Update app.py

app.py

@@ -3,15 +3,21 @@ import os, re, json, textwrap, traceback
 from decimal import Decimal
 from typing import List, Tuple
 
-from flask import Flask, request, render_template
+from flask import Flask, request, render_template, url_for
 from flask_cors import CORS
-from rag_mini import search, ensure_ready, DEFAULT_TOPK
+
+from rag_mini import (
+    search,
+    ensure_ready,
+    DEFAULT_TOPK,
+    rag_debug_info,   # for /debug/rag
+)
 
 # ------------ LLM config ------------
-LLM_PROVIDER = (os.getenv("LLM_PROVIDER") or "openai").strip().lower()
-LLM_MODEL    = (os.getenv("LLM_MODEL") or "gpt-4o-mini").strip()
-LLM_API_KEY      = os.getenv("OPENAI_API_KEY") or os.getenv("LLM_API_KEY")
-OPENAI_BASE_URL  = os.getenv("OPENAI_BASE_URL")  # optional
+LLM_PROVIDER    = (os.getenv("LLM_PROVIDER") or "openai").strip().lower()
+LLM_MODEL       = (os.getenv("LLM_MODEL") or "gpt-4o-mini").strip()
+LLM_API_KEY     = os.getenv("OPENAI_API_KEY") or os.getenv("LLM_API_KEY")
+OPENAI_BASE_URL = os.getenv("OPENAI_BASE_URL")  # optional (Azure/proxy)
 
 app = Flask(__name__)
 app.secret_key = os.getenv("FLASK_SECRET_KEY", "change-me-please")
@@ -24,8 +30,8 @@ Return two things:
   "candidates": [
     {
       "name": "string",
-      "score": 0,            // 0..400
-      "score_pct": 0,        // 0..100
+      "score": 0,            // 0..400 (sum of 4 independent 0..100 utilities)
+      "score_pct": 0,        // 0..100 normalized display
       "reasons": ["..."],
       "tradeoffs": ["..."],
       "citations": ["[1]", "[2]"]
@@ -36,7 +42,7 @@ Return two things:
 Rules:
 - Use only the provided context; cite with [1], [2]. No fabrication.
 - Utilities per criterion are in [0,1]. Cost utility increases as cost decreases.
-- Weights are independent 0..100 (not normalized).
+- Weights (performance, stability, cost, availability) are independent 0..100 (NOT normalized).
 """
 
 ANSWER_TEMPLATE = """User constraints:
@@ -148,6 +154,10 @@ def healthz():
         "has_api_key": bool(LLM_API_KEY),
     }, 200
 
+@app.get("/debug/rag")
+def debug_rag():
+    return rag_debug_info(), 200
+
 @app.get("/")
 def index():
     return render_template("index.html", default_k=DEFAULT_TOPK)
@@ -175,15 +185,14 @@ def recommend():
                     f"Consider budget={budget} and process={process}. "
                     f"Rank by performance, stability, cost, and availability.")
 
-        # RAG search can fail if embeddings libs are missing – catch it
+        # RAG search (never crash UI)
         try:
             hits = search(question, k=k)
+            rag_error = ""
         except Exception as e:
             app.logger.exception("RAG search failed")
             hits = []
             rag_error = f"RAG error: {type(e).__name__}: {e}"
-        else:
-            rag_error = ""
 
         ctx, cites = format_context(hits)
 
@@ -195,7 +204,7 @@ def recommend():
             context=ctx, citations=cites
         )
 
-        # LLM call – catch and show the error text instead of 500
+        # LLM call (never crash UI)
         raw = ""
         try:
             raw = call_llm_cloud(SYSTEM_RULES, user_prompt)
@@ -206,7 +215,6 @@ def recommend():
         parsed = extract_json_block(raw) if raw else None
         candidates = (parsed or {}).get("candidates", []) if parsed else []
 
-        # Append any RAG error to the raw output so you see it in UI
         if rag_error:
             raw = f"{rag_error}\n\n{raw}"
 
@@ -219,7 +227,6 @@ def recommend():
             raw_output=raw or "",
         )
     except Exception as e:
-        # Last-resort guard: never return 500 to the user without context
         app.logger.exception("recommend() hard failure")
         tb = traceback.format_exc()
         return render_template(