fix: truncated JSON, missing pitch for placeholder sites, token limit

- max_output_tokens 2048→4096 (main truncation fix)
- page snippet 2000→800 chars, search results capped at 600 chars
- JSON schema reordered: lead_quality/pitch_angle/services_needed first,
  so most important fields survive even if output is truncated
- RULES block in prompt: placeholder = HOT lead, pitch_angle is MANDATORY,
  services_needed must have ≥2 items, keep values ≤15 words to avoid truncation
- _parse_output: truncated JSON repair — closes open [] and {} brackets
  and strips trailing incomplete key-value before retrying json.loads

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

app/replicate_ai.py

@@ -67,7 +67,7 @@ def _build_prompt(a: dict, search_results: str = "") -> str:
     webmaster   = ", ".join(a.get("webmaster_verified") or []) or "none"
     lorem_str   = ", ".join(a.get("lorem_matches") or []) or "none"
     ph_str      = ", ".join(a.get("placeholder_matches") or []) or "none"
-    snippet     = (a.get("visible_text_snippet") or "")[:2000]
+    snippet     = (a.get("visible_text_snippet") or "")[:800]
     social_str  = ", ".join(a.get("social_links") or []) or "none detected"
     gmb_str     = f"✅ Found — {a.get('gmb_url','')}" if a.get("has_gmb") else "❌ Not detected"
     copyright_yr = a.get("copyright_year") or "not found"
@@ -138,47 +138,54 @@ Profiles found on site: {social_str}
 === PAGE TEXT SAMPLE ===
 {snippet}
 
-=== WEB SEARCH RESULTS (use these to find contact info, verify business details) ===
+=== WEB SEARCH RESULTS (use to find contacts, verify business identity) ===
-{search_results if search_results else "No search results available."}
+{(search_results or "No results.")[:600]}
 
 === INSTRUCTIONS ===
 The client sells: web redesign, SEO, hosting migration, SSL renewal,
 security audits, GDPR compliance, accessibility fixes, Google Ads,
-maintenance contracts, AI tools for SMEs, Google My Business setup/optimisation,
+maintenance contracts, AI tools for SMEs, GMB setup, social media management.
-social media management (Instagram, Facebook, LinkedIn, TikTok).
 
-IMPORTANT — use the WEB SEARCH RESULTS above to:
+RULES — you MUST follow all of these:
-1. Find any phone numbers, emails, or WhatsApp not visible on the homepage.
+1. A placeholder / minimal / blank site (few words, no images, no CMS) is one of
-2. Identify the business owner name if available.
+   the BEST leads — they need a complete website build + all digital services.
-3. Populate best_contact_value with a real phone/email you found.
+   Score it lead_quality=HOT or WARM and write an enthusiastic pitch.
-4. Use the copyright year and Last-Modified date to estimate when the site was last updated.
+2. pitch_angle is MANDATORY. Never leave it empty. Write 1 punchy Spanish sentence
-5. Determine the actual CMS from code signals and visible text (not just the heuristic).
+   tailored to the business type. Even "Hola, su web necesita una renovación
+   completa — podemos tenerla lista en 2 semanas." is better than nothing.
+3. services_needed must list at LEAST 2 services. For a blank/placeholder site
+   always include "diseño web" and "posicionamiento SEO".
+4. Use the WEB SEARCH RESULTS to find the real phone/email — put the best one
+   in best_contact_value.
+5. Use copyright_year + Last-Modified to estimate site_last_updated.
+6. Keep every string value SHORT (≤ 15 words). Arrays: max 4 items.
+   This keeps the JSON small and avoids truncation.
 
 Respond ONLY with valid JSON, no markdown fences, no text outside the JSON:
 {{
-  "summary": "2-3 sentence executive summary of the site's state",
+  "lead_quality": "HOT|WARM|COLD",
+  "lead_reasoning": "1-2 sentences why",
+  "pitch_angle": "1 punchy cold-outreach sentence in Spanish — NEVER empty",
+  "services_needed": ["service1","service2"],
+  "best_contact_channel": "email|phone|whatsapp|social|web_form|unknown",
+  "best_contact_value": "real email/phone from page or search results",
+  "all_contacts": {{"emails":[],"phones":[],"whatsapp":[],"social":[]}},
+  "summary": "2-3 sentence executive summary",
   "site_quality_score": <0-10>,
-  "content_issues": ["specific issues found in page content"],
+  "cms_detected": "wordpress|wix|custom|unknown",
-  "performance_notes": "load time, size, mobile assessment",
+  "site_last_updated": "year or estimate",
-  "seo_status": "SEO health — what's missing or broken",
-  "hosting_notes": "ASN/ISP name, EU vs non-EU, any concerns",
-  "gdpr_compliance": "cookie banner status, privacy policy, GDPR gaps",
-  "accessibility_issues": ["specific a11y problems found"],
-  "cms_detected": "wordpress|wix|squarespace|custom|unknown",
-  "site_last_updated": "year or estimate, e.g. '2019' or 'likely 2021'",
   "kit_digital_confirmed": true/false,
+  "kit_digital_reasoning": "1 sentence",
   "has_gmb": true/false,
   "has_social_media": true/false,
-  "kit_digital_reasoning": "1 sentence",
   "is_local_sme": true/false,
-  "lead_quality": "HOT|WARM|COLD",
+  "urgency_signals": ["issue1","issue2"],
-  "lead_reasoning": "1-2 sentences",
+  "content_issues": ["issue1"],
-  "best_contact_channel": "email|phone|whatsapp|social|web_form|unknown",
+  "accessibility_issues": ["issue1"],
-  "best_contact_value": "actual email/phone/URL or empty string",
+  "performance_notes": "brief",
-  "all_contacts": {{"emails":[],"phones":[],"whatsapp":[],"social":[]}},
+  "seo_status": "brief",
-  "pitch_angle": "1 cold-outreach sentence in Spanish",
+  "hosting_notes": "brief",
-  "services_needed": ["service1","service2"],
+  "gdpr_compliance": "brief",
-  "urgency_signals": ["specific urgent issues like 'SSL expires in 12 days', 'lorem ipsum on homepage', 'no cookie banner'"],
   "outreach_notes": "sales rep context"
 }}"""
 
@@ -187,13 +194,25 @@ def _parse_output(raw: str) -> dict:
     text = re.sub(r"```(?:json)?", "", raw).strip().rstrip("`").strip()
     m = re.search(r"\{[\s\S]+\}", text)
     if m:
+        candidate = m.group(0)
         try:
-            return json.loads(m.group(0))
+            return json.loads(candidate)
+        except json.JSONDecodeError:
+            # Truncated JSON: close any open arrays/objects and retry
+            fixed = candidate
+            # Count unclosed brackets
+            depth_obj = fixed.count("{") - fixed.count("}")
+            depth_arr = fixed.count("[") - fixed.count("]")
+            # Strip trailing incomplete key-value (e.g. `,"foo": "bar` with no closing `"`)
+            fixed = re.sub(r',\s*"[^"]*"?\s*:\s*[^,\}\]]*$', '', fixed)
+            fixed += "]" * max(0, depth_arr) + "}" * max(0, depth_obj)
+            try:
+                return json.loads(fixed)
             except json.JSONDecodeError:
                 pass
     logger.warning("Could not parse Gemini JSON output, raw: %s", raw[:300])
     return {
-        "summary": raw[:400],
+        "summary": raw[:400] if raw.strip() else "AI assessment failed — no output.",
         "lead_quality": "COLD",
         "best_contact_channel": "unknown",
         "best_contact_value": "",
@@ -220,7 +239,7 @@ async def assess_domain(analysis: dict) -> dict:
                 "top_p":   0.9,
                 "temperature": 0.2,
                 "thinking_level": "low",
-                "max_output_tokens": 2048,
+                "max_output_tokens": 4096,
             }
         }
         try: