"""The agentic cost curve with the prefix cache ON + FIXED_CHUNKS (audit 258).
Baseline from audit 250.3 (cache off): 15.5, 29.0, 44.5, 54.0, 72.0, 85.3 s as the prompt grew 2.7k -> 15.9k.
"""
import json, time, urllib.request
URL="http://127.0.0.1:8300/v1/chat/completions"
def ask(msgs, mt=8):
    body={"model":"glm-5.3-exl3","messages":msgs,"max_tokens":mt,"temperature":0}
    r=urllib.request.Request(URL,data=json.dumps(body).encode(),headers={"Content-Type":"application/json"})
    t0=time.perf_counter(); d=json.loads(urllib.request.urlopen(r,timeout=3600).read()); dt=time.perf_counter()-t0
    st=d.get("tensorfold",{}) or {}
    return dt, d.get("usage",{}).get("prompt_tokens"), st.get("cached"), d["choices"][0]["message"].get("content") or ""
unit=("Log entry: the service handled a request, validated the payload, wrote two rows, and returned 201. ")
blk=unit*120
msgs=[{"role":"system","content":"You are an agent. Reply with just OK."}]
print(f"{'turn':>4s} {'prompt':>8s} {'cached':>8s} {'wall s':>8s}   (cache-off baseline)")
base=[15.5,29.0,44.5,54.0,72.0,85.3]
for t in range(1,7):
    msgs.append({"role":"user","content":f"{blk}\nStep {t}: acknowledge."})
    dt,pt,ca,got=ask(msgs)
    msgs.append({"role":"assistant","content":got or "OK"})
    b=base[t-1] if t<=len(base) else None
    sp=f"{b/dt:5.1f}x" if b else ""
    print(f"{t:4d} {pt:8} {str(ca):>8} {dt:8.2f}   {b}s  {sp}", flush=True)
print("### GROWTH DONE")