision→synth Δ"
+                            value={
+                                indexedRun.confidence
+                                    .revisionToSynthesizerDelta != null
+                                    ? indexedRun.confidence.revisionToSynthesizerDelta.toFixed(
+                                          3,
+                                      )
+                                    : "—"
+                            }
+                            helpKey="revisionToSynthesizerDelta"
+                        />
+                        <MetricCard
+                            label="Calibrated−synth Δ"
+                            value={
+                                indexedRun.confidence.calibratedAdjusted !=
+                                    null &&
+                                indexedRun.confidence.synthesizer != null
+                                    ? (
+                                          indexedRun.confidence
+                                              .calibratedAdjusted -
+                                          indexedRun.confidence.synthesizer
+                                      ).toFixed(3)
+                                    : "—"
+                            }
+                            helpKey="calibratedMinusSynthDelta"
+                        />
+                        <MetricCard
+                            label="Evidence risk"
+                            value={
+                                indexedRun.research?.evidenceRiskLevel != null
+                                    ? indexedRun.research.evidenceRiskLevel.toFixed(
+                                          1,
+                                      )
+                                    : "—"
+                            }
+                            helpKey="evidenceRiskLevel"
+                        />
+                    </div>
+                    <div
+                        className="page-actions"
+                        style={{
+                            display: "flex",
+                            gap: 10,
+                            flexWrap: "wrap",
+                            marginTop: 12,
+                        }}
+                    >
+                        <Link href="/drift" className="button secondary">
+                            Drift explorer
+                        </Link>
+                        <Link
+                            href={driftCompareHref}
+                            className="button secondary"
+                        >
+                            Compare drift peer
+                        </Link>
+                    </div>
+                </div>
+            ) : null}
+
+            <div className="card">
+                <h2 style={{ marginTop: 0 }}>Question insight shortcuts</h2>
+                <p className="small muted" style={{ marginBottom: "1rem" }}>
+                    Open filtered insight pages scoped to this research question.
+                </p>
+                <div
+                    className="page-actions"
+                    style={{ display: "flex", gap: 10, flexWrap: "wrap" }}
+                >
+                    {insightLinks.map((link) => (
+                        <Link
+                            key={link.page}
+                            href={link.href}
+                            className="button secondary"
+                        >
+                            {link.label}
+                        </Link>
+                    ))}
+                </div>
+            </div>
+
             {stepTimingSummary.timedStepCount > 0 ? (
                 <div className="card">
                     <h2 style={{ marginTop: 0 }}>Step timing</h2>
@@ -368,6 +477,51 @@ export default async function RunTracePage({
                 <CritiqueBreakdown entries={critiqueByType} />
             </div>
 
+            {critiqueNotes.length > 0 ? (
+                <div className="card">
+                    <h2 style={{ marginTop: 0 }}>Critique notes</h2>
+                    <p className="small muted" style={{ marginBottom: "1rem" }}>
+                        Verbatim issue notes from skeptic and red team critique
+                        steps in this trace.
+                    </p>
+                    <ResponsiveTable
+                        columns={[
+                            { key: "agentName", label: "Agent" },
+                            {
+                                key: "severity",
+                                label: "Severity",
+                                helpKey: "maxSeverity",
+                            },
+                            {
+                                key: "type",
+                                label: "Type",
+                                render: (row) => (
+                                    <Link
+                                        href={`/issues?type=${encodeURIComponent(
+                                            (row as { type: string }).type,
+                                        )}&q=${encodeURIComponent(run.question)}`}
+                                    >
+                                        {(row as { type: string }).type}
+                                    </Link>
+                                ),
+                            },
+                            {
+                                key: "note",
+                                label: "Note",
+                                cellClass: "cell-question",
+                            },
+                        ]}
+                        data={critiqueNotes.map((note, idx) => ({
+                            ...note,
+                            rowKey: `${note.agentName}-${note.type}-${idx}`,
+                        }))}
+                        getRowId={(row) =>
+                            (row as { rowKey: string }).rowKey
+                        }
+                    />
+                </div>
+            ) : null}
+
             <div className="card trace-final-answer">
                 <div
                     style={{

apps/web/app/runs/page.tsx

@@ -341,6 +341,34 @@ export default async function RunsPage({
                                               : value.toFixed(2);
                                       },
                                   },
+                                  {
+                                      key: "evidenceRisk",
+                                      label: "Evidence risk",
+                                      helpKey: "evidenceRiskLevel",
+                                      hideOnMobile: true,
+                                      render: (row: Record<string, unknown>) => {
+                                          const value = (
+                                              row as { evidenceRisk?: number }
+                                          ).evidenceRisk;
+                                          return value == null
+                                              ? "—"
+                                              : value.toFixed(1);
+                                      },
+                                  },
+                                  {
+                                      key: "solverDrift",
+                                      label: "Solver→rev Δ",
+                                      helpKey: "solverToRevisionDelta",
+                                      hideOnMobile: true,
+                                      render: (row: Record<string, unknown>) => {
+                                          const value = (
+                                              row as { solverDrift?: number }
+                                          ).solverDrift;
+                                          return value == null
+                                              ? "—"
+                                              : value.toFixed(2);
+                                      },
+                                  },
                                   {
                                       key: "coherence",
                                       label: "Coherence",
@@ -428,6 +456,8 @@ export default async function RunsPage({
                             issues: indexed?.issueCount,
                             avgSeverity: indexed?.avgSeverity,
                             solverConf: indexed?.solverConfidence,
+                            evidenceRisk: indexed?.evidenceRiskLevel,
+                            solverDrift: indexed?.solverToRevisionDelta,
                             coherence: indexed?.coherence,
                             factualRisk: indexed?.factualRisk,
                         };

apps/web/lib/indexRunLookup.test.ts

@@ -16,14 +16,21 @@ function sampleIndex(): AnalysisIndex {
                 fastMode: false,
                 finalAnswerPreview: "A",
                 stepCount: 10,
-                confidence: { solver: 0.8 },
+                confidence: {
+                    solver: 0.8,
+                    solverToRevisionDelta: -0.12,
+                    revisionToSynthesizerDelta: 0.05,
+                },
                 critique: {
                     issueCount: 5,
                     maxSeverity: 4,
                     avgSeverity: 3.2,
                     byType: { factual: 2 },
                 },
-                research: { evidenceRiskLevel: 3 },
+                research: {
+                    evidenceRiskLevel: 3,
+                    topCounterfactualFailureMode: "missing evidence",
+                },
                 quality: { coherence: 4.1, factualRisk: 2.5 },
             },
         ],
@@ -51,9 +58,12 @@ describe("buildIndexRunLookup", () => {
             maxSeverity: 4,
             avgSeverity: 3.2,
             solverConfidence: 0.8,
+            solverToRevisionDelta: -0.12,
+            revisionToSynthesizerDelta: 0.05,
             evidenceRiskLevel: 3,
             coherence: 4.1,
             factualRisk: 2.5,
+            topCounterfactualFailureMode: "missing evidence",
         });
         expect(lookup.get("missing")).toBeUndefined();
     });