From 8e5f500557ed2159780b338526507bca9078e161 Mon Sep 17 00:00:00 2001
From: Annie Weng <126124914+Anniew31@users.noreply.github.com>
Date: Tue, 26 May 2026 17:27:20 -0400
Subject: [PATCH] fix: improve signal accuracy and optimize tweet ingestion

---
 api/cron/collect-tweets.ts         | 13 ++++++++++---
 src/analysis/keyword-matcher.ts    |  4 ++--
 src/analysis/sentiment-analyzer.ts | 14 ++++++--------
 src/analysis/signal-generator.ts   |  2 +-
 src/api/kalshi-client.ts           |  2 +-
 vercel.json                        |  4 ----
 6 files changed, 20 insertions(+), 19 deletions(-)

diff --git a/api/cron/collect-tweets.ts b/api/cron/collect-tweets.ts
index 20b5018..a2d0590 100644
--- a/api/cron/collect-tweets.ts
+++ b/api/cron/collect-tweets.ts
@@ -134,6 +134,8 @@ export default async function handler(
 
       totalCollected += result.tweets.length;
 
+      const tweetsToStore: AnalyzedTweet[] = [];
+
       for (const rawTweet of result.tweets) {
         totalAnalyzed++;
 
@@ -168,12 +170,17 @@ export default async function handler(
           collected_at: new Date().toISOString(),
         };
 
-        // Store tweet in KV
-        await storeTweet(analyzedTweet);
-        totalStored++;
+        tweetsToStore.push(analyzedTweet);
       }
+      await Promise.allSettled(
+        tweetsToStore.map(tweet => storeTweet(tweet))
+      );
+
+      totalStored += tweetsToStore.length;
     }
 
+  
+
     // Step 6: Update feed indices
     await updateFeedIndices();
 
diff --git a/src/analysis/keyword-matcher.ts b/src/analysis/keyword-matcher.ts
index d2da844..80820bd 100644
--- a/src/analysis/keyword-matcher.ts
+++ b/src/analysis/keyword-matcher.ts
@@ -853,8 +853,8 @@ function isPromotionalContent(text: string): boolean {
   }
 
   // Excessive emoji usage (often in spam)
-  const emojiCount = (text.match(/[\uD800-\uDFFF]/g) || []).length;
-  if (emojiCount > 15 && text.length < 200) {
+  const emojiCount = [...text].filter(c => (c.codePointAt(0) ?? 0) > 0xFFFF).length;
+  if (emojiCount > 8 && text.length < 200) {
     return true; // More than 15 emoji chars in a short tweet is suspicious
   }
 
diff --git a/src/analysis/sentiment-analyzer.ts b/src/analysis/sentiment-analyzer.ts
index 21d73b0..d7e8623 100644
--- a/src/analysis/sentiment-analyzer.ts
+++ b/src/analysis/sentiment-analyzer.ts
@@ -15,7 +15,7 @@ const BULLISH_KEYWORDS = [
   'bullish', 'moon', 'rally', 'pump', 'surge', 'soar', 'skyrocket',
   'buy', 'long', 'calls', 'green', 'win', 'winning', 'yes', 'definitely',
   'confirmed', 'happening', 'inevitable', 'obvious', 'clearly', 'certain',
-  'guarantee', 'lock', 'easy', 'confident', 'predict', 'will happen',
+  'guarantee', 'lock', 'easy', 'confident', 'predict', 'happen',
   'going to', 'up', 'rise', 'increase', 'gain', 'profit', 'success',
   'boom', 'growth', 'explosive', 'parabolic', 'breakout'
 ];
@@ -53,13 +53,11 @@ export function analyzeSentiment(tweetText: string): SentimentResult {
 
   for (let i = 0; i < words.length; i++) {
     const word = words[i].replace(/[^a-z]/g, '');
-    const prevWord = i > 0 ? words[i - 1].replace(/[^a-z]/g, '') : '';
-
-    // Check for negation
-    const isNegated = NEGATIONS.includes(prevWord);
-
-    // Check for strong modifier
-    const isStrong = STRONG_MODIFIERS.includes(prevWord);
+    // Look back 2 words so negation passes through a modifier
+    const prev1 = i > 0 ? words[i-1].replace(/[^a-z]/g,'') : '';
+    const prev2 = i > 1 ? words[i-2].replace(/[^a-z]/g,'') : '';
+    const isNegated = NEGATIONS.includes(prev1) || (STRONG_MODIFIERS.includes(prev1) && NEGATIONS.includes(prev2));
+    const isStrong = STRONG_MODIFIERS.includes(prev1) && !NEGATIONS.includes(prev2);
     const weight = isStrong ? 2 : 1;
 
     // Check bullish
diff --git a/src/analysis/signal-generator.ts b/src/analysis/signal-generator.ts
index 45c1eaf..a981a4c 100644
--- a/src/analysis/signal-generator.ts
+++ b/src/analysis/signal-generator.ts
@@ -173,7 +173,7 @@ function generateSuggestedAction(
   urgency: UrgencyLevel
 ): SuggestedAction {
   // Don't suggest action if edge is too low
-  if (edge < 0.10) {
+  if (edge < 0.05) {
     return {
       direction: 'HOLD',
       confidence: 0,
diff --git a/src/api/kalshi-client.ts b/src/api/kalshi-client.ts
index f1653a6..906bb88 100644
--- a/src/api/kalshi-client.ts
+++ b/src/api/kalshi-client.ts
@@ -163,7 +163,7 @@ function toMarket(km: KalshiMarket): Market {
     keywords: generateKeywords(km.title),
     yesPrice: +safeYes.toFixed(2),
     noPrice: safeNo,
-    volume24h: km.volume_24h ?? km.volume ?? 0,
+    volume24h: km.volume_24h ?? 0,
     url: marketUrl,
     category: inferCategory(km.series_ticker || km.event_ticker || km.ticker),
     lastUpdated: new Date().toISOString(),
diff --git a/vercel.json b/vercel.json
index 5d0587d..4c4c49d 100644
--- a/vercel.json
+++ b/vercel.json
@@ -44,10 +44,6 @@
     {
       "source": "/api/(.*)",
       "headers": [
-        {
-          "key": "Access-Control-Allow-Credentials",
-          "value": "true"
-        },
         {
           "key": "Access-Control-Allow-Origin",
           "value": "*"