"""Criterion 2 under PRODUCTION (top2): the reused answer must be the CURRENT prompt's answer.
Noise divergence is allowed; answering the PREVIOUS turn's question is not.
W asks for ALPHA, P asks for BETA. The reused P must not answer ALPHA.
"""
import json, urllib.request, hashlib, re
URL="http://127.0.0.1:8300/v1/chat/completions"
def ask(prompt, mt=64, temp=0):
    body={"model":"glm-5.3-exl3","messages":[{"role":"user","content":prompt}],"max_tokens":mt,
          "temperature":temp,"return_token_ids":True}
    r=urllib.request.Request(URL,data=json.dumps(body).encode(),headers={"Content-Type":"application/json"})
    d=json.loads(urllib.request.urlopen(r,timeout=3600).read())
    st=d.get("tensorfold",{}) or {}
    m=d["choices"][0]["message"]
    txt=(m.get("content") or "") + "\n" + (m.get("reasoning_content") or "")
    ids=st.get("token_ids") or []
    return st.get("cached"), hashlib.sha256(json.dumps(ids).encode()).hexdigest()[:16], txt.upper(), ids
unit="The service validated the payload and emitted a structured record with stable ids. "
stale=0; tot=0; noise=0
for it in range(6):
    base=unit*(70+it*9)
    W=base+"\nReply with exactly the single word: ALPHA"
    P=base+"\nReply with exactly the single word: BETA"
    cW,sW,tW,_=ask(W)
    cR,sR,tR,idR=ask(P)
    cC,sC,tC,idC=ask(P)
    beta_r = "BETA" in tR; alpha_r = "ALPHA" in tR and not beta_r
    beta_c = "BETA" in tC
    same = sR==sC
    tot+=1
    stale += alpha_r
    noise += (not same)
    print(f"it{it}: warm={cW} reused={cR} cold={cC} | reused_says_BETA={beta_r} says_ALPHA={alpha_r} "
          f"cold_says_BETA={beta_c} | identical={same}", flush=True)
print(f"\nSTALE (reused answered the WARM question): {stale}/{tot}")
print(f"noise-level divergence (sha differs, allowed): {noise}/{tot}")
print("### SEMANTIC DONE")