Upload 18 files

app.py

@@ -66,16 +66,25 @@ def ui_explain_detail(ticker):
     if not raw:
         yield "Run the analysis and select a ticker first."
         return
-    # Same pattern as Sector Rotation: feed the deterministic ENGLISH raw read
-    # to the agent — no Chinese in, no Chinese out, fast (short prompt).
-    prompt = ("You are an equity analyst. Based ONLY on this factual read of a US "
-              "stock's multi-timeframe Chan-theory verdict, write a short plain-"
-              "English summary for a long-term holder: what's the situation today, "
-              "should they act or wait, and the key price levels. ≤90 words, no "
-              "disclaimers.\n\nRAW READ:\n" + raw)
+    # The full Chan ruling chain (Chinese) is kept BACKSTAGE in STATE and fed to
+    # the model alongside the English raw read, so the summary reflects the real
+    # multi-timeframe reasoning — but the chain is never shown and output is
+    # English only. (This restores the merged raw-read + ruling-chain logic.)
+    chain = automation.STATE.get("signals_details", {}).get(ticker or "", "")
+    chain_core = chain.split("日线买卖点逐项诊断")[0].strip()[:2000] if chain else ""
+    prompt = ("You are an equity analyst. Write a SHORT plain-English summary "
+              "(≤100 words) for a long-term holder of a US stock: the situation "
+              "today, whether to act or wait, and the key price levels.\n"
+              "Use the FACT LINE for the numbers, and the RULING CHAIN (a "
+              "Chinese multi-timeframe Chan-theory decision log) for the reasoning "
+              "— translate and synthesize it; output ENGLISH ONLY, no Chinese "
+              "characters, do not quote the log, no disclaimers.\n\n"
+              f"FACT LINE:\n{raw}")
+    if chain_core:
+        prompt += f"\n\nRULING CHAIN (translate & synthesize, don't quote):\n{chain_core}"
     yield "🤖 _Translator sub-agent (Qwen3-1.7B · llama.cpp) is summarizing…_"
     final = ""
-    for acc in llm_local.chat_stream(prompt, max_tokens=240, temperature=0.2,
+    for acc in llm_local.chat_stream(prompt, max_tokens=260, temperature=0.2,
                                      worker="translator"):
         final = acc
         yield "🤖 **AI narrative (Translator sub-agent · Qwen3-1.7B):**\n\n" + acc
@@ -304,8 +313,12 @@ with gr.Blocks(title="Chan Compass · US", **_style_kw) as demo:
                               elem_id="detail-log")
         traces_md = gr.Markdown(research_agent.list_traces())
         auto_log_timer = gr.Timer(2.0)
-        auto_log_timer.tick(lambda: "\n".join(automation.STATE["log"][-40:])
-                            or "(no log yet)", None, auto_log)
+
+        def _auto_tick():
+            log = "\n".join(automation.STATE["log"][-40:]) or "(no log yet)"
+            return log, research_agent.list_traces()
+
+        auto_log_timer.tick(_auto_tick, None, [auto_log, traces_md])
 
     with gr.Tab("🧠 Model"):
         gr.Markdown("All AI runs **locally** through **llama.cpp** (llama-cpp-python) with "

research_agent.py

@@ -330,7 +330,10 @@ def read_report(fname: str) -> str:
 
 def list_traces() -> str:
     try:
-        fs = sorted(os.listdir(paths.TRACES_DIR), reverse=True)[:20]
+        fs = [f for f in os.listdir(paths.TRACES_DIR) if f.endswith(".json")]
+        fs.sort(key=lambda f: os.path.getmtime(os.path.join(paths.TRACES_DIR, f)),
+                reverse=True)
+        fs = fs[:20]
         if not fs:
             return "_No traces yet — run a research note first._"
         return ("**Saved agent traces** (`" + paths.TRACES_DIR + "`):\n" +