CONVERSATION_SCRIPT = [
    ("user", "Merhaba, siparişim hakkında soru sormak istiyorum."),
    ("user", "Sipariş numaram XYZ-123."),
    ("user", "Kargo durumu ne?"),
    ("user", "Ne zaman teslim olur?"),
    ("user", "İstanbul'a ne kadar sürede gelir genelde?"),
    ("user", "Kapıda ödeme yapabilir miyim?"),
    # ... 44 daha (gerçek senaryo için)
]
# Toplam 50 user mesaj, modelin 50 cevabı = 100 mesaj

import anthropic
import time
 
client = anthropic.Anthropic()
SYSTEM = "Müşteri destek asistanısın. Türkçe ve net cevap ver. " * 30  # ~1500 token
KB = "..." * 5000  # ~12K KB
 
conversation = []
total_cost = 0.0
total_cache_creation = 0
total_cache_read = 0
 
for turn, (role, msg) in enumerate(CONVERSATION_SCRIPT, 1):
    # Conversation history'ye breakpoint ekle (incremental cache)
    if conversation:
        last = conversation[-1]
        history_cache = [
            *conversation[:-1],
            {
                **last,
                "content": [{"type": "text", "text": last["content"], "cache_control": {"type": "ephemeral", "ttl": "5m"}}],
            },
        ]
    else:
        history_cache = []
 
    messages = [*history_cache, {"role": role, "content": msg}]
 
    start = time.perf_counter()
    resp = client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=300,
        system=[
            {"type": "text", "text": KB + SYSTEM, "cache_control": {"type": "ephemeral", "ttl": "1h"}},
        ],
        messages=messages,
    )
    latency = time.perf_counter() - start
 
    u = resp.usage
    cw = u.cache_creation_input_tokens or 0
    cr = u.cache_read_input_tokens or 0
    total_cache_creation += cw
    total_cache_read += cr
 
    cost = (
        u.input_tokens / 1e6 * 3.0
        + cw / 1e6 * 3.75
        + cr / 1e6 * 0.30
        + u.output_tokens / 1e6 * 15.0
    )
    total_cost += cost
 
    hit_rate = cr / max(1, cr + cw) * 100
    if turn in [1, 5, 10, 20, 30, 40, 50]:
        print(f"Turn {turn:>3}: hit={hit_rate:>5.1f}% | cw={cw:>5} cr={cr:>6} | cost=${cost:.4f} | latency={latency:.2f}s")
 
    # Assistant cevabını ekle
    conversation.append({"role": "user", "content": msg})
    conversation.append({"role": "assistant", "content": resp.content[0].text})
 
# Özet
overall_hit_rate = total_cache_read / max(1, total_cache_read + total_cache_creation) * 100
print(f"\n═══ ÖZET ═══")
print(f"Toplam 50 turn maliyet: ${total_cost:.4f}  |  {total_cost*33.5:.2f} TL")
print(f"Overall cache hit rate: {overall_hit_rate:.1f}%")
print(f"Cache write toplam:     {total_cache_creation:>10,} token")
print(f"Cache read toplam:      {total_cache_read:>10,} token")