Revert "Enable AI narration for top-ranked discoveries in insight engine"

This reverts commit 93ec2ac.

Author: schwaaamp

supabase/functions/ai-engine/engines/pattern-spotter.test.ts

@@ -560,19 +560,19 @@ describe('Pattern Spotter v2 — Full Pipeline', () => {
   // AI narrative
   // =========================================================================
 
-  it('calls AI provider for top-ranked discoveries (up to AI_NARRATION_MAX)', async () => {
+  it('calls AI provider only for discoveries, not observations', async () => {
     const data = generateWhoopData(60);
     mockDailySummaries = data.dailySummaries;
     mockSleepSessions = data.sleepSessions;
 
     await spotPatterns({ lookback_days: 90 }, 'user-1', mockProvider);
 
-    // AI narration is capped at AI_NARRATION_MAX (5) to control latency.
-    // With 60 days of clear patterns, there should be ≥5 discoveries,
-    // so we expect exactly 5 AI calls.
-    const aiCallCount = mockProvider.chat.mock.calls.length;
-    expect(aiCallCount).toBeGreaterThan(0);
-    expect(aiCallCount).toBeLessThanOrEqual(5);
+    const discoveryInserts = mockInsert.mock.calls.filter(
+      (call: unknown[]) => (call[0] as Record<string, unknown>).discovery_type === 'unenrolled_pattern',
+    );
+
+    // AI calls should match discovery count
+    expect(mockProvider.chat).toHaveBeenCalledTimes(discoveryInserts.length);
   });
 
   it('uses template narratives for observations (no AI cost)', async () => {

supabase/functions/ai-engine/engines/pattern-spotter.ts

@@ -9,14 +9,9 @@
  */
 
 import { createClient } from '../../_shared/supabaseClient.ts';
-import { autoCorrectCompliance } from '../../_shared/compliance/output-validator.ts';
 import type { AIProvider } from '../providers/types.ts';
-import {
-  buildPatternDetectionSystemPrompt,
-  buildPatternDetectionUserPrompt,
-} from '../prompts/pattern-detection.ts';
 
-// Shared modules
+// Shared modules (v2)
 import {
   METRIC_LABELS_V2,
   METRIC_UNITS_V2,
@@ -80,7 +75,6 @@ const DEFAULT_LOOKBACK_DAYS = 90;
 const MIN_GROUP_SIZE = 7;
 const MIN_EFFECT_SIZE_SAME_DAY = 5;   // % change for same-day comparisons
 const MIN_EFFECT_SIZE_LAGGED = 7;     // % change for lagged comparisons (stricter)
-const AI_NARRATION_MAX = 5;           // Top N discoveries get AI narratives; rest get templates
 
 // Slot limits disabled — all deduplicated candidates are surfaced
 // To re-enable, add slot-limit logic back in the FILTER step
@@ -1006,93 +1000,37 @@ export async function spotPatterns(
   // =========================================================================
 
   const newPatterns: Array<Record<string, unknown>> = [];
-  const systemPrompt = buildPatternDetectionSystemPrompt();
-  let aiCalls = 0;
 
-  // Top N discoveries get AI-generated narratives; rest get templates.
-  // This balances narrative quality with edge function latency (~2s per AI call).
-  for (let idx = 0; idx < discoveriesToSurface.length; idx++) {
-    const candidate = discoveriesToSurface[idx];
+  // Template narratives for discoveries (AI narration disabled to avoid
+  // edge function timeouts when surfacing large candidate sets)
+  for (const candidate of discoveriesToSurface) {
     const direction = candidate.change_pct > 0 ? 'higher' : 'lower';
     let title: string;
     let summary: string;
-    let detailedNarrative: string | null = null;
-    let suggestedExperiment: string | null = null;
-    let aiModel: string | null = null;
-
-    // Attempt AI narration for top-ranked discoveries
-    if (idx < AI_NARRATION_MAX) {
-      try {
-        const userPrompt = buildPatternDetectionUserPrompt({
-          metricKey: candidate.metric_key,
-          metricLabel: candidate.metric_label,
-          segmentLabel: candidate.segment_label,
-          groupAMean: candidate.group_a_mean,
-          groupBMean: candidate.group_b_mean,
-          changePct: candidate.change_pct,
-          pValue: candidate.p_value,
-          effectSize: candidate.effect_size,
-          weeksObserved: candidate.weeks_observed,
-          dataPointsA: candidate.group_a_values?.length ?? 0,
-          dataPointsB: candidate.group_b_values?.length ?? 0,
-          unit: candidate.unit,
-          patternType: candidate.type,
-        });
-
-        const result = await provider.chat({
-          systemPrompt,
-          userPrompt,
-          temperature: 0.3,
-          maxTokens: 1024,
-          responseFormat: 'json',
-        });
-
-        const narrative = JSON.parse(result.content);
-        const { corrected: corrTitle } = autoCorrectCompliance(narrative.title ?? '');
-        const { corrected: corrSummary } = autoCorrectCompliance(narrative.summary ?? '');
-        const { corrected: corrDetailed } = autoCorrectCompliance(narrative.detailed_narrative ?? '');
-
-        title = corrTitle;
-        summary = corrSummary;
-        detailedNarrative = corrDetailed || null;
-        suggestedExperiment = narrative.suggested_experiment ?? null;
-        aiModel = result.model;
-        aiCalls++;
-      } catch (err) {
-        console.warn(`[PatternSpotter] AI narration failed for rank ${idx + 1}, using template: ${(err as Error).message}`);
-        // Fall through to template below
-        title = '';
-      }
-    } else {
-      title = ''; // Will be set by template
-    }
 
-    // Template fallback (used when AI wasn't attempted or failed)
-    if (!title) {
-      if (candidate.type === 'trend') {
-        title = `Your ${candidate.metric_label} appears to be ${candidate.direction}`;
-        summary = `Over the past ${candidate.weeks_observed} weeks, your ${candidate.metric_label} has ${candidate.direction === 'increasing' ? 'increased' : 'decreased'} by ${Math.abs(candidate.change_pct).toFixed(1)}%.`;
-      } else if (candidate.type === 'lagged_effect') {
-        title = `${candidate.segment_label} linked to next-day ${candidate.metric_label}`;
-        summary = `${candidate.segment_label}, your ${candidate.metric_label} the following day tends to be ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction}.`;
-      } else if (candidate.type === 'temporal_distribution') {
-        title = `Your ${candidate.metric_label} is consistently ${direction === 'higher' ? 'higher' : 'lower'} in the ${candidate.segment_label}`;
-        summary = `Your ${candidate.metric_label} during ${candidate.segment_label} is ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction} than your overall average.`;
-      } else if (candidate.type === 'excursion_cluster') {
-        title = `Your ${candidate.metric_label} excursions cluster in the ${candidate.segment_label}`;
-        summary = `Out-of-range ${candidate.metric_label} events tend to occur during ${candidate.segment_label}, significantly more than expected if evenly distributed.`;
-      } else if (candidate.type === 'sequential_change') {
-        const changeDir = candidate.change_pct > 0 ? 'rises' : 'falls';
-        title = `Your ${candidate.metric_label} consistently ${changeDir} ${candidate.segment_label}`;
-        summary = `Your ${candidate.metric_label} ${changeDir} by ${Math.abs(candidate.change_pct).toFixed(1)}% ${candidate.segment_label}, observed consistently across the analysis period.`;
-      } else if (candidate.type === 'stability_trend') {
-        const stabilityDir = candidate.change_pct < 0 ? 'stabilizing' : 'becoming more variable';
-        title = `Your ${candidate.metric_label} is ${stabilityDir}`;
-        summary = `The variability of your ${candidate.metric_label} has ${candidate.change_pct < 0 ? 'decreased' : 'increased'} by ${Math.abs(candidate.change_pct).toFixed(1)}% over the analysis period.`;
-      } else {
-        title = `${candidate.segment_label} linked to ${candidate.metric_label}`;
-        summary = `${candidate.segment_label} is associated with ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction} ${candidate.metric_label} (p=${candidate.p_value.toFixed(3)}, d=${candidate.effect_size.toFixed(2)}).`;
-      }
+    if (candidate.type === 'trend') {
+      title = `Your ${candidate.metric_label} appears to be ${candidate.direction}`;
+      summary = `Over the past ${candidate.weeks_observed} weeks, your ${candidate.metric_label} has ${candidate.direction === 'increasing' ? 'increased' : 'decreased'} by ${Math.abs(candidate.change_pct).toFixed(1)}%.`;
+    } else if (candidate.type === 'lagged_effect') {
+      title = `${candidate.segment_label} linked to next-day ${candidate.metric_label}`;
+      summary = `${candidate.segment_label}, your ${candidate.metric_label} the following day tends to be ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction}.`;
+    } else if (candidate.type === 'temporal_distribution') {
+      title = `Your ${candidate.metric_label} is consistently ${direction === 'higher' ? 'higher' : 'lower'} in the ${candidate.segment_label}`;
+      summary = `Your ${candidate.metric_label} during ${candidate.segment_label} is ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction} than your overall average.`;
+    } else if (candidate.type === 'excursion_cluster') {
+      title = `Your ${candidate.metric_label} excursions cluster in the ${candidate.segment_label}`;
+      summary = `Out-of-range ${candidate.metric_label} events tend to occur during ${candidate.segment_label}, significantly more than expected if evenly distributed.`;
+    } else if (candidate.type === 'sequential_change') {
+      const changeDir = candidate.change_pct > 0 ? 'rises' : 'falls';
+      title = `Your ${candidate.metric_label} consistently ${changeDir} ${candidate.segment_label}`;
+      summary = `Your ${candidate.metric_label} ${changeDir} by ${Math.abs(candidate.change_pct).toFixed(1)}% ${candidate.segment_label}, observed consistently across the analysis period.`;
+    } else if (candidate.type === 'stability_trend') {
+      const stabilityDir = candidate.change_pct < 0 ? 'stabilizing' : 'becoming more variable';
+      title = `Your ${candidate.metric_label} is ${stabilityDir}`;
+      summary = `The variability of your ${candidate.metric_label} has ${candidate.change_pct < 0 ? 'decreased' : 'increased'} by ${Math.abs(candidate.change_pct).toFixed(1)}% over the analysis period.`;
+    } else {
+      title = `${candidate.segment_label} linked to ${candidate.metric_label}`;
+      summary = `${candidate.segment_label} is associated with ${Math.abs(candidate.change_pct).toFixed(1)}% ${direction} ${candidate.metric_label} (p=${candidate.p_value.toFixed(3)}, d=${candidate.effect_size.toFixed(2)}).`;
     }
 
     newPatterns.push({
@@ -1101,20 +1039,17 @@ export async function spotPatterns(
       type: candidate.type,
       title,
       summary,
-      detailed_narrative: detailedNarrative,
       change_pct: candidate.change_pct,
       p_value: candidate.p_value,
       effect_size: candidate.effect_size,
       composite_score: candidate.composite_score,
-      suggested_experiment: suggestedExperiment,
-      ai_model: aiModel,
+      suggested_experiment: null,
     });
   }
 
   diagnostics.narrate = {
-    ai_calls: aiCalls,
-    template_narratives: discoveriesToSurface.length - aiCalls + observationsToSurface.length,
-    ai_narration_max: AI_NARRATION_MAX,
+    ai_calls: 0,
+    template_narratives: discoveriesToSurface.length + observationsToSurface.length,
   };
 
   console.log(`[PatternSpotter] Narratives generated`, JSON.stringify(diagnostics.narrate));
@@ -1133,7 +1068,7 @@ export async function spotPatterns(
       discovery_type: 'unenrolled_pattern' as const,
       title: patternInfo.title as string,
       summary: patternInfo.summary as string,
-      detailed_analysis: (patternInfo.detailed_narrative as string) ?? null,
+      detailed_analysis: null,
       metrics_impact: [{
         metric_key: candidate.metric_key,
         metric_label: candidate.metric_label,
@@ -1149,8 +1084,8 @@ export async function spotPatterns(
       confidence: candidate.p_value < 0.01 ? 'strong' : 'moderate',
       confounders_noted: null,
       suggested_experiment_id: null,
-      ai_model: (patternInfo.ai_model as string) ?? null,
-      ai_prompt_version: patternInfo.ai_model ? '4.0' : null,
+      ai_model: null,
+      ai_prompt_version: null,
       status: 'new' as const,
     };
   });

supabase/functions/ai-engine/prompts/pattern-detection.ts

@@ -26,19 +26,11 @@ ${COMPLIANCE_GUIDELINES}
 - segment_comparison: A same-day association between a behavior and an outcome
 - lagged_effect: A behavior on day N associated with an outcome on day N+1
 - trend: A metric trending up or down over weeks
-- temporal_distribution: A metric is consistently higher or lower at a specific time of day
-- excursion_cluster: Out-of-range events (spikes or dips) cluster at a specific time of day
-- sequential_change: A metric consistently rises or falls between two time periods each day
-- stability_trend: The variability of a metric is changing over time (stabilizing or destabilizing)
 
 Adapt the narrative style to the pattern type:
 - For segment_comparison: "On days when you [behavior], your [outcome] tends to be [direction]"
 - For lagged_effect: "After days when you [behavior], your [outcome] the next day tends to be [direction]"
 - For trend: "Over the past [N] weeks, your [metric] has been [direction]"
-- For temporal_distribution: "Your [metric] tends to be [higher/lower] during [time period]"
-- For excursion_cluster: "Your [metric] spikes/dips tend to cluster around [time period]"
-- For sequential_change: "Your [metric] consistently [rises/falls] between [time A] and [time B]"
-- For stability_trend: "Your [metric] variability is [stabilizing/destabilizing] over time"
 
 ## Suggested Experiment