feat: Feature 2.1 — Parlay Scan with correlation detection + monetization

POST /api/scan/parlay — authenticated parlay analysis:
- Supabase JWT auth middleware (auth.getUser verification)
- 5 correlation types detected between legs (same_game, same_team,
  same_player_conflicting, positive_correlation, blowout_cascade)
- Overall parlay grading (A/B/C/D) with correlation penalty adjustments
- Free tier: 5 scans/month, atomic scan count increment
- Scan 5: full analysis + personalized upgrade pitch
- Scan 6+: 403 block with upgrade pitch
- Pitch personalization from scan history (top stats, grades, tier rec)
- DB writes: picks + scan_sessions per scan

30 new tests, 158 total (131 Node.js + 27 Python), all passing

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

src/services/correlationEngine.js

@@ -0,0 +1,126 @@
+function detectCorrelations(analyzedLegs, spreads) {
+  const flags = [];
+
+  for (let i = 0; i < analyzedLegs.length; i++) {
+    for (let j = i + 1; j < analyzedLegs.length; j++) {
+      const a = analyzedLegs[i];
+      const b = analyzedLegs[j];
+
+      const aGame = a.reasoning?.steps?.season_avg ? getGameKey(a) : null;
+      const bGame = b.reasoning?.steps?.season_avg ? getGameKey(b) : null;
+      const sameGame = aGame && bGame && aGame === bGame;
+
+      const aTeam = a._team;
+      const bTeam = b._team;
+
+      // 1. same_player_conflicting
+      if (a.player.toLowerCase() === b.player.toLowerCase()) {
+        if (isConflicting(a, b)) {
+          flags.push({
+            type: 'same_player_conflicting',
+            legs: [i, j],
+            detail: `${a.player}: ${a.stat_type} ${a.direction} conflicts with ${b.stat_type} ${b.direction}`,
+            impact: 'major_negative',
+          });
+        } else {
+          // 4. positive_correlation (same player, complementary)
+          flags.push({
+            type: 'positive_correlation',
+            legs: [i, j],
+            detail: `${a.player}: ${a.stat_type} ${a.direction} + ${b.stat_type} ${b.direction} are correlated`,
+            impact: 'positive',
+          });
+        }
+        continue;
+      }
+
+      if (!sameGame) continue;
+
+      // 2. same_game_same_team
+      if (aTeam && bTeam && aTeam === bTeam) {
+        flags.push({
+          type: 'same_game_same_team',
+          legs: [i, j],
+          detail: `${a.player} and ${b.player} are both ${aTeam} — usage overlap possible`,
+          impact: 'minor_negative',
+        });
+        continue;
+      }
+
+      // 3. same_game_opposing_players
+      if (a.stat_type === b.stat_type && a.direction === 'over' && b.direction === 'over') {
+        flags.push({
+          type: 'same_game_opposing_players',
+          legs: [i, j],
+          detail: `${a.player} and ${b.player} in same game, both ${a.stat_type} overs`,
+          impact: 'minor_negative',
+        });
+      }
+    }
+  }
+
+  // 5. blowout_cascade — 2+ legs from a high-spread game
+  const gameLegs = groupByGame(analyzedLegs);
+  for (const [gameKey, indices] of Object.entries(gameLegs)) {
+    if (indices.length < 2) continue;
+    const gameSpread = findSpreadForGame(analyzedLegs[indices[0]], spreads);
+    if (gameSpread != null && Math.abs(gameSpread) > 8) {
+      flags.push({
+        type: 'blowout_cascade',
+        legs: indices,
+        detail: `${indices.length} legs from a game with ${gameSpread > 0 ? '+' : ''}${gameSpread} spread — blowout risk compounds`,
+        impact: 'major_negative',
+      });
+    }
+  }
+
+  return flags;
+}
+
+function getGameKey(leg) {
+  // Use home_team + away_team from reasoning to identify the game
+  const sit = leg.reasoning?.steps?.situational;
+  const lineCmp = leg.reasoning?.steps?.line_comparison;
+  // Fallback: use the first line's game context
+  // We'll use _gameTime attached by the scan service
+  return leg._gameTime || null;
+}
+
+function isConflicting(a, b) {
+  // Same player, opposite directions on related stats
+  if (a.direction !== b.direction) return true;
+
+  // Over points + under PRA (points is component of PRA)
+  const complementary = [
+    ['points', 'pra'], ['rebounds', 'pra'], ['assists', 'pra'],
+  ];
+  for (const [s1, s2] of complementary) {
+    if ((a.stat_type === s1 && b.stat_type === s2) || (a.stat_type === s2 && b.stat_type === s1)) {
+      if (a.direction !== b.direction) return true;
+    }
+  }
+
+  return false;
+}
+
+function groupByGame(legs) {
+  const groups = {};
+  for (let i = 0; i < legs.length; i++) {
+    const key = legs[i]._gameTime;
+    if (!key) continue;
+    if (!groups[key]) groups[key] = [];
+    groups[key].push(i);
+  }
+  return groups;
+}
+
+function findSpreadForGame(leg, spreads) {
+  if (!spreads || !leg._team) return null;
+  const spread = spreads.find((s) =>
+    s.home_team === leg._team || s.away_team === leg._team
+  );
+  if (!spread) return null;
+  return spread.home_spread;
+}
+
+module.exports = { detectCorrelations };

src/services/parlayGrader.js

@@ -0,0 +1,63 @@
+function gradeParlayFromLegs(legResults, correlationFlags) {
+  if (legResults.length === 0) {
+    return { grade: 'D', confidence: 30, composite: 0 };
+  }
+
+  // Extract leg composites and grades
+  const legComposites = legResults.map((l) => l._composite || 0);
+  const legGrades = legResults.map((l) => l.grade);
+  const legConfidences = legResults.map((l) => l.confidence);
+  const dCount = legGrades.filter((g) => g === 'D').length;
+
+  // Average of leg composites
+  const legAvg = legComposites.reduce((a, b) => a + b, 0) / legComposites.length;
+
+  // Correlation penalties
+  let correlationPenalty = 0;
+  let hasMajorNegative = false;
+  for (const flag of correlationFlags) {
+    if (flag.impact === 'minor_negative') correlationPenalty -= 0.3;
+    if (flag.impact === 'major_negative') {
+      correlationPenalty -= 1.0;
+      hasMajorNegative = true;
+    }
+  }
+
+  const parlayComposite = legAvg + correlationPenalty;
+
+  // Grade assignment
+  let grade;
+  if (parlayComposite >= 2.5 && dCount === 0 && !hasMajorNegative) {
+    grade = 'A';
+  } else if (parlayComposite >= 1.5 && dCount <= 1) {
+    grade = 'B';
+  } else if (parlayComposite >= 0.5) {
+    grade = 'C';
+  } else {
+    grade = 'D';
+  }
+
+  // 2+ D legs forces grade D
+  if (dCount >= 2) grade = 'D';
+
+  // Major negative caps at B
+  if (hasMajorNegative && (grade === 'A')) {
+    grade = 'B';
+  }
+
+  // Confidence: average of legs, adjusted for correlations
+  let confidence = legConfidences.reduce((a, b) => a + b, 0) / legConfidences.length;
+  for (const flag of correlationFlags) {
+    if (flag.impact === 'minor_negative') confidence -= 5;
+    if (flag.impact === 'major_negative') confidence -= 15;
+  }
+  confidence = Math.max(30, Math.min(95, Math.round(confidence)));
+
+  return {
+    grade,
+    confidence,
+    composite: Math.round(parlayComposite * 100) / 100,
+  };
+}
+
+module.exports = { gradeParlayFromLegs };

src/services/parlayScanService.js

@@ -0,0 +1,173 @@
+const { analyzeProp } = require('./propAnalyzer');
+const { getOdds } = require('./oddsService');
+const { detectCorrelations } = require('./correlationEngine');
+const { gradeParlayFromLegs } = require('./parlayGrader');
+const { generateUpgradePitch } = require('./upgradePitch');
+const { getSupabaseServiceClient } = require('../utils/supabase');
+
+async function scanParlay(user, legs) {
+  const supabase = getSupabaseServiceClient();
+  const isFree = user.tier === 'free';
+
+  // Scan count check (atomic for free tier)
+  if (isFree) {
+    if (user.scan_count >= 5) {
+      // Already exhausted — return 403 with pitch
+      const pitch = await generateUpgradePitch(supabase, user.id, null);
+      return {
+        blocked: true,
+        scan_count: user.scan_count,
+        scans_remaining: 0,
+        upgrade_pitch: pitch,
+      };
+    }
+  }
+
+  // Analyze all legs
+  const legResults = [];
+  for (const leg of legs) {
+    const result = await analyzeProp(leg);
+    legResults.push(result);
+  }
+
+  // Fetch odds data for correlation detection (spreads, game context)
+  let spreads = [];
+  try {
+    const oddsData = await getOdds('nba');
+    spreads = oddsData.spreads || [];
+
+    // Attach game context to leg results for correlation detection
+    for (const leg of legResults) {
+      const matchingProps = (oddsData.props || []).filter(
+        (p) => p.player.toLowerCase().includes(leg.player.toLowerCase())
+      );
+      if (matchingProps.length > 0) {
+        const prop = matchingProps[0];
+        leg._gameTime = prop.game_time;
+        // Resolve team from season avg
+        const seasonStep = leg.reasoning?.steps?.season_avg;
+        const team = leg._resolvedTeam || null;
+        // Use the team from the analysis context
+        if (leg.reasoning?.steps?.situational?.home_away?.context === 'home') {
+          leg._team = prop.home_team;
+        } else if (leg.reasoning?.steps?.situational?.home_away?.context === 'away') {
+          leg._team = prop.away_team;
+        }
+      }
+    }
+  } catch (_) {
+    // Correlation detection is best-effort
+  }
+
+  // Detect correlations
+  const correlationFlags = detectCorrelations(legResults, spreads);
+
+  // Grade the parlay
+  // Attach composite scores from individual analyses for parlay grading
+  for (const leg of legResults) {
+    // Reconstruct composite from the reasoning steps
+    const steps = leg.reasoning?.steps;
+    if (steps) {
+      const seasonDelta = steps.season_avg?.vs_line || 0;
+      const recentDelta = steps.recent_form?.vs_line || 0;
+      leg._composite = (Math.abs(seasonDelta) + Math.abs(recentDelta)) / 2;
+    } else {
+      leg._composite = 0;
+    }
+  }
+
+  const { grade: parlayGrade, confidence: parlayConfidence } = gradeParlayFromLegs(
+    legResults,
+    correlationFlags
+  );
+
+  // Write to database
+  const pickIds = [];
+  for (const leg of legResults) {
+    const { data: pick, error } = await supabase
+      .from('picks')
+      .insert({
+        user_id: user.id,
+        player: leg.player,
+        stat_type: leg.stat_type,
+        line: leg.line,
+        book: leg.book || 'unknown',
+        direction: leg.direction,
+        grade: leg.grade,
+        edge_pct: leg.edge_pct,
+        reasoning: leg.reasoning?.summary || '',
+        kill_conditions: (leg.kill_conditions_triggered || []).map((k) => k.code),
+        confidence: leg.confidence,
+      })
+      .select('id')
+      .single();
+
+    if (pick) pickIds.push(pick.id);
+  }
+
+  // Write scan session
+  const { data: session } = await supabase
+    .from('scan_sessions')
+    .insert({
+      user_id: user.id,
+      legs: pickIds,
+      final_grade: parlayGrade,
+      kill_conditions: correlationFlags
+        .filter((f) => f.impact !== 'positive')
+        .map((f) => f.type),
+      correlation_notes: JSON.stringify(correlationFlags),
+    })
+    .select('id')
+    .single();
+
+  // Atomic scan count increment for free tier
+  let newScanCount = user.scan_count;
+  if (isFree) {
+    const { data: updated } = await supabase
+      .from('users')
+      .update({ scan_count: user.scan_count + 1 })
+      .eq('id', user.id)
+      .eq('scan_count', user.scan_count)
+      .select('scan_count')
+      .single();
+
+    newScanCount = updated?.scan_count ?? user.scan_count + 1;
+  }
+
+  // Build response legs (stripped of internal fields)
+  const responseLegs = legResults.map((leg, i) => ({
+    index: i,
+    player: leg.player,
+    stat_type: leg.stat_type,
+    line: leg.line,
+    direction: leg.direction,
+    grade: leg.grade,
+    confidence: leg.confidence,
+    edge_pct: leg.edge_pct,
+    kill_conditions: leg.kill_conditions_triggered || [],
+    reasoning_summary: leg.reasoning?.summary || '',
+  }));
+
+  // Generate upgrade pitch at scan 5
+  let upgradePitch = null;
+  if (isFree && newScanCount >= 5) {
+    upgradePitch = await generateUpgradePitch(supabase, user.id, {
+      grade: parlayGrade,
+      legs: responseLegs,
+    });
+  }
+
+  return {
+    blocked: false,
+    scan_id: session?.id || null,
+    parlay_grade: parlayGrade,
+    parlay_confidence: parlayConfidence,
+    correlation_flags: correlationFlags,
+    legs: responseLegs,
+    scan_count: newScanCount,
+    scans_remaining: isFree ? Math.max(0, 5 - newScanCount) : null,
+    upgrade_pitch: upgradePitch,
+  };
+}
+
+module.exports = { scanParlay };

src/services/upgradePitch.js

@@ -0,0 +1,86 @@
+async function generateUpgradePitch(supabase, userId, currentScanResults) {
+  // Fetch prior scan sessions + picks
+  const { data: sessions } = await supabase
+    .from('scan_sessions')
+    .select('id, final_grade, legs, created_at')
+    .eq('user_id', userId)
+    .order('created_at', { ascending: false })
+    .limit(10);
+
+  const { data: picks } = await supabase
+    .from('picks')
+    .select('stat_type, direction, grade, player')
+    .eq('user_id', userId)
+    .order('created_at', { ascending: false })
+    .limit(50);
+
+  const allPicks = picks || [];
+  const allSessions = sessions || [];
+  const totalScans = allSessions.length + 1; // +1 for current
+
+  // Count good grades (A or B)
+  const priorGrades = allSessions.map((s) => s.final_grade).filter(Boolean);
+  if (currentScanResults?.grade) priorGrades.push(currentScanResults.grade);
+  const goodCount = priorGrades.filter((g) => g === 'A' || g === 'B').length;
+
+  // Most common stat type
+  const statCounts = {};
+  for (const pick of allPicks) {
+    statCounts[pick.stat_type] = (statCounts[pick.stat_type] || 0) + 1;
+  }
+  // Include current scan legs
+  if (currentScanResults?.legs) {
+    for (const leg of currentScanResults.legs) {
+      const st = leg.stat_type;
+      statCounts[st] = (statCounts[st] || 0) + 1;
+    }
+  }
+  const topStatType = Object.entries(statCounts).sort((a, b) => b[1] - a[1])[0]?.[0] || 'props';
+
+  // Most common direction for top stat
+  const dirCounts = { over: 0, under: 0 };
+  for (const pick of allPicks) {
+    if (pick.stat_type === topStatType) {
+      dirCounts[pick.direction] = (dirCounts[pick.direction] || 0) + 1;
+    }
+  }
+  const topDirection = dirCounts.over >= dirCounts.under ? 'over' : 'under';
+
+  // Average leg count
+  const legCounts = allSessions.map((s) => (s.legs || []).length);
+  if (currentScanResults?.legs) legCounts.push(currentScanResults.legs.length);
+  const avgLegs = legCounts.length > 0
+    ? legCounts.reduce((a, b) => a + b, 0) / legCounts.length
+    : 3;
+
+  // Unique players scanned
+  const uniquePlayers = new Set(allPicks.map((p) => p.player));
+
+  // Compliment
+  let compliment;
+  if (goodCount >= 3) compliment = "you've got a good eye";
+  else if (goodCount >= 2) compliment = "you're getting sharper";
+  else if (goodCount >= 1) compliment = "BetonBLK is helping you filter";
+  else compliment = "let's find better edges together";
+
+  // Tier recommendation
+  const tierRecommended = (avgLegs > 4 || uniquePlayers.size >= 5) ? 'desk' : 'analyst';
+
+  const tierBenefit = tierRecommended === 'desk'
+    ? 'Desk tier adds full bet tracking, ROI analytics, and priority cascade alerts.'
+    : 'Analyst tier gives you unlimited scans plus line movement alerts so you never miss a soft number.';
+
+  const founderPrice = tierRecommended === 'desk' ? '$34.99/mo' : '$14.99/mo';
+  const standardPrice = tierRecommended === 'desk' ? '$49.99/mo' : '$19.99/mo';
+
+  return {
+    hook: `You've scanned ${totalScans} parlays this month. ${goodCount} graded B or higher — ${compliment}.`,
+    insight: `Your best edge has been ${topStatType} ${topDirection}s. ${tierBenefit}`,
+    cta: `Unlock unlimited scans for ${founderPrice} (founder rate)`,
+    tier_recommended: tierRecommended,
+    founder_price: founderPrice,
+    standard_price: standardPrice,
+  };
+}
+
+module.exports = { generateUpgradePitch };