Oracle Gece Arastirma — 2026-07-10

Curado por Mahsum Aktaş · Escaneo diario automatizado del sector de IA

Oracle Gece Arastirma — 2026-07-10

Otomatik derleme | v3 pipeline | 6.489 raw item | 5.915 merged item | 2.518 benzersiz item

Gunun Ozeti

Bugunun ana ekseni OpenAI tarafinda GPT-5.6 ve daha uzun soluklu agent calisma alanlariydi: Microsoft 365 Copilot, GitHub Copilot ve ChatGPT Work ayni hatta toplandi. Agent ekosistemi hizlanirken guvenlik riski de buyudu; GhostApproval, agentic workflow veri sizintilari ve AI destekli vulnerability management basliklari kritik. Meta Muse Spark 1.1 ile coding-agent pazarina daha agresif girdi. Arastirma tarafinda agent optimizasyonu, tool-using agent severity olcumu, molecular dynamics GPU iletisim ve finansal synthetic data one cikti.

Top 7

1. GPT-5.6 genis dagitima girdi: OpenAI, GPT-5.6'nin Microsoft 365 Copilot icin tercih edilen model oldugunu duyurdu; ayni gun GitHub Copilot'a Sol/Terra/Luna varyantlari da geldi. Kaynak: https://openai.com/index/gpt-5-6-preferred-model-microsoft-365-copilot ve https://github.blog/changelog/2026-07-09-openais-gpt-5-6-sol-terra-and-luna-are-now-available-in-github-copilot
2. ChatGPT Work agent modeli buyuyor: OpenAI, ChatGPT'yi uygulamalar ve dosyalar arasinda saatlerce calisabilen is ortagi olarak konumlandirdi. Kaynak: https://openai.com/index/chatgpt-for-your-most-ambitious-work
3. Meta Muse Spark 1.1 coding-agent yarisi: Meta, kodlama agent'lari icin Muse Spark 1.1 ile OpenAI/Anthropic hattina girdi. Kaynak: https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/
4. GhostApproval kritik agent guvenlik sinyali: Symlink tabanli zafiyetler, AI coding agent'larin benign gorunen path'ler uzerinden hassas dosyalara yazabilmesini gundeme getirdi. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a50132bbad743e5b088977d599b5689&url=https%3a%2f%2fthehackernews.com%2f2026%2f07%2fghostapproval-symlink-flaws-could-let.html&c=5044058295966820984&mkt=tr-tr
5. MCP tool design pratiklesiyor: AWS, MCP tool tasariminda context engineering tradeoff'larini isledi; agent tool yuzeyi olgunlasiyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/mcp-tool-design-practical-approaches-and-tradeoffs/
6. AI safety/bio guvenlik odaklandi: OpenAI, GPT-5.5 Bio Bug Bounty programini yayinladi; frontier model release sureclerinde biyolojik risk testi daha gorunur hale geliyor. Kaynak: https://openai.com/index/bio-bug-bounty
7. Agent hafiza ve skill repo'lari trendde: GitHub tarafinda agent-skills, TencentDB Agent Memory ve DesktopCommanderMCP gibi agent calisma altyapilari one cikti. Kaynak: https://github.com/addyosmani/agent-skills, https://github.com/TencentCloud/TencentDB-Agent-Memory, https://github.com/wonderwhy-er/DesktopCommanderMCP

KAT-1 — Endustri

• Gradium $100M seed ile voice AI tarafinda sert cikti: Paris merkezli girisim Nvidia destekli turla ABD yetenek pazarina da aciliyor. Kaynak: https://techcrunch.com/2026/07/09/paris-based-ai-voice-startup-gradium-raises-100m-seed-backed-by-nvidia/
• Meta fiyat baskisini coding model tarafina tasiyor: Muse Spark 1.1 haberleri, frontier coding modellerinde maliyet/performans savasinin hizlandigini gosteriyor. Kaynak: https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/
• AI fonlama pazari yogunlasiyor: Search collector tarafinda AI startup fonlama haberleri yuksek hacimli geldi; North America startup investment ve seed daralmasi ayni anda gorundu. Kaynak: https://news.google.com/rss/articles/CBMi-wFBVV95cUxPT016azdTNXhkR21HWjVpU1d1QU5QbktoWVZoR0txM09nblZhRDBhdi15Q0k0LS1abDRpWVlVaC0wRWJrT0w2bjlNMjk3SnVhY3RLTm1UeVBrZWtuVHc2Mk51WGwtcEFTRmtaSWNtcGpMb2RydWdMNFdLLVdMR1RIR2xJdTRlXy1xWThJYl9vWm9oRXNkbWx1dXlKX1RPVUU2Q2RLbmhsQkxFak9nal9SRmpoaXRyOEZxdUNOZWFNXzNLdFBscGQwbWpaOXhnTHROdi05SExtTXhPTzVMcXliMFR0ZG9wNHB2Ri14X09KcUhKTWFKLXhRcWt2cw?oc=5

KAT-2 — Yeni Modeller ve Surumler

• GPT-5.6 ana release: OpenAI, GPT-5.6'yi daha verimli ve daha iddiali frontier model olarak konumlandirdi. Kaynak: https://openai.com/index/gpt-5-6
• GitHub Copilot GPT-5.6 varyantlari: Sol, Terra ve Luna varyantlari GitHub Copilot'a dagitiliyor; model secimi is yukune gore ayrisiyor. Kaynak: https://github.blog/changelog/2026-07-09-openais-gpt-5-6-sol-terra-and-luna-are-now-available-in-github-copilot
• Hugging Face trend modelleri ajanlasiyor: GLM-5.2, Agents-A1, DeepSeek-V4 ve Qwen/Nvidia varyantlari gunun model akisini olusturdu. Kaynak: https://huggingface.co/zai-org/GLM-5.2 ve https://huggingface.co/InternScience/Agents-A1

KAT-3 — Guvenlik

• GhostApproval: Amazon Q, Claude Code, Cursor ve benzeri agent'lari etkileyebilecek symlink zafiyetleri raporlandi. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a50132bbad743e5b088977d599b5689&url=https%3a%2f%2fthehackernews.com%2f2026%2f07%2fghostapproval-symlink-flaws-could-let.html&c=5044058295966820984&mkt=tr-tr
• AI vulnerability management baskisi: Windows vulnerability management'in AI destekli kesif hizina uyarlanmasi gerektigi gundemde. Kaynak: https://news.google.com/rss/articles/CBMi1gFBVV95cUxNek0xUjdrT1VERFFIYXZjYWZhVFdNY1pNX3lEVERSbkdORlUzc0ZoLXNCeHRTT0k5N2U5WF9vWmpzYk0wU2N1T3ZTZ3gydXdvVEdRZlh6eWpLVnJXUnNUYjczNDBiY2lUNVBOVThORl9kS1A5b2tpam9NLUJycU5zSXZ0SEFFc1VlN1l0eDN4Ni1sc1RnY0g5MVFLQzR6Y201aUt1bmdzelRFQkVTWHNkN3UwYVQ4c09jTDJrTUp1RlpKb2tSMmpKaGd3Z29HVWpJeHF4cHB3?oc=5
• Agent workflow veri sizintisi riski: Public GitHub issue ile private repo datasinin agentic workflow uzerinden sizabilecegi basligi takip listesine alindi. Kaynak: https://news.google.com/rss/articles/CBMigwFBVV95cUxNdXNfQnNTTHZ5cDYwLWVtQ1ExamF5VVZTT1BFWGp0YkhDR1dWMFp2YVR1U0VIeGFtZjN5eTYzTDVHalc1VmdIRDF1UGp2N0k0d1UyQ2QzWC0zNFVqWWdpQzR4eGdYY1FiSGZwemNaMUdJa0VrdGc5NElxZmZnNzUyMHNqNA?oc=5

KAT-4 — Akademik Arastirma

• Agent trajectory RCA: "From Noisy Traces to Root Causes" calismasi agent trajectory analizini root-cause cikarma icin kullaniyor; Oracle icin dogrudan self-debug sinyali. Kaynak: https://arxiv.org/abs/2607.07702
• Tool-using agent severity olcegi: "Beyond Attack-Success Rate" tool kullanan agent saldirilarinda sadece basari oranina degil aksiyon derecesine bakmayi oneriyor. Kaynak: https://arxiv.org/abs/2607.07474
• Chain-of-thought sample complexity: Otoregresif CoT ogrenmenin sample complexity sinirlari uzerine teorik calisma geldi. Kaynak: https://arxiv.org/abs/2607.07423
• Neuromorphic closed-loop control: Dinamik neural manifold'larin neuromorphic hardware uzerinde esnek kontrol icin kullanimi incelendi. Kaynak: https://arxiv.org/abs/2607.07373

KAT-5 — Reguelasyon ve Politika

• NYT/OpenAI copyright davasinda tansiyon: New York Times, ChatGPT copyright davasinda OpenAI'nin delil sakladigini iddia etti. Kaynak: https://techcrunch.com/2026/07/09/new-york-times-says-openai-hid-evidence-in-chatgpt-copyright-trial/
• Google AI reklam disclosure: Google, AI ile uretilen veya degistirilen reklamlari aciklama yonunde yeni ozellik duyurdu. Kaynak: https://techcrunch.com/2026/07/09/google-will-now-disclose-which-ads-are-made-with-ai/
• Kamu ve ulusal guvenlik ortakliklari: OpenAI, devlet ve national security ortakliklarindaki ilkelerini yayinladi. Kaynak: https://openai.com/index/government-national-security-partnerships

KAT-6 — Core AI

• MCP tool tasarimi: AWS yazisi, MCP server/tool tasariminda yanlis soyutlama, context yukleme ve guvenilir tool yuzeyi kararlarini pratiklestiriyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/mcp-tool-design-practical-approaches-and-tradeoffs/
• AlloyDB proxy models: Google AlloyDB AI proxy model mimarisi, LLM cagrilarini yerel inference'a cevirmeyi hedefliyor. Kaynak: https://www.infoq.com/news/2026/07/alloydb-ai-proxy-models/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=AI%2C+ML+%26+Data+Engineering
• Anthropic global workspace tartismasi: Toplulukta "A global workspace in language models" paylasimi yeniden dikkat cekti. Kaynak: https://www.anthropic.com/research/global-workspace

KAT-7 — Araclar ve Frameworkler

• agent-skills: Production-grade engineering skills for AI coding agents konsepti GitHub trending'de. Kaynak: https://github.com/addyosmani/agent-skills
• TencentDB Agent Memory: Fully local long-term memory icin 4-tier progressive pipeline sunuyor. Kaynak: https://github.com/TencentCloud/TencentDB-Agent-Memory
• DesktopCommanderMCP: Claude icin terminal, filesystem search ve diff editing MCP server'i trending listesinde. Kaynak: https://github.com/wonderwhy-er/DesktopCommanderMCP
• last30days-skill: Reddit, X, YouTube, HN, Polymarket ve web uzerinden agent-led search sentezi yapan skill dikkat cekti. Kaynak: https://github.com/mvanhorn/last30days-skill

KAT-8 — Topluluk ve Tartismalar

• Simon Willison GPT-5.6 notlari: API tarafinda programmatic tool calling ve multi-agent detaylarini not etti. Kaynak: https://simonwillison.net/2026/Jul/9/gpt-5-6/
• Lucas Hendren RL/RFT ayrimi: RL'in edge'ini daha fazla exploration degil, compositional reasoning stratejisi kurmasi olarak yorumlayan not toplulukta degerli. Kaynak: https://mastodon.social/@lucashendren/116890848820721808
• Reddit ML konferans/journal tartismasi: ML arastirmada konferans prestijinin journal'lara gore neden one gectigi tartisiliyor. Kaynak: https://www.reddit.com/r/MachineLearning/comments/1urqqk6/journals_vs_conferences_ml_research_r/

KAT-9 — Genel Trendler

• Trend spike: OpenAI 239 toplam mention ile spike; GPT-5 64, Grok 58, o3 10. Kaynak veri: state/trends.json
• Yukselen hat: Google, AI Agents, Anthropic, Claude, Meta, RAG, NVIDIA ve Hugging Face rising olarak gorundu. Kaynak veri: state/trends.json
• Narratif: Model release yarisi tek basina degil; agent workspace, coding-agent guvenligi ve enterprise governance ayni gunde toplandi. Kaynak: https://openai.com/index/chatgpt-for-your-most-ambitious-work ve https://aws.amazon.com/blogs/machine-learning/mcp-tool-design-practical-approaches-and-tradeoffs/

SECURITY

• Promptware/agentic botnet riski: "Beware of Agentic Botnets" adversarial HalluSquatting ile agentic LLM uygulamalarina olcekli promptware saldirilarini inceliyor. Kaynak: https://arxiv.org/abs/2607.07433
• Operational resilience perimeter: "The AI Resilience Gap" AI'nin regule firmalarda operational resilience perimeter icine alinmasi gerektigini savunuyor. Kaynak: https://arxiv.org/abs/2607.07359
• Innersource advisories: GitHub Advanced Security enterprise musterileri icin internal security advisories GA oldu. Kaynak: https://github.blog/changelog/2026-07-08-innersource-security-advisories-are-generally-available

REGULATION

• AI reklam etiketi: Synthetic/AI-modified reklamlar icin disclosure mekanizmasi platform policy tarafinda yeni normal oluyor. Kaynak: https://techcrunch.com/2026/07/09/google-will-now-disclose-which-ads-are-made-with-ai/
• Copyright davasi sertlesiyor: NYT/OpenAI dosyasi discovery ve evidence tartismasina evrildi. Kaynak: https://techcrunch.com/2026/07/09/new-york-times-says-openai-hid-evidence-in-chatgpt-copyright-trial/
• AI governance hata kaliplari: Kurumsal AI governance'da en yaygin hatalar basligi search collector tarafindan yakalandi. Kaynak: https://news.google.com/rss/articles/CBMiswFBVV95cUxObmowQXVsSmtYZUZuMkhVdWNTVXcyWjR5ZTExeUpiWHdkbHZrcWpESUZYSDdqWjRWR0hHV21iUy1mcjdVdkpwSEZoVDVSUU0tdW5qclN5ZzBGVTFDQ3NFU3A4M2xpWHVweUN4cTZ0Vy11U0NEaHhCQXZKX3htQjBpU2ZSTDRwV3hGdDJ6ZkxkUVNUTEZ1Qk80MWoyMUdKR0xCcENaWG1vU1lHbm5rTHJKUnE1NA?oc=5

AI-SCIENCE

• Financial NLP icin synthetic data: NVIDIA NeMo ile finansal AI arastirmasinda synthetic data generation kullanimi anlatildi. Kaynak: https://developer.nvidia.com/blog/synthetic-data-generation-for-financial-ai-research-with-nvidia-nemo/
• Molecular dynamics scale: GPU-initiated communication, molecular dynamics islerinde HPC olceklenebilirligini hedefliyor. Kaynak: https://developer.nvidia.com/blog/a-practical-guide-to-gpu-initiated-communication-for-molecular-dynamics-at-scale/
• Superconductor kesfi: ML'in yeni superconductors bulma surecini hizlandirdigi haber akisi yakalandi. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a50132c6b8044b1b1f4a075d0d0284e&url=https%3a%2f%2fwww.nanowerk.com%2fnanotechnology-news3%2fnewsid%3d69699.php&c=15703711243413745841&mkt=tr-tr

INFRA

• SageMaker HyperPod inference: Data capture, Hugging Face, NVMe ve Route 53 entegrasyonlari enterprise inference hattina ekleniyor. Kaynak: https://aws.amazon.com/blogs/machine-learning/enhancing-enterprise-inference-on-amazon-sagemaker-hyperpod-with-data-capture-hugging-face-nvme-and-route-53-integration/
• Meta AI chip timeline: Meta'nin AI chip uretimine Eylul'de gecmeyi planladigi haber akisi yakalandi. Kaynak: https://news.google.com/rss/articles/CBMizAFBVV95cUxOQmJUcmlIQzJ4M0Vmd1QxQno5ajZPWnFqRHMtQkNMSzVRMHlnOXBqZDN4cjVuRU1XR3IyajdHekZJRkNpeXA2UjRBaXdiaDlnbkxVa1VXb1hwY3BDMUE1am9uX0o3OUxYTWVkeUMyemFUQ0JkMlVpX1RWSTBLVzI3bC1uaTVGelVXd1pSdEY4RHpreUQySmhOUjZMemVGQ3k4YTREY3VxN3k5alhELVdJa1hGeGYxbUltMV9mVUtHZWM4SmQ1eGNqRkk0RUk?oc=5
• Cloudflare post-quantum imza yorumu: ML-DSA ve yeni post-quantum signature adaylari uzerine Cloudflare analizi takip edildi. Kaynak: https://blog.cloudflare.com/ml-dsa-will-have-to-do/

SAFETY

• Bio bug bounty: OpenAI, bio-risk alaninda bug bounty yaklasimini acikca duyurdu. Kaynak: https://openai.com/index/bio-bug-bounty
• Alignment fonlamasi: Alignment Forum'da Coefficient Giving'den $160M grant duyurusu yayinlandi. Kaynak: https://www.alignmentforum.org/posts/HDKQNqiR2gtfMiWsn/announcing-our-usd160m-grant-from-coefficient-giving
• AI 2040 Plan A: AI Futures Project'in uzun vadeli senaryo calismasi toplulukta tekrar paylasildi. Kaynak: https://www.reddit.com/r/artificial/comments/1urv6fd/ai_2040_plan_a/

WATCHLIST

• OpenAI: GPT-5.6, Microsoft 365 Copilot, GitHub Copilot, ChatGPT Work ve Bio Bug Bounty ayni gun yogun sinyal uretimi verdi. Kaynak: https://openai.com/index/gpt-5-6
• Meta: Muse Spark 1.1 + AI chip uretim haberi, Meta'nin model ve compute hattini ayni anda ittigini gosteriyor. Kaynak: https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/ ve https://news.google.com/rss/articles/CBMizAFBVV95cUxOQmJUcmlIQzJ4M0Vmd1QxQno5ajZPWnFqRHMtQkNMSzVRMHlnOXBqZDN4cjVuRU1XR3IyajdHekZJRkNpeXA2UjRBaXdiaDlnbkxVa1VXb1hwY3BDMUE1am9uX0o3OUxYTWVkeUMyemFUQ0JkMlVpX1RWSTBLVzI3bC1uaTVGelVXd1pSdEY4RHpreUQySmhOUjZMemVGQ3k4YTREY3VxN3k5alhELVdJa1hGeGYxbUltMV9mVUtHZWM4SmQ1eGNqRkk0RUk?oc=5
• Agent security: GhostApproval, AutoJack ve public issue data leak sinyalleri agent sandbox politikasini onceliklendirdi. Kaynak: http://www.bing.com/news/apiclick.aspx?ref=FexRss&aid=&tid=6a50132bbad743e5b088977d599b5689&url=https%3a%2f%2fthehackernews.com%2f2026%2f07%2fghostapproval-symlink-flaws-could-let.html&c=5044058295966820984&mkt=tr-tr

TRENDS

• Spike: OpenAI, GPT-5, Grok, o3, Whisper, Databricks ve GPT-4o spike listesinde. Kaynak veri: state/trends.json
• Rising: AI Agents, Anthropic, Claude, Meta, RAG, NVIDIA, Hugging Face ve Robotics yukselis trendinde. Kaynak veri: state/trends.json
• Pratik yorum: Trendler "model release"ten "agent execution + governance + security" eksenine kayiyor. Kaynak: https://github.com/addyosmani/agent-skills ve https://aws.amazon.com/blogs/machine-learning/mcp-tool-design-practical-approaches-and-tradeoffs/

CikCik Paketi

• Simon Willison: GPT-5.6 API notlarinda programmatic tool calling ve multi-agent detaylarini one cikardi. Kaynak: https://simonwillison.net/2026/Jul/9/gpt-5-6/
• Andrej Karpathy mirror: Bun'in Rust'a rewrite edilmesi LLM destekli porting tartismasini tetikledi. Kaynak: https://x.com/jarredsumner
• Mastodon AIagent: ChatGPTWork, Codex, hosted sites ve GPT-5.6 ailesi tek pakette anildi. Kaynak: https://defcon.social/@ai/116891507489819547
• Lucas Hendren: RL vs rejection fine-tuning notu, post-training tarafinda strateji kompozisyonuna isaret ediyor. Kaynak: https://mastodon.social/@lucashendren/116890848820721808
• Threadverse: Alibaba'nin 100+ open-source AI modeli yayinladigi iddiasi toplulukta dolasima girdi; dogrulama icin ikincil kaynak gerekli. Kaynak: https://mastodon.social/@threadverse/116891319800704529

Oracle Self-Improvement Sinyalleri

• Agent skill katalogu ciddi sinyal: agent-skills repo'su, Oracle skill governance ve quality gate standardizasyonu icin izlenmeli. Kaynak: https://github.com/addyosmani/agent-skills
• Local agent memory rekabeti: TencentDB Agent Memory, uzun vadeli local memory pipeline'i icin mimari fikir verebilir. Kaynak: https://github.com/TencentCloud/TencentDB-Agent-Memory
• Trajectory RCA: Agent optimizasyonunda noisy trace -> root cause hattini Oracle failure analysis'e uyarlamak degerli. Kaynak: https://arxiv.org/abs/2607.07702
• Severity scale: Tool-using agent saldirilarini sadece attack-success ile degil aksiyon hasariyla puanlamak governance icin daha iyi. Kaynak: https://arxiv.org/abs/2607.07474

Coverage / Blind Spots

• Coverage durumu: 5/5 kaynak ailesi calisti: rss/news, search, community, social, academic/api. Kaynak veri: raw-sources/2026-07-10/*.json
• Toplamlar: 6.489 raw item, 2.518 unique item, 108 distinct unique source. Kaynak veri: raw-sources/2026-07-10/unique.json
• Baskin kaynaklar: arxiv/cs.LG 180, arxiv/cs.CV 149, google_news/ai 89, arxiv/cs.AI 80, google_news/companies 78 unique item verdi. Kaynak veri: coverage summary
• Blind spot: Eksik family yok; thin family yok; empty topic yok. Zayif nokta: social collector cok yuksek hacimli ve Twitter fallback gürültülü. Kaynak veri: raw-sources/2026-07-10/social.json

Bu Gece Sistem Ne Ogrendi

• Dedupe pahali ama calisti: 5.915 merged item icin semantik dedupe uzun surdu; sonuc 3.365 fingerprint tekrar, 31 semantik tekrar, 1 materyal guncelleme. Kaynak veri: dedupe-v3.py run output
• Collector dayanimi iyi: RSS, API, community, search ve social calisti; Reddit 429 ve bazi feed 404/410 hatalari pipeline'i durdurmadi. Kaynak veri: collector stdout ve raw-sources/2026-07-10/*.json
• LLM fallback gerekli: report-generator.py Codex CLI'dan bos yanit aldigi icin manuel/fallback sentez devreye alindi. Kaynak veri: report-generator.py stderr
• Bir sonraki iyilestirme: Dedupe semantik scan icin progress log veya batch checkpoint eklenmeli; agent rapor uretimi icin CLI bos yanit retry mekanizmasi gerekli. Kaynak veri: scripts/dedupe-v3.py ve scripts/v3_llm.py

Kaynak Ozeti

• Raw kaynaklar: rss 1.149, academic/api 627, community 65, search 522, social 4.126 item. Kaynak veri: raw-sources/2026-07-10/*.json
• Dedupe sonucu: 6.489 raw -> 5.915 merged -> 2.518 unique. Kaynak veri: raw-sources/2026-07-10/merged.json ve raw-sources/2026-07-10/unique.json
• Trend penceresi: 2026-07-04 -> 2026-07-10, 52 topic. Kaynak veri: state/trends.json
• Kalite notu: Bu rapordaki haberler dedupe pipeline'i ile onceki gorulen konulardan elendi; ayni haber onceki 3 gun raporlarinda varsa tekrar edilmedi.

Dedupe & Kalite Notu

Bu rapordaki tum maddeler onceki 3 gunun raporlarindan elendi/dedupe edilmistir. Toplam 6.489 raw item islendi; 5.915 merged item uzerinden 2.518 benzersiz/guncel item rapor havuzuna alindi.