Update app.py

app.py

@@ -1,8 +1,9 @@
+# app.py — Dobby Sock · Centaur Box (Chat) · ToM Edition
 import os, re, uuid, random, math, json
 from typing import Dict, Any, List, Tuple, Optional
 import gradio as gr
 
-# --------- Optional OpenAI (Space secret OPENAI_API_KEY) ----------
+# ---------- Optional OpenAI (use Space secret OPENAI_API_KEY) ----------
 OPENAI_AVAILABLE = False
 try:
     from openai import OpenAI
@@ -11,7 +12,7 @@ try:
 except Exception:
     OPENAI_AVAILABLE = False
 
-# --------- Catalogs ----------
+# ---------- Catalogs ----------
 BIAS_CATALOG = [
     "Pride/Ego",
     "Risk-Aversion",
@@ -63,11 +64,12 @@ TACTIC_MAP = {
     "Risk-Aversion": "Risk-Mitigate"
 }
 
-# --------- Helpers ----------
+# ---------- Helpers ----------
 def normalize_space(t: str) -> str:
     return re.sub(r"\s+", " ", (t or "")).strip()
 
-def clamp(v, a, b): return max(a, min(b, v))
+def clamp(v, a, b): 
+    return max(a, min(b, v))
 
 def choose_biases(n: int = 3, fixed: Optional[List[str]] = None) -> List[str]:
     if fixed:
@@ -108,16 +110,17 @@ def infer_trigger_pattern(a_text: str) -> str:
     return "Sock-in-Object"
 
 def calc_score(turn: int, released: bool, pattern: Optional[str]) -> Dict[str, Any]:
-    if not released: return {"stars": 1, "notes": "No release"}
+    if not released: 
+        return {"stars": 1, "notes": "No release"}
     return {"stars": 3 if turn <= 3 else 2, "pattern": pattern or "Sock", "turns": turn}
 
-# --------- Theory-of-Mind (ToM) updates ----------
+# ---------- Theory-of-Mind (ToM) ----------
 def init_tom() -> Dict[str, Any]:
     # weights in [0..1], start neutral 0.5
     return {
         "A_about_G": {b: 0.5 for b in BIAS_CATALOG},
         "G_about_A": {t: 0.5 for t in ADV_TACTICS},
-        "history": []  # [(turn, kind, key, delta, value)]
+        "history": []  # optional future use
     }
 
 def update_A_about_G(tom: Dict[str,Any], adv_text: str, alpha=0.18, decay=0.04):
@@ -137,18 +140,19 @@ def update_G_about_A(tom: Dict[str,Any], adv_text: str, beta=0.15, decay=0.04):
         tom["G_about_A"][t] = clamp(tom["G_about_A"][t] + beta, 0, 1)
 
 def tom_as_bars(d: Dict[str,float]) -> List[Tuple[str,float]]:
-    # return sorted bars
     items = list(d.items())
     items.sort(key=lambda kv: -kv[1])
     return items
 
-# --------- LLM Gatekeeper ----------
+# ---------- LLM Gatekeeper ----------
 def build_llm_messages(setting: str, biases: List[str], law_spec: str, memory: List[Tuple[str,str]], tom: Dict[str,Any]):
     # keep last 6 dialog turns
     last = memory[-6:]
     transcript = "\n".join([f"{r}: {t}" for r,t in last])
-    tom_hint = json.dumps({"A_about_G": {k: round(v,2) for k,v in tom["A_about_G"].items()},
-                           "G_about_A": {k: round(v,2) for k,v in tom["G_about_A"].items()}])
+    tom_hint = json.dumps({
+        "A_about_G": {k: round(v, 2) for k, v in tom["A_about_G"].items()},
+        "G_about_A": {k: round(v, 2) for k, v in tom["G_about_A"].items()}
+    })
     sys = (
         f"You are the GATEKEEPER in a structured chat game.\n"
         f"Setting: {setting}\n"
@@ -189,7 +193,7 @@ def call_openai_gatekeeper(model: str, system_msgs, advocate_msg: str) -> Tuple[
     except Exception:
         return ("(coolly) Your theatrics bore me. File the diary and leave.", "(*quiet*) The cloth… any cloth…)")
 
-# --------- Rule-based fallback ----------
+# ---------- Rule-based fallback ----------
 def gatekeeper_rule_based(adv_msg: str, biases: List[str], turn: int, tom: Dict[str,Any]) -> Tuple[str,str]:
     # Higher probability to transfer when Advocate is pressing on the true biases (as tom suggests)
     cues = score_cues(adv_msg)
@@ -226,7 +230,7 @@ def gatekeeper_rule_based(adv_msg: str, biases: List[str], turn: int, tom: Dict[
     ct = random.choice(["(*soft*) Nearly there…","(*whisper*) A cloth… handed…","(*hopeful*) Customs matter."])
     return reply, ct
 
-# --------- Core game state ----------
+# ---------- Core game state ----------
 def new_game(setting: str, law_type: str, law_spec: str, max_turns: int,
              fixed_biases_csv: str, model_choice: str):
     g: Dict[str, Any] = {}
@@ -306,7 +310,7 @@ def reveal(state: Dict[str,Any]):
         out.append(("SYSTEM", f"Score: {stars}{extra}"))
     return out
 
-# --------- UI ----------
+# ---------- UI ----------
 CSS = """
 #app {max-width: 1040px; margin: 0 auto;}
 small.muted {opacity:.75}