fix: correct base64 encoding for Vision Analysis screenshots

- Previous chunked btoa approach encoded each chunk independently,
  producing corrupted base64 that Gemini couldn't parse (returned {})
- Now builds complete binary string first, then encodes once with btoa
- Added screenshot debug info to channel errors for diagnostics
- Confirmed: foundingYear 2004, doctors, gangnamunni data all extracted

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

supabase/functions/_shared/visionAnalysis.ts

@@ -42,12 +42,17 @@ export interface VisionAnalysisResult {
  * Firecrawl v2 returns a GCS URL (not base64). We download it and convert to base64
  * so Gemini Vision can consume it via inlineData.
  */
+// Track per-screenshot errors for debugging
+export const screenshotErrors: string[] = [];
+
 async function captureScreenshot(
   url: string,
   firecrawlKey: string,
 ): Promise<{ screenshotUrl: string; base64: string } | null> {
   try {
     console.log(`[vision] Capturing screenshot: ${url}`);
+    console.log(`[vision] Firecrawl key present: ${!!firecrawlKey}, length: ${firecrawlKey?.length}`);
+
     // Firecrawl v2: use "screenshot@fullPage" format (no separate screenshotOptions)
     const res = await fetchWithRetry(`${FIRECRAWL_BASE}/scrape`, {
       method: "POST",
@@ -64,13 +69,17 @@ async function captureScreenshot(
 
     if (!res.ok) {
       const errText = await res.text().catch(() => "");
-      console.error(`[vision] Screenshot failed for ${url}: HTTP ${res.status} — ${errText.slice(0, 200)}`);
+      const msg = `Screenshot HTTP ${res.status} for ${url}: ${errText.slice(0, 200)}`;
+      console.error(`[vision] ${msg}`);
+      screenshotErrors.push(msg);
       return null;
     }
     const data = await res.json();
     const screenshotUrl: string | null = data.data?.screenshot || null;
     if (!screenshotUrl) {
-      console.warn(`[vision] Screenshot response OK but no screenshot URL for ${url}. Keys: ${JSON.stringify(Object.keys(data.data || {}))}`);
+      const msg = `Screenshot OK but no URL for ${url}. Keys: ${JSON.stringify(Object.keys(data.data || {}))}`;
+      console.warn(`[vision] ${msg}`);
+      screenshotErrors.push(msg);
       return null;
     }
 
@@ -83,21 +92,32 @@ async function captureScreenshot(
       maxRetries: 1,
     });
     if (!imgRes.ok) {
-      console.error(`[vision] Failed to download screenshot image: HTTP ${imgRes.status}`);
+      const msg = `Screenshot download failed: HTTP ${imgRes.status} for ${url}`;
+      console.error(`[vision] ${msg}`);
+      screenshotErrors.push(msg);
       return null;
     }
     const imgBuffer = await imgRes.arrayBuffer();
     const bytes = new Uint8Array(imgBuffer);
 
-    // Use Deno's standard base64 encoding (efficient for large binaries)
-    const { encode: encodeBase64 } = await import("https://deno.land/std@0.224.0/encoding/base64.ts");
-    const base64 = encodeBase64(bytes);
+    // Convert to base64 — build binary string in chunks, then encode once
+    const chunkSize = 8192;
+    let binaryStr = "";
+    for (let i = 0; i < bytes.length; i += chunkSize) {
+      const chunk = bytes.subarray(i, Math.min(i + chunkSize, bytes.length));
+      for (let j = 0; j < chunk.length; j++) {
+        binaryStr += String.fromCharCode(chunk[j]);
+      }
+    }
+    const base64 = btoa(binaryStr);
 
     console.log(`[vision] Screenshot captured & converted: ${url} (${Math.round(base64.length / 1024)}KB base64, ${Math.round(bytes.length / 1024)}KB raw)`);
 
     return { screenshotUrl, base64 };
   } catch (err) {
-    console.error(`[vision] Screenshot error for ${url}:`, err instanceof Error ? err.message : err);
+    const msg = `Screenshot exception for ${url}: ${err instanceof Error ? err.message : String(err)}`;
+    console.error(`[vision] ${msg}`);
+    screenshotErrors.push(msg);
     return null;
   }
 }

supabase/functions/collect-channel-data/index.ts

@@ -2,7 +2,7 @@ import "@supabase/functions-js/edge-runtime.d.ts";
 import { createClient } from "https://esm.sh/@supabase/supabase-js@2";
 import type { VerifiedChannels } from "../_shared/verifyHandles.ts";
 import { PERPLEXITY_MODEL } from "../_shared/config.ts";
-import { captureAllScreenshots, runVisionAnalysis, type ScreenshotResult } from "../_shared/visionAnalysis.ts";
+import { captureAllScreenshots, runVisionAnalysis, screenshotErrors, type ScreenshotResult } from "../_shared/visionAnalysis.ts";
 import { fetchWithRetry, fetchJsonWithRetry, wrapChannelTask, type ChannelTaskResult } from "../_shared/retry.ts";
 
 const corsHeaders = {
@@ -395,17 +395,22 @@ Deno.serve(async (req) => {
           channelData.visionPerPage = vision.perPage;
         }
 
-        // Store screenshots (without base64 — just metadata for report)
+        // Store screenshots metadata (NOT base64 — use the GCS URL from Firecrawl)
         channelData.screenshots = screenshots.map(ss => ({
           id: ss.id,
-          url: ss.base64 ? `data:image/png;base64,${ss.base64}` : ss.url,
+          url: ss.url,  // GCS signed URL (valid ~7 days)
           channel: ss.channel,
           capturedAt: ss.capturedAt,
           caption: ss.caption,
           sourceUrl: ss.sourceUrl,
         }));
 
-        if (screenshots.length === 0) throw new Error("No screenshots captured");
+        if (screenshots.length === 0) {
+          const debugInfo = screenshotErrors.length > 0
+            ? screenshotErrors.join(" | ")
+            : "No errors recorded — check FIRECRAWL_API_KEY";
+          throw new Error(`No screenshots captured: ${debugInfo}`);
+        }
       }));
     }