Update admin dashboard for v3 batched AI narration diagnostics

The Narrate tab now correctly shows 'AI narration active (batched)' when
the v3 pipeline runs with batch_ai mode. Handles both v2 (ai_calls count)
and v3 (mode/ai_call_succeeded) diagnostic formats. Updated tooltips to
reflect that observations are now AI-narrated alongside discoveries.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: schwaaamp

docs/planning/insight-engine-v3.md

@@ -1,6 +1,6 @@
 # Insight Engine v3 — Dynamic Metric Discovery & Intraday Analysis
 
-**Status:** DEPLOYED — Phases 0-5 complete, blacklist seeded (212 entries), BH family correction live
+**Status:** DEPLOYED — Phases 0-5 complete, blacklist seeded (212 entries), BH family correction live, batched AI narration live (prompt v5.0)
 **Author:** Claude Code / schwaaamp
 **Date:** 2026-03-16
 **Supersedes:** `pattern-spotter-v2-rewrite.md` (complete, deployed)
@@ -1591,7 +1591,13 @@ catch:
 - `_shared/pattern-ranker.ts` — no changes needed
 - All analyzer modules — no changes needed
 
-**Estimated effort:** Medium — the batched prompt design and experiment matching logic are new. The narration infrastructure (imports, compliance, fallback) was built and reverted once already.
+**Status: DONE (2026-03-19)**
+- Batched AI narration live: single GPT-4o-mini call per pipeline run
+- All discoveries and observations get plain-language titles and summaries
+- OpenAI json_object wrapper format handled (extracts array from `{ "discoveries": [...] }`)
+- Template fallback on AI failure — 21 vitest passing
+- Admin dashboard updated to show v3 narration diagnostics (mode, patterns narrated, AI succeeded)
+- Experiment matching NOT YET IMPLEMENTED — `suggested_experiment_id` still null (deferred to next iteration)
 
 ---
 

web/src/app/routes/admin/insights.$runId.tsx

@@ -358,10 +358,16 @@ export default function InsightRunDetail() {
     all_candidates?: CandidateRow[];
   }) | undefined;
   const narrateDiag = diag.narrate as {
+    // v2 fields
     ai_calls?: number;
     template_narratives?: number;
     model?: string;
     prompt_version?: string;
+    // v3 fields (batched narration)
+    mode?: 'batch_ai' | 'template_fallback';
+    discoveries_narrated?: number;
+    ai_call_succeeded?: boolean;
+    template_fallback?: boolean;
   } | undefined;
   const persistDiag = diag.persist as Record<string, number> | undefined;
 
@@ -1265,36 +1271,58 @@ export default function InsightRunDetail() {
           <div>
             <StepHeader
               title="Narrate"
-              description="Generates user-facing titles and summaries. AI narration via GPT-4o-mini for discoveries, templates for observations."
+              description="Generates plain-language titles and summaries via a single batched GPT-4o-mini call for all discoveries and observations. Falls back to templates if AI call fails."
               inputLabel={`Input: ${discoveriesCount + observationsCount} classified patterns from Filter`}
               outputLabel={`Output: narrated patterns to Persist`}
             />
 
             {/* AI Status Banner */}
             {(() => {
-              const aiCalls = narrateDiag?.ai_calls ?? 0;
-              const isAiDisabled = aiCalls === 0;
+              // Support both v2 (ai_calls) and v3 (mode/ai_call_succeeded) diagnostics
+              const isV3 = narrateDiag?.mode != null;
+              const aiActive = isV3
+                ? narrateDiag?.ai_call_succeeded === true
+                : (narrateDiag?.ai_calls ?? 0) > 0;
+              const narratedCount = isV3
+                ? narrateDiag?.discoveries_narrated ?? 0
+                : narrateDiag?.ai_calls ?? 0;
+              const templateCount = isV3
+                ? (narrateDiag?.template_fallback ? narrateDiag?.discoveries_narrated ?? 0 : 0)
+                : narrateDiag?.template_narratives ?? 0;
+
               return (
                 <>
                   <div style={{
                     padding: "12px 16px",
                     borderRadius: "8px",
                     marginBottom: "16px",
-                    background: isAiDisabled ? "rgba(234, 179, 8, 0.1)" : "rgba(34, 197, 94, 0.1)",
-                    border: `1px solid ${isAiDisabled ? "var(--warning)" : "var(--success)"}`,
-                    color: isAiDisabled ? "var(--warning)" : "var(--success)",
+                    background: aiActive ? "rgba(34, 197, 94, 0.1)" : "rgba(234, 179, 8, 0.1)",
+                    border: `1px solid ${aiActive ? "var(--success)" : "var(--warning)"}`,
+                    color: aiActive ? "var(--success)" : "var(--warning)",
                     fontSize: "13px",
                     fontWeight: 500,
                   }}>
-                    {isAiDisabled
-                      ? "AI NARRATION DISABLED \u2014 using template narratives for all patterns"
-                      : `AI narration active \u2014 ${aiCalls} AI call${aiCalls !== 1 ? "s" : ""} made`}
+                    {aiActive
+                      ? `AI narration active (batched) \u2014 ${narratedCount} pattern${narratedCount !== 1 ? "s" : ""} narrated`
+                      : narrateDiag?.template_fallback
+                        ? "AI narration failed \u2014 using template fallback for all patterns"
+                        : "AI narration disabled \u2014 using template narratives for all patterns"}
                   </div>
 
                   <div style={{ marginBottom: "20px", padding: "16px", background: "var(--bg-primary)", borderRadius: "8px" }}>
                     <div style={{ display: "flex", gap: "24px", flexWrap: "wrap" }}>
-                      <StatChip label="AI Calls" value={aiCalls} />
-                      <StatChip label="Template Narratives" value={narrateDiag?.template_narratives} />
+                      {isV3 ? (
+                        <>
+                          <StatChip label="Mode" value={narrateDiag?.mode === 'batch_ai' ? 'Batched AI' : 'Template'} />
+                          <StatChip label="Patterns Narrated" value={narratedCount} />
+                          <StatChip label="AI Succeeded" value={aiActive ? 'Yes' : 'No'} />
+                        </>
+                      ) : (
+                        <>
+                          <StatChip label="AI Calls" value={narrateDiag?.ai_calls ?? 0} />
+                          <StatChip label="Template Narratives" value={templateCount} />
+                        </>
+                      )}
                       {narrateDiag?.model && (
                         <div>
                           <span style={{ color: "var(--text-dimmed)", fontSize: "12px" }}>Model</span>
@@ -1317,12 +1345,6 @@ export default function InsightRunDetail() {
                   {!narrateDiag && (
                     <div className="empty-state">No narrate diagnostics available (pipeline may have exited early)</div>
                   )}
-
-                  {narrateDiag && aiCalls === 0 && (
-                    <div style={{ fontSize: "13px", color: "var(--text-dimmed)", fontStyle: "italic" }}>
-                      When AI narration is re-enabled: per-discovery narrative quality, token usage, latency per call
-                    </div>
-                  )}
                 </>
               );
             })()}

web/src/app/routes/admin/insights.tsx

@@ -341,7 +341,7 @@ export default function InsightsDashboard() {
                       <th title="Total rows fetched across all 6 source tables (daily_summary, sleep_sessions, activities, glucose_data, glucose_alarms, intraday_data). Needs >= 14 to proceed.">Data Rows</th>
                       <th title="Pattern candidates before BH correction → after. The arrow shows how many survived the Benjamini-Hochberg false discovery rate filter (alpha=0.10).">Candidates</th>
                       <th title="Strong patterns (p < 0.05, |effect size| > 0.3) actually inserted into user_discoveries. Limited to 3 active + 2 per week.">Discoveries</th>
-                      <th title="Weaker patterns inserted as observations. Limited to 5 active + 3 per week. Use template narratives (no AI call).">Observations</th>
+                      <th title="Weaker patterns inserted as observations. Narrated by the same batched AI call as discoveries.">Observations</th>
                       <th title="Data points outside plausibility bounds (e.g., HR > 220 bpm) that were excluded from analysis and flagged as data quality issues.">Outliers</th>
                       <th title="Wall-clock time for the entire pipeline (fetch through persist) for this user.">Duration</th>
                       <th title="Pipeline exit reason: 'success' = produced candidates, 'insufficient_data' = < 14 rows, 'no_candidates' = 0 passed significance test, 'no_bh_survivors' = all removed by BH correction, 'error' = exception thrown.">Status</th>