Upload folder using huggingface_hub

Updates/only-routers_ai_poc_hf_fixed_v10_1.ipynb

@@ -0,0 +1,1614 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "410c42fc",
+   "metadata": {},
+   "source": [
+    "# Only-Routers (v10.1)\n",
+    "\n",
+    "Fixes escaped docstrings causing SyntaxError.\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "cbcb1594",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import re\n",
+    "import json\n",
+    "import math\n",
+    "import hashlib\n",
+    "import tempfile\n",
+    "from dataclasses import dataclass\n",
+    "from datetime import datetime, date\n",
+    "from typing import Any, Dict, List, Optional, Tuple\n",
+    "\n",
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "\n",
+    "import fitz  # PyMuPDF\n",
+    "import faiss\n",
+    "from sentence_transformers import SentenceTransformer\n",
+    "from rapidfuzz import fuzz, process\n",
+    "\n",
+    "import gradio as gr\n",
+    "from openai import OpenAI\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Settings\n",
+    "# ============================\n",
+    "TODAY = date(2026, 1, 18)\n",
+    "OPENAI_MODEL = \"gpt-5.2\"\n",
+    "OPENAI_REASONING = {\"effort\": \"high\"}\n",
+    "MATCH_OK = 80\n",
+    "\n",
+    "EMBED_MODEL_NAME = \"sentence-transformers/all-MiniLM-L6-v2\"\n",
+    "PARSEC_CONTEXT_BEFORE = 900\n",
+    "PARSEC_CONTEXT_AFTER = 1600\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# OpenAI client (HF Space secret: OPENAI_API_KEY)\n",
+    "# ============================\n",
+    "API_KEY = os.getenv(\"OPENAI_API_KEY\", \"\").strip()\n",
+    "client = OpenAI(api_key=API_KEY) if API_KEY else None\n",
+    "\n",
+    "# ----------------------------\n",
+    "# Gradio state helpers\n",
+    "# Keep state as a JSON STRING to avoid schema issues on Hugging Face.\n",
+    "# ----------------------------\n",
+    "def state_load(st_json: str) -> Dict[str, Any]:\n",
+    "    try:\n",
+    "        if not st_json:\n",
+    "            return {}\n",
+    "        return json.loads(st_json) if isinstance(st_json, str) else {}\n",
+    "    except Exception:\n",
+    "        return {}\n",
+    "\n",
+    "def state_dump(st: Dict[str, Any]) -> str:\n",
+    "    try:\n",
+    "        return json.dumps(st or {}, ensure_ascii=False)\n",
+    "    except Exception:\n",
+    "        return \"{}\"\n",
+    "\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Helpers\n",
+    "# ============================\n",
+    "def norm_text(s: Any) -> str:\n",
+    "    try:\n",
+    "        if s is None or (isinstance(s, float) and math.isnan(s)) or pd.isna(s):\n",
+    "            return \"\"\n",
+    "    except Exception:\n",
+    "        pass\n",
+    "    s = str(s).strip().lower()\n",
+    "    s = re.sub(r\"[^a-z0-9\\s\\-\\/]\", \" \", s)\n",
+    "    s = re.sub(r\"\\s+\", \" \", s).strip()\n",
+    "    return s\n",
+    "\n",
+    "def safe_str(v: Any) -> str:\n",
+    "    if v is None or (isinstance(v, float) and pd.isna(v)) or pd.isna(v):\n",
+    "        return \"\"\n",
+    "    return str(v).strip()\n",
+    "\n",
+    "def is_5g(modem_type: Any) -> bool:\n",
+    "    s = norm_text(modem_type)\n",
+    "    return (\"5g\" in s) or (\"nr\" in s)\n",
+    "\n",
+    "def json_load_safe(s: str) -> Dict[str, Any]:\n",
+    "    try:\n",
+    "        return json.loads(s)\n",
+    "    except Exception:\n",
+    "        return {}\n",
+    "\n",
+    "def gpt_json(system: str, payload: Dict[str, Any], max_tokens: int = 600) -> Dict[str, Any]:\n",
+    "    if client is None:\n",
+    "        return {}\n",
+    "    resp = client.responses.create(\n",
+    "        model=OPENAI_MODEL,\n",
+    "        reasoning=OPENAI_REASONING,\n",
+    "        input=[{\"role\":\"system\",\"content\":system},{\"role\":\"user\",\"content\":json.dumps(payload)}],\n",
+    "        max_output_tokens=max_tokens,\n",
+    "    )\n",
+    "    return json_load_safe(getattr(resp, \"output_text\", \"\") or \"\")\n",
+    "\n",
+    "\n",
+    "def gpt_answer_md(system: str, user: str, max_tokens: int = 650) -> str:\n",
+    "    \"\"\"Return a rep-friendly markdown answer.\"\"\"\n",
+    "    if client is None:\n",
+    "        return \"No API key is configured, so I can't answer detailed questions right now.\"\n",
+    "    resp = client.responses.create(\n",
+    "        model=OPENAI_MODEL,\n",
+    "        reasoning=OPENAI_REASONING,\n",
+    "        input=[\n",
+    "            {\"role\": \"system\", \"content\": system},\n",
+    "            {\"role\": \"user\", \"content\": user},\n",
+    "        ],\n",
+    "        max_output_tokens=max_tokens,\n",
+    "    )\n",
+    "    return (getattr(resp, \"output_text\", \"\") or \"\").strip()\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Load data\n",
+    "# ============================\n",
+    "EOS_PATH = \"routers_eos_eol_by_sku.csv\"\n",
+    "DEC_PATH = \"dec2025routers.csv\"\n",
+    "PARSEC_PDF = \"ParsecCatalog.pdf\"\n",
+    "\n",
+    "if not os.path.exists(EOS_PATH):\n",
+    "    raise FileNotFoundError(f\"Missing {EOS_PATH} in repo.\")\n",
+    "if not os.path.exists(DEC_PATH):\n",
+    "    raise FileNotFoundError(f\"Missing {DEC_PATH} in repo.\")\n",
+    "if not os.path.exists(PARSEC_PDF):\n",
+    "    raise FileNotFoundError(f\"Missing {PARSEC_PDF} in repo.\")\n",
+    "\n",
+    "df_eos = pd.read_csv(EOS_PATH).copy()\n",
+    "df_dec = pd.read_csv(DEC_PATH).copy()\n",
+    "\n",
+    "\n",
+    "def _canonize_eos_columns(df: pd.DataFrame) -> pd.DataFrame:\n",
+    "    \"\"\"Normalize lifecycle CSV column names (case-insensitive) and create expected columns.\"\"\"\n",
+    "    # Map various header spellings to canonical names used by the app\n",
+    "    mapping = {}\n",
+    "    for c in df.columns:\n",
+    "        k = str(c).strip().lower().replace(\" \", \"_\")\n",
+    "        if k in {\"sku\", \"model\", \"device\", \"device_sku\"}:\n",
+    "            mapping[c] = \"sku\"\n",
+    "        elif k in {\"manufacturer\", \"make\", \"vendor\"}:\n",
+    "            mapping[c] = \"manufacturer\"\n",
+    "        elif k in {\"device_type\", \"type\"}:\n",
+    "            mapping[c] = \"device_type\"\n",
+    "        elif k in {\"end_of_sale\", \"eos\", \"end_sale\", \"end_of_sales\"}:\n",
+    "            mapping[c] = \"end_of_sale\"\n",
+    "        elif k in {\"end_of_life\", \"eol\", \"end_life\"}:\n",
+    "            mapping[c] = \"end_of_life\"\n",
+    "        elif k in {\"suggested_replacement\", \"replacement_4g\", \"lte_replacement\", \"replacement_lte\", \"replacement\"}:\n",
+    "            mapping[c] = \"suggested_replacement\"\n",
+    "        elif k in {\"advanced_5g_option\", \"replacement_5g\", \"fiveg_replacement\", \"5g_replacement\", \"upgrade_5g\"}:\n",
+    "            mapping[c] = \"advanced_5g_option\"\n",
+    "        elif k in {\"region\", \"market\"}:\n",
+    "            mapping[c] = \"region\"\n",
+    "        elif k in {\"notes\", \"note\"}:\n",
+    "            mapping[c] = \"notes\"\n",
+    "        elif k in {\"description\", \"device_description\", \"name\"}:\n",
+    "            mapping[c] = \"description\"\n",
+    "\n",
+    "    df = df.rename(columns=mapping).copy()\n",
+    "\n",
+    "    # Create expected columns if missing\n",
+    "    if \"sku\" not in df.columns:\n",
+    "        # Try the common capitalized header as a fallback\n",
+    "        if \"SKU\" in df.columns:\n",
+    "            df[\"sku\"] = df[\"SKU\"].astype(str)\n",
+    "        else:\n",
+    "            df[\"sku\"] = \"\"\n",
+    "\n",
+    "    if \"manufacturer\" not in df.columns:\n",
+    "        df[\"manufacturer\"] = \"\"\n",
+    "\n",
+    "    if \"device_type\" not in df.columns:\n",
+    "        df[\"device_type\"] = \"\"\n",
+    "\n",
+    "    if \"description\" not in df.columns:\n",
+    "        # If the simplified file removed description, use SKU as description (still searchable)\n",
+    "        df[\"description\"] = df[\"sku\"].astype(str)\n",
+    "\n",
+    "    if \"notes\" not in df.columns:\n",
+    "        df[\"notes\"] = \"\"\n",
+    "\n",
+    "    if \"region\" not in df.columns:\n",
+    "        df[\"region\"] = \"\"\n",
+    "\n",
+    "    if \"suggested_replacement\" not in df.columns:\n",
+    "        df[\"suggested_replacement\"] = \"\"\n",
+    "\n",
+    "    if \"advanced_5g_option\" not in df.columns:\n",
+    "        df[\"advanced_5g_option\"] = \"\"\n",
+    "\n",
+    "    if \"end_of_sale\" not in df.columns:\n",
+    "        df[\"end_of_sale\"] = \"\"\n",
+    "\n",
+    "    if \"end_of_life\" not in df.columns:\n",
+    "        df[\"end_of_life\"] = \"\"\n",
+    "\n",
+    "    return df\n",
+    "\n",
+    "df_eos = _canonize_eos_columns(df_eos)\n",
+    "\n",
+    "\n",
+    "def region_ok(x: Any) -> bool:\n",
+    "    s = str(x or \"\").strip().lower()\n",
+    "    if not s:\n",
+    "        return True\n",
+    "    if \"not specified\" in s:\n",
+    "        return True\n",
+    "    if \"north america\" in s:\n",
+    "        return True\n",
+    "    if re.search(r\"\\busa\\b\", s):\n",
+    "        return True\n",
+    "    if re.search(r\"\\bunited\\s+states\\b\", s):\n",
+    "        return True\n",
+    "    if re.search(r\"\\bu\\.?s\\.?\\b\", s):\n",
+    "        return True\n",
+    "    return False\n",
+    "\n",
+    "if \"region\" in df_eos.columns:\n",
+    "    df_eos = df_eos[df_eos[\"region\"].apply(region_ok)].reset_index(drop=True)\n",
+    "\n",
+    "# Maker mapping (includes Teltonika)\n",
+    "CANON_MAKER = {\n",
+    "    \"CRADLEPOINT\": {\"cradlepoint\", \"ericsson\", \"ericsson enterprise wireless\"},\n",
+    "    \"SIERRA\": {\"sierra\", \"sierra wireless\", \"semtech\", \"airlink\"},\n",
+    "    \"FEENEY\": {\"feeney\", \"feeney wireless\", \"inseego\"},\n",
+    "    \"DIGI\": {\"digi\", \"accelerated\", \"accelerated concepts\"},\n",
+    "    \"CISCO_MERAKI\": {\"meraki\", \"cisco meraki\"},\n",
+    "    \"CISCO\": {\"cisco\"},\n",
+    "    \"TELTONIKA\": {\"teltonika\"},\n",
+    "}\n",
+    "\n",
+    "def canon_maker_from_text(s: Any) -> str:\n",
+    "    t = norm_text(s)\n",
+    "    for canon, terms in CANON_MAKER.items():\n",
+    "        for term in terms:\n",
+    "            if term in t:\n",
+    "                return canon\n",
+    "    return \"UNKNOWN\"\n",
+    "\n",
+    "df_eos[\"_canon_make\"] = df_eos[\"manufacturer\"].apply(canon_maker_from_text) if \"manufacturer\" in df_eos.columns else \"UNKNOWN\"\n",
+    "df_eos[\"_norm_sku\"] = df_eos[\"sku\"].apply(norm_text) if \"sku\" in df_eos.columns else \"\"\n",
+    "df_eos[\"_norm_desc\"] = df_eos[\"description\"].apply(norm_text) if \"description\" in df_eos.columns else \"\"\n",
+    "df_eos[\"_norm_notes\"] = df_eos[\"notes\"].apply(norm_text) if \"notes\" in df_eos.columns else \"\"\n",
+    "\n",
+    "df_dec[\"_canon_make\"] = df_dec[\"Make\"].apply(canon_maker_from_text) if \"Make\" in df_dec.columns else \"UNKNOWN\"\n",
+    "df_dec[\"_norm_model\"] = df_dec[\"Model\"].apply(norm_text) if \"Model\" in df_dec.columns else \"\"\n",
+    "df_dec[\"_is5g\"] = df_dec[\"Modem Type\"].apply(is_5g) if \"Modem Type\" in df_dec.columns else False\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Date helpers\n",
+    "# ============================\n",
+    "@dataclass\n",
+    "class ParsedDate:\n",
+    "    raw: str\n",
+    "    kind: str\n",
+    "    value: Optional[date]\n",
+    "\n",
+    "def parse_date_field(x: Any) -> ParsedDate:\n",
+    "    raw = str(x or \"\").strip()\n",
+    "    if not raw:\n",
+    "        return ParsedDate(raw=\"\", kind=\"missing\", value=None)\n",
+    "\n",
+    "    # Common US formats: M/D/YY or M/D/YYYY (e.g., 6/24/24, 9/30/21)\n",
+    "    for fmt in (\"%m/%d/%y\", \"%m/%d/%Y\", \"%-m/%-d/%y\", \"%-m/%-d/%Y\"):\n",
+    "        try:\n",
+    "            dt = datetime.strptime(raw, fmt).date()\n",
+    "            return ParsedDate(raw=raw, kind=\"full\", value=dt)\n",
+    "        except Exception:\n",
+    "            pass\n",
+    "\n",
+    "    # ISO-ish: YYYY\n",
+    "    if re.fullmatch(r\"\\d{4}\", raw):\n",
+    "        y = int(raw)\n",
+    "        if y == TODAY.year:\n",
+    "            return ParsedDate(raw=raw, kind=\"year\", value=date(y, 1, 1))\n",
+    "        if y < TODAY.year:\n",
+    "            return ParsedDate(raw=raw, kind=\"year\", value=date(y, 1, 1))\n",
+    "        return ParsedDate(raw=raw, kind=\"year\", value=date(y, 12, 31))\n",
+    "\n",
+    "    # YYYY-MM\n",
+    "    if re.fullmatch(r\"\\d{4}-\\d{2}\", raw):\n",
+    "        try:\n",
+    "            y, m = raw.split(\"-\")\n",
+    "            return ParsedDate(raw=raw, kind=\"year_month\", value=date(int(y), int(m), 1))\n",
+    "        except Exception:\n",
+    "            return ParsedDate(raw=raw, kind=\"bad\", value=None)\n",
+    "\n",
+    "    # YYYY-MM-DD\n",
+    "    if re.fullmatch(r\"\\d{4}-\\d{2}-\\d{2}\", raw):\n",
+    "        try:\n",
+    "            dt = datetime.strptime(raw, \"%Y-%m-%d\").date()\n",
+    "            return ParsedDate(raw=raw, kind=\"full\", value=dt)\n",
+    "        except Exception:\n",
+    "            return ParsedDate(raw=raw, kind=\"bad\", value=None)\n",
+    "\n",
+    "    # Last resort: leave as raw (unparsed)\n",
+    "    return ParsedDate(raw=raw, kind=\"bad\", value=None)\n",
+    "\n",
+    "    if re.fullmatch(r\"\\d{4}-\\d{2}-\\d{2}\", raw):\n",
+    "        try:\n",
+    "            dt = datetime.strptime(raw, \"%Y-%m-%d\").date()\n",
+    "            return ParsedDate(raw=raw, kind=\"full\", value=dt)\n",
+    "        except Exception:\n",
+    "            return ParsedDate(raw=raw, kind=\"bad\", value=None)\n",
+    "\n",
+    "    return ParsedDate(raw=raw, kind=\"bad\", value=None)\n",
+    "\n",
+    "def display_date(pd_: ParsedDate) -> str:\n",
+    "    if pd_.kind == \"missing\":\n",
+    "        return \"Not listed\"\n",
+    "    if pd_.kind == \"bad\":\n",
+    "        return pd_.raw or \"Not listed\"\n",
+    "    return pd_.raw\n",
+    "\n",
+    "def status_from_eos_eol(eos: ParsedDate, eol: ParsedDate) -> str:\n",
+    "    if eos.value is None and eol.value is None:\n",
+    "        return \"Unknown\"\n",
+    "    if eol.value is not None and eol.value <= TODAY:\n",
+    "        return \"End of Life\"\n",
+    "    if eos.value is not None and eos.value <= TODAY:\n",
+    "        return \"End of Sale\"\n",
+    "    return \"Active\"\n",
+    "\n",
+    "def row_to_dates_and_status(row: pd.Series) -> Tuple[str, str, str]:\n",
+    "    eos = parse_date_field(row.get(\"end_of_sale\"))\n",
+    "    eol = parse_date_field(row.get(\"end_of_life\"))\n",
+    "    return display_date(eos), display_date(eol), status_from_eos_eol(eos, eol)\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Embeddings + Parsec index\n",
+    "# ============================\n",
+    "embedder = SentenceTransformer(EMBED_MODEL_NAME)\n",
+    "\n",
+    "def extract_pdf_text_pages(path: str) -> List[str]:\n",
+    "    doc = fitz.open(path)\n",
+    "    return [doc[i].get_text(\"text\") for i in range(len(doc))]\n",
+    "\n",
+    "def build_parsec_cards(pages: List[str]) -> List[str]:\n",
+    "    cards = []\n",
+    "    for p in pages:\n",
+    "        for m in re.finditer(r\"Standard\\s+SKU:\", p):\n",
+    "            start = max(0, m.start() - PARSEC_CONTEXT_BEFORE)\n",
+    "            end = min(len(p), m.start() + PARSEC_CONTEXT_AFTER)\n",
+    "            c = p[start:end].strip()\n",
+    "            if len(c) >= 200:\n",
+    "                cards.append(c)\n",
+    "    out, seen = [], set()\n",
+    "    for c in cards:\n",
+    "        h = hashlib.sha1(c.encode(\"utf-8\")).hexdigest()\n",
+    "        if h not in seen:\n",
+    "            seen.add(h); out.append(c)\n",
+    "    return out\n",
+    "\n",
+    "parsec_cards = build_parsec_cards(extract_pdf_text_pages(PARSEC_PDF))\n",
+    "parsec_emb = embedder.encode(parsec_cards, batch_size=64, show_progress_bar=False, normalize_embeddings=True)\n",
+    "parsec_emb = np.asarray(parsec_emb, dtype=np.float32)\n",
+    "parsec_index = faiss.IndexFlatIP(parsec_emb.shape[1])\n",
+    "parsec_index.add(parsec_emb)\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Device resolution\n",
+    "# ============================\n",
+    "def label_for_row(i: int) -> str:\n",
+    "    r = df_eos.iloc[i]\n",
+    "    return f\"{r.get('sku','')} — {r.get('manufacturer','')} — {r.get('description','')}\"[:220]\n",
+    "\n",
+    "EOS_LABELS = [label_for_row(i) for i in range(len(df_eos))]\n",
+    "EOS_CORPUS = []\n",
+    "for _, r in df_eos.iterrows():\n",
+    "    EOS_CORPUS.append(\" \".join([r.get(\"_norm_sku\",\"\"), r.get(\"_canon_make\",\"\"), r.get(\"_norm_desc\",\"\"), r.get(\"_norm_notes\",\"\")]))\n",
+    "\n",
+    "def local_candidates(query: str, top_k: int = 6) -> List[Tuple[int, int, str]]:\n",
+    "    q = norm_text(query)\n",
+    "    hits = process.extract(q, EOS_CORPUS, scorer=fuzz.WRatio, limit=top_k)\n",
+    "    return [(int(idx), int(score), EOS_LABELS[int(idx)]) for _, score, idx in hits]\n",
+    "\n",
+    "def gpt_choose_device(user_text: str, candidates: List[Tuple[int,int,str]]) -> Dict[str, Any]:\n",
+    "    if client is None:\n",
+    "        return {}\n",
+    "    sys = \"Pick which router the user meant. Never invent. Return strict JSON only.\"\n",
+    "    payload = {\n",
+    "        \"user_input\": user_text,\n",
+    "        \"candidates\": [{\"row_idx\": i, \"score\": s, \"label\": lbl} for (i,s,lbl) in candidates],\n",
+    "        \"rules\": [\n",
+    "            \"If one is clearly correct, return mode='ok' with row_idx.\",\n",
+    "            \"If two are plausible, return mode='pick' with top 2 options.\"\n",
+    "        ],\n",
+    "        \"output_schema\": {\"mode\":\"ok|pick\",\"row_idx\":\"int\",\"options\":[{\"row_idx\":\"int\",\"label\":\"string\"}]}\n",
+    "    }\n",
+    "    return gpt_json(sys, payload, max_tokens=280)\n",
+    "\n",
+    "def resolve_device(user_text: str) -> Dict[str, Any]:\n",
+    "    q = norm_text(user_text)\n",
+    "    exact = df_eos.index[df_eos[\"_norm_sku\"] == q].tolist()\n",
+    "    if len(exact) == 1:\n",
+    "        return {\"mode\":\"ok\",\"row_idx\": int(exact[0])}\n",
+    "    if len(exact) > 1:\n",
+    "        opts = [{\"row_idx\": int(i), \"label\": EOS_LABELS[int(i)]} for i in exact[:2]]\n",
+    "        return {\"mode\":\"pick\",\"options\": opts}\n",
+    "\n",
+    "    cands = local_candidates(user_text, top_k=6)\n",
+    "    if not cands:\n",
+    "        return {\"mode\":\"not_found\"}\n",
+    "\n",
+    "    if cands[0][1] >= 95 and (len(cands) == 1 or (cands[0][1] - cands[1][1]) >= 8):\n",
+    "        return {\"mode\":\"ok\",\"row_idx\": cands[0][0]}\n",
+    "\n",
+    "    g = gpt_choose_device(user_text, cands)\n",
+    "    if g.get(\"mode\") == \"ok\" and isinstance(g.get(\"row_idx\"), int):\n",
+    "        return {\"mode\":\"ok\",\"row_idx\": int(g[\"row_idx\"])}\n",
+    "\n",
+    "    if g.get(\"mode\") == \"pick\":\n",
+    "        opts = g.get(\"options\", []) or []\n",
+    "        opts2 = [{\"row_idx\": int(o[\"row_idx\"]), \"label\": str(o[\"label\"])} for o in opts[:2] if \"row_idx\" in o]\n",
+    "        if opts2:\n",
+    "            return {\"mode\":\"pick\",\"options\": opts2}\n",
+    "\n",
+    "    if len(cands) > 1:\n",
+    "        return {\"mode\":\"pick\",\"options\":[{\"row_idx\":cands[0][0],\"label\":cands[0][2]},{\"row_idx\":cands[1][0],\"label\":cands[1][2]}]}\n",
+    "    return {\"mode\":\"pick\",\"options\":[{\"row_idx\":cands[0][0],\"label\":cands[0][2]}]}\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Replacements — lifecycle CSV source of truth\n",
+    "# ============================\n",
+    "def extract_model_token(text: str) -> str:\n",
+    "    s = safe_str(text)\n",
+    "    if not s:\n",
+    "        return \"\"\n",
+    "    parts = [p.strip() for p in s.split(\"|\") if p.strip()]\n",
+    "    candidates = parts[::-1] if parts else [s]\n",
+    "    for cand in candidates:\n",
+    "        m = re.search(r\"\\bRUT[A-Z]?\\d{2,4}\\b\", cand.upper())\n",
+    "        if m:\n",
+    "            return m.group(0).upper()\n",
+    "        m = re.search(r\"\\bIX\\d{2}\\b\", cand, flags=re.IGNORECASE)\n",
+    "        if m:\n",
+    "            return m.group(0).upper()\n",
+    "        m = re.search(r\"\\b(R\\d{3,4}|E\\d{3,4}|S\\d{3,4})\\b\", cand, flags=re.IGNORECASE)\n",
+    "        if m:\n",
+    "            return m.group(0).upper()\n",
+    "        m = re.search(r\"\\b[A-Z]{1,6}\\d{2,4}[A-Z]?\\b\", cand.upper())\n",
+    "        if m:\n",
+    "            return m.group(0).upper()\n",
+    "    return candidates[0][:60]\n",
+    "\n",
+    "def device_is_4g(row: pd.Series) -> bool:\n",
+    "    # Detect LTE/4G even when the description uses \"Cat 4 / Cat6 / Cat 12\" without saying \"LTE\"\n",
+    "    t = norm_text(row.get(\"description\",\"\")) + \" \" + norm_text(row.get(\"notes\",\"\")) + \" \" + norm_text(row.get(\"sku\",\"\"))\n",
+    "\n",
+    "    # If it explicitly says 5G/NR, treat as not 4G-only\n",
+    "    if (\"5g\" in t) or (\"nr\" in t):\n",
+    "        return False\n",
+    "\n",
+    "    # Classic signals\n",
+    "    if (\"lte\" in t) or (\"4g\" in t):\n",
+    "        return True\n",
+    "\n",
+    "    # LTE category signals (Cat 1..20 are LTE categories; Cat M1/M2 are LTE-M)\n",
+    "    if re.search(r\"\\bcat\\s*[-]?\\s*(m1|m2)\\b\", t):\n",
+    "        return True\n",
+    "\n",
+    "    m = re.search(r\"\\bcat\\s*[-]?\\s*(\\d{1,2})\\b\", t)\n",
+    "    if m:\n",
+    "        try:\n",
+    "            cat = int(m.group(1))\n",
+    "            if 0 < cat <= 20:\n",
+    "                return True\n",
+    "        except Exception:\n",
+    "            pass\n",
+    "\n",
+    "    # If \"cat\" appears at all, it's almost always LTE-family\n",
+    "    if \"cat\" in t:\n",
+    "        return True\n",
+    "\n",
+    "    return False\n",
+    "\n",
+    "    # If it explicitly says 5G/NR, treat as not 4G-only\n",
+    "    if (\"5g\" in t) or (\"nr\" in t):\n",
+    "        return False\n",
+    "\n",
+    "    # Classic signals\n",
+    "    if (\"lte\" in t) or (\"4g\" in t):\n",
+    "        return True\n",
+    "\n",
+    "    # LTE category signals (Cat 1..20 are LTE categories; Cat M1/M2 are LTE-M)\n",
+    "    if re.search(r\"\\bcat\\s*[-]?\\s*(m1|m2)\\b\", t):\n",
+    "        return True\n",
+    "\n",
+    "    m = re.search(r\"\\bcat\\s*[-]?\\s*(\\d{1,2})\\b\", t)\n",
+    "    if m:\n",
+    "        try:\n",
+    "            cat = int(m.group(1))\n",
+    "            if 0 < cat <= 20:\n",
+    "                return True\n",
+    "        except Exception:\n",
+    "            pass\n",
+    "\n",
+    "    # If \"cat\" appears at all, it's almost always LTE-family\n",
+    "    if \"cat\" in t:\n",
+    "        return True\n",
+    "\n",
+    "    return False\n",
+    "\n",
+    "\n",
+    "def candidate_5g_models_from_lifecycle(manufacturer: str) -> List[str]:\n",
+    "    mfr = norm_text(manufacturer)\n",
+    "    pool = df_eos[df_eos[\"manufacturer\"].astype(str).str.lower().eq(mfr)].copy() if \"manufacturer\" in df_eos.columns else df_eos.copy()\n",
+    "    vals = pool[\"advanced_5g_option\"].tolist() if \"advanced_5g_option\" in pool.columns else []\n",
+    "    out, seen = [], set()\n",
+    "    for v in vals:\n",
+    "        tok = extract_model_token(v)\n",
+    "        if tok and tok.lower() != \"nan\" and tok not in seen:\n",
+    "            seen.add(tok); out.append(tok)\n",
+    "    return out\n",
+    "\n",
+    "def candidate_4g_models_from_lifecycle(manufacturer: str) -> List[str]:\n",
+    "    mfr = norm_text(manufacturer)\n",
+    "    pool = df_eos[df_eos[\"manufacturer\"].astype(str).str.lower().eq(mfr)].copy() if \"manufacturer\" in df_eos.columns else df_eos.copy()\n",
+    "    vals = pool[\"suggested_replacement\"].tolist() if \"suggested_replacement\" in pool.columns else []\n",
+    "    out, seen = [], set()\n",
+    "    for v in vals:\n",
+    "        tok = extract_model_token(v)\n",
+    "        if tok and tok.lower() != \"nan\" and tok not in seen:\n",
+    "            seen.add(tok); out.append(tok)\n",
+    "    return out\n",
+    "\n",
+    "def gpt_pick_from_candidates(old_row: pd.Series, candidates: List[str], need: str) -> str:\n",
+    "    if client is None or not candidates:\n",
+    "        return \"\"\n",
+    "    sys = \"Pick the best replacement model. Choose only from candidates. Return strict JSON only.\"\n",
+    "    payload = {\n",
+    "        \"old_device\": {\n",
+    "            \"sku\": str(old_row.get(\"sku\",\"\")),\n",
+    "            \"manufacturer\": str(old_row.get(\"manufacturer\",\"\")),\n",
+    "            \"description\": str(old_row.get(\"description\",\"\")),\n",
+    "            \"need\": need,\n",
+    "        },\n",
+    "        \"candidates\": candidates[:40],\n",
+    "        \"output_schema\": {\"choice\":\"string\"}\n",
+    "    }\n",
+    "    out = gpt_json(sys, payload, max_tokens=240) or {}\n",
+    "    choice = str(out.get(\"choice\",\"\") or \"\").strip()\n",
+    "    return choice if choice in candidates else \"\"\n",
+    "\n",
+    "def fallback_5g_from_dec(canon_make: str) -> str:\n",
+    "    pool5 = df_dec[(df_dec[\"_canon_make\"] == canon_make) & (df_dec[\"_is5g\"] == True)]\n",
+    "    return str(pool5.iloc[0][\"Model\"]).strip() if not pool5.empty else \"\"\n",
+    "\n",
+    "def pick_replacements_lifecycle(row: pd.Series, status: str, use_gpt: bool = True) -> Dict[str, Any]:\n",
+    "    canon = str(row.get(\"_canon_make\",\"UNKNOWN\"))\n",
+    "    manufacturer = str(row.get(\"manufacturer\",\"\") or \"\")\n",
+    "\n",
+    "    sug_raw = safe_str(row.get(\"suggested_replacement\",\"\"))\n",
+    "    adv_raw = safe_str(row.get(\"advanced_5g_option\",\"\"))\n",
+    "\n",
+    "    has_4g_alt = bool(sug_raw.strip())\n",
+    "    has_5g_alt = bool(adv_raw.strip())\n",
+    "\n",
+    "    # Treat as 4G if the description indicates LTE OR lifecycle provides a 4G suggested replacement\n",
+    "    is_4g = device_is_4g(row) or has_4g_alt\n",
+    "\n",
+    "    # Provide 5G option if the unit is 4G, EOS/EOL, or lifecycle explicitly provides advanced_5g_option\n",
+    "    want_5g = is_4g or (status in {\"End of Sale\",\"End of Life\"}) or has_5g_alt\n",
+    "\n",
+    "    # 4G alternative: show whenever lifecycle provides it (or device appears 4G)\n",
+    "    repl_4g = \"Not applicable\"\n",
+    "    if is_4g or has_4g_alt:\n",
+    "        repl_4g = extract_model_token(sug_raw)\n",
+    "        if not repl_4g:\n",
+    "            cand4 = candidate_4g_models_from_lifecycle(manufacturer)\n",
+    "            repl_4g = (gpt_pick_from_candidates(row, cand4, \"4G alternative\") if (use_gpt and client) else \"\") or (cand4[0] if cand4 else \"\")\n",
+    "        if not repl_4g:\n",
+    "            repl_4g = \"Not applicable\"\n",
+    "\n",
+    "    # 5G replacement: prefer lifecycle advanced_5g_option whenever present\n",
+    "    repl_5g = \"Not listed\"\n",
+    "    if want_5g:\n",
+    "        repl_5g = extract_model_token(adv_raw)\n",
+    "        if not repl_5g:\n",
+    "            cand5 = candidate_5g_models_from_lifecycle(manufacturer)\n",
+    "            repl_5g = (gpt_pick_from_candidates(row, cand5, \"5G replacement/upgrade\") if (use_gpt and client) else \"\") or (cand5[0] if cand5 else \"\")\n",
+    "        if not repl_5g:\n",
+    "            repl_5g = fallback_5g_from_dec(canon) or \"Not listed\"\n",
+    "\n",
+    "    if repl_5g.lower() == \"nan\":\n",
+    "        repl_5g = \"Not listed\"\n",
+    "\n",
+    "    return {\"repl_4g\": repl_4g, \"repl_5g\": repl_5g, \"sources\": [\"lifecycle_csv\"] + ([\"gpt\"] if (use_gpt and client) else [])}\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Antennas (Parsec-only)\n",
+    "# ============================\n",
+    "PARSEC_FAMILY_WORDS = {\"chinook\",\"labrador\",\"boxer\",\"bloodhound\",\"husky\",\"beagle\",\"mastiff\",\"collie\",\"shepherd\",\"belgian\",\"australian\",\"terrier\",\"pyrenees\"}\n",
+    "BAD_NAME_MARKERS = {\"customization\",\"standard connectors\",\"connectors\",\"features\",\"benefits\",\"specifications\",\"mechanical\",\"electrical\",\"mounting\",\"accessories\",\"description:\",\"standard sku\"}\n",
+    "\n",
+    "def clean_line(s: str) -> str:\n",
+    "    s = re.sub(r\"\\s+\", \" \", str(s or \"\").strip())\n",
+    "    if re.fullmatch(r\"-[a-z0-9]+\", s.lower()):\n",
+    "        return \"\"\n",
+    "    return s\n",
+    "\n",
+    "def is_bad_name_line(line: str) -> bool:\n",
+    "    low = line.lower()\n",
+    "    if any(m in low for m in BAD_NAME_MARKERS):\n",
+    "        return True\n",
+    "    if re.search(r\"\\b-[a-z0-9]{1,4}\\b\", low) and len(low) <= 25:\n",
+    "        return True\n",
+    "    return False\n",
+    "\n",
+    "def family_from_line(line: str) -> str:\n",
+    "    low = line.lower()\n",
+    "    for fam in PARSEC_FAMILY_WORDS:\n",
+    "        if fam in low:\n",
+    "            return fam.capitalize()\n",
+    "    return \"\"\n",
+    "\n",
+    "def parsec_connectors_from_card(t: str) -> str:\n",
+    "    m = re.search(r\"Standard\\s+Connectors:\\s*(.+)\", t, flags=re.IGNORECASE)\n",
+    "    if m:\n",
+    "        return re.sub(r\"\\s+\", \" \", m.group(1).strip())[:80]\n",
+    "    return \"\"\n",
+    "\n",
+    "def parsec_mounts_from_card(t: str) -> List[str]:\n",
+    "    mounts = []\n",
+    "    for m in re.finditer(r\"Mount:\\s*(.+)\", t, flags=re.IGNORECASE):\n",
+    "        val = re.sub(r\"\\s+\", \" \", m.group(1).strip())\n",
+    "        parts = [p.strip().lower() for p in val.split(\",\") if p.strip()]\n",
+    "        mounts.extend(parts)\n",
+    "    out = []\n",
+    "    seen = set()\n",
+    "    for x in mounts:\n",
+    "        if x not in seen:\n",
+    "            seen.add(x); out.append(x)\n",
+    "    return out\n",
+    "\n",
+    "def parsec_name_from_card(card_text: str) -> str:\n",
+    "    lines = [clean_line(ln) for ln in str(card_text or \"\").splitlines()]\n",
+    "    lines = [ln for ln in lines if ln]\n",
+    "\n",
+    "    for ln in lines:\n",
+    "        if is_bad_name_line(ln):\n",
+    "            continue\n",
+    "        fam = family_from_line(ln)\n",
+    "        if fam:\n",
+    "            return fam\n",
+    "\n",
+    "    sku_i = None\n",
+    "    for i, ln in enumerate(lines):\n",
+    "        if \"standard sku\" in ln.lower():\n",
+    "            sku_i = i\n",
+    "            break\n",
+    "    if sku_i is not None:\n",
+    "        window = lines[max(0, sku_i - 12):sku_i]\n",
+    "        for ln in reversed(window):\n",
+    "            if is_bad_name_line(ln):\n",
+    "                continue\n",
+    "            if 3 <= len(ln) <= 40 and re.search(r\"[A-Za-z]\", ln):\n",
+    "                return ln.split()[0].capitalize()\n",
+    "\n",
+    "    return \"Parsec antenna\"\n",
+    "\n",
+    "def parsec_part_from_card(t: str) -> str:\n",
+    "    m = re.search(r\"Standard\\s+SKU:\\s*([A-Z0-9]+)\", t)\n",
+    "    return m.group(1).strip() if m else \"\"\n",
+    "\n",
+    "def parsec_desc_from_card(t: str) -> str:\n",
+    "    m = re.search(r\"Description:\\s*(.+?)(?:\\n|$)\", t, flags=re.IGNORECASE)\n",
+    "    return re.sub(r\"\\s+\",\" \",m.group(1).strip())[:220] if m else \"\"\n",
+    "\n",
+    "def parsec_retrieve(query: str, top_k: int = 12) -> List[Dict[str, Any]]:\n",
+    "    qv = embedder.encode([query], normalize_embeddings=True)\n",
+    "    qv = np.asarray(qv, dtype=np.float32)\n",
+    "    scores, ids = parsec_index.search(qv, top_k)\n",
+    "    out: List[Dict[str, Any]] = []\n",
+    "    for sc, i in zip(scores[0].tolist(), ids[0].tolist()):\n",
+    "        if 0 <= int(i) < len(parsec_cards):\n",
+    "            card = parsec_cards[int(i)]\n",
+    "            out.append({\n",
+    "                \"score\": float(sc),\n",
+    "                \"name\": parsec_name_from_card(card),\n",
+    "                \"part_number\": parsec_part_from_card(card),\n",
+    "                \"description\": parsec_desc_from_card(card),\n",
+    "                \"connectors\": parsec_connectors_from_card(card),\n",
+    "                \"mounts\": parsec_mounts_from_card(card),\n",
+    "                \"_card\": card.lower(),\n",
+    "            })\n",
+    "    return out\n",
+    "\n",
+    "def choose_best_parsec(cands: List[Dict[str, Any]], mode: str) -> Dict[str, Any]:\n",
+    "    best = None\n",
+    "    best_score = -1e9\n",
+    "\n",
+    "    for c in cands:\n",
+    "        card = c.get(\"_card\",\"\")\n",
+    "        mounts = c.get(\"mounts\", []) or []\n",
+    "        score = float(c.get(\"score\", 0.0))\n",
+    "\n",
+    "        if \"omni\" in card:\n",
+    "            score += 0.6\n",
+    "        if \"directional\" in card:\n",
+    "            score -= 1.5\n",
+    "\n",
+    "        if mode == \"vehicle\":\n",
+    "            if any(\"magnetic\" in m for m in mounts):\n",
+    "                score += 3.0\n",
+    "            if any(\"through\" in m for m in mounts):\n",
+    "                score += 2.0\n",
+    "            if any(\"wall\" in m for m in mounts) or any(\"pole\" in m for m in mounts):\n",
+    "                score -= 1.2\n",
+    "            if \"app: fixed\" in card and \"mobile\" not in card:\n",
+    "                score -= 2.0\n",
+    "\n",
+    "        if mode == \"stationary\":\n",
+    "            if any(\"wall\" in m for m in mounts):\n",
+    "                score += 2.0\n",
+    "            if any(\"pole\" in m for m in mounts):\n",
+    "                score += 1.8\n",
+    "\n",
+    "        if score > best_score:\n",
+    "            best_score = score\n",
+    "            best = c\n",
+    "\n",
+    "    if not best:\n",
+    "        return {\"name\":\"Parsec antenna\",\"part_number\":\"\",\"description\":\"\",\"connectors\":\"\",\"mounts\":[]}\n",
+    "\n",
+    "    best = dict(best)\n",
+    "    best.pop(\"_card\", None)\n",
+    "    return best\n",
+    "\n",
+    "\n",
+    "def infer_mimo_for_5g(repl_5g_model: str) -> str:\n",
+    "    \"\"\"Rule: every 5G router uses a 4x4 antenna.\"\"\"\n",
+    "    return \"4x4\"\n",
+    "\n",
+    "    # If the model name hints 5G, lean 4x4\n",
+    "    if \"5g\" in model.lower() or model.upper().startswith((\"R\", \"E\", \"S\", \"IX\", \"RUTM\")):\n",
+    "        default = \"4x4\"\n",
+    "    else:\n",
+    "        default = \"2x2\"\n",
+    "\n",
+    "    # Use dec2025routers.csv if we can match the model under the same maker family\n",
+    "    try:\n",
+    "        pool = df_dec[df_dec[\"_canon_make\"] == canon_make].copy()\n",
+    "        if pool.empty:\n",
+    "            return default\n",
+    "        hit = process.extractOne(norm_text(model), pool[\"_norm_model\"].tolist(), scorer=fuzz.WRatio)\n",
+    "        if not hit or hit[1] < MATCH_OK:\n",
+    "            return default\n",
+    "        row = pool.iloc[int(hit[2])]\n",
+    "        txt2 = (str(row.get(\"Antennas (internal/external/both)\", \"\")) + \" \" + str(row.get(\"Modem Type\", \"\")) + \" \" + str(row.get(\"Special notes\",\"\"))).lower()\n",
+    "        if \"4x4\" in txt2 or \"4 x 4\" in txt2 or \"4x 4\" in txt2:\n",
+    "            return \"4x4\"\n",
+    "        if \"2x2\" in txt2 or \"2 x 2\" in txt2:\n",
+    "            return \"2x2\"\n",
+    "        # If modem type includes 5G, lean 4x4\n",
+    "        if \"5g\" in txt2 or \"nr\" in txt2:\n",
+    "            return \"4x4\"\n",
+    "        return default\n",
+    "    except Exception:\n",
+    "        return default\n",
+    "\n",
+    "def antenna_options_for(router_model: str, tech: str, mimo: str) -> Dict[str, Any]:\n",
+    "    q_stationary = f\"{router_model} {tech} {mimo} omni stationary pole wall fixed site Parsec\"\n",
+    "    q_vehicle = f\"{router_model} {tech} {mimo} omni vehicle mobile magnetic through-bolt Parsec\"\n",
+    "\n",
+    "    cand_stationary = parsec_retrieve(q_stationary, top_k=12)\n",
+    "    cand_vehicle = parsec_retrieve(q_vehicle, top_k=12)\n",
+    "\n",
+    "    s = choose_best_parsec(cand_stationary, mode=\"stationary\")\n",
+    "    v = choose_best_parsec(cand_vehicle, mode=\"vehicle\")\n",
+    "\n",
+    "    s.update({\"mimo\": mimo, \"why\": \"Stationary omni best match.\"})\n",
+    "    v.update({\"mimo\": mimo, \"why\": \"Vehicle omni best match.\"})\n",
+    "\n",
+    "    return {\"stationary_omni\": s, \"vehicle_omni\": v, \"sources\":[\"parsec_rag\"]}\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Install-ready checklist\n",
+    "# ============================\n",
+    "def install_ready_checklist(current_sku: str, repl: Dict[str,Any], ant: Dict[str,Any]) -> str:\n",
+    "    st = ant.get(\"stationary_omni\", {})\n",
+    "    vh = ant.get(\"vehicle_omni\", {})\n",
+    "    if client is not None:\n",
+    "        sys = \"Create a short, install-ready checklist for a Verizon rep. Return markdown only.\"\n",
+    "        payload = {\"current_device\": current_sku, \"replacements\": repl, \"antennas\": {\"stationary\": st, \"vehicle\": vh}}\n",
+    "        resp = client.responses.create(\n",
+    "            model=OPENAI_MODEL,\n",
+    "            reasoning=OPENAI_REASONING,\n",
+    "            input=[{\"role\":\"system\",\"content\":sys},{\"role\":\"user\",\"content\":json.dumps(payload)}],\n",
+    "            max_output_tokens=520,\n",
+    "        )\n",
+    "        return (getattr(resp, \"output_text\", \"\") or \"\").strip()\n",
+    "    return \"\\n\".join([\n",
+    "        \"### Install-ready checklist\",\n",
+    "        f\"- Current device: {current_sku}\",\n",
+    "        f\"- 5G replacement: {repl.get('repl_5g','')}\",\n",
+    "        f\"- 4G alternative: {repl.get('repl_4g','Not applicable')}\",\n",
+    "        f\"- Stationary omni antenna: {st.get('name','')} (PN {st.get('part_number','')})\",\n",
+    "        f\"- Vehicle omni antenna: {vh.get('name','')} (PN {vh.get('part_number','')})\",\n",
+    "        \"- Next steps: confirm mounting + cable lengths + power; place order; schedule install.\",\n",
+    "    ])\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Batch mode (NO GPT)\n",
+    "# ============================\n",
+    "def parse_batch_inputs(text_blob: str, file_obj: Any) -> List[str]:\n",
+    "    items: List[str] = []\n",
+    "    if file_obj is not None:\n",
+    "        try:\n",
+    "            path = file_obj.name if hasattr(file_obj, \"name\") else str(file_obj)\n",
+    "            df = pd.read_csv(path)\n",
+    "            col = df.columns[0]\n",
+    "            items.extend([str(x).strip() for x in df[col].tolist() if str(x).strip()])\n",
+    "        except Exception:\n",
+    "            pass\n",
+    "    if text_blob:\n",
+    "        for ln in str(text_blob).splitlines():\n",
+    "            ln = ln.strip()\n",
+    "            if ln:\n",
+    "                items.append(ln)\n",
+    "    seen=set()\n",
+    "    out=[]\n",
+    "    for x in items:\n",
+    "        k=norm_text(x)\n",
+    "        if k and k not in seen:\n",
+    "            seen.add(k); out.append(x)\n",
+    "    return out\n",
+    "\n",
+    "def run_batch(text_blob: str, file_obj: Any, include_antennas: bool):\n",
+    "    inputs = parse_batch_inputs(text_blob, file_obj)\n",
+    "    if not inputs:\n",
+    "        return \"\", None, None, \"\"\n",
+    "\n",
+    "    rows=[]\n",
+    "    for item in inputs:\n",
+    "        res = resolve_device(item)\n",
+    "        if res.get(\"mode\") != \"ok\":\n",
+    "            rows.append({\"Input\": item, \"Matched\":\"\", \"Status\":\"Needs review\", \"EOS\":\"\", \"EOL\":\"\", \"4G alternative\":\"\", \"5G replacement\":\"\", \"Notes\":\"Not found/ambiguous\"})\n",
+    "            continue\n",
+    "\n",
+    "        life_row = df_eos.iloc[int(res[\"row_idx\"])]\n",
+    "        eos, eol, status = row_to_dates_and_status(life_row)\n",
+    "        repl = pick_replacements_lifecycle(life_row, status, use_gpt=False)\n",
+    "\n",
+    "        rows.append({\n",
+    "            \"Input\": item,\n",
+    "            \"Matched\": str(life_row.get(\"sku\",\"\")),\n",
+    "            \"Status\": status,\n",
+    "            \"EOS\": eos,\n",
+    "            \"EOL\": eol,\n",
+    "            \"4G alternative\": repl.get(\"repl_4g\",\"\"),\n",
+    "            \"5G replacement\": repl.get(\"repl_5g\",\"\"),\n",
+    "            \"Notes\": \"\",\n",
+    "        })\n",
+    "\n",
+    "    out_df = pd.DataFrame(rows)\n",
+    "    counts = out_df[\"Status\"].value_counts(dropna=False).to_dict()\n",
+    "    top_5g = out_df[\"5G replacement\"].value_counts(dropna=False).head(5).to_dict()\n",
+    "    summary = f\"Rows: {len(out_df)} | \" + \" | \".join([f\"{k}: {v}\" for k,v in counts.items()])\n",
+    "    rollup = \"Top 5G recommendations:\\n\" + \"\\n\".join([f\"- {k}: {v}\" for k,v in top_5g.items() if str(k).strip()])\n",
+    "\n",
+    "    tmp = tempfile.NamedTemporaryFile(delete=False, suffix=\".csv\")\n",
+    "    out_df.to_csv(tmp.name, index=False)\n",
+    "\n",
+    "    return summary, out_df, tmp.name, rollup\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Replacement feature table + manufacturer link (5G device)\n",
+    "# ============================\n",
+    "\n",
+    "FEATURE_COLS = [\"Device\", \"Modem technology\", \"WiFi\", \"Ports\", \"Antennas\", \"Ruggedness\", \"Use case\"]\n",
+    "\n",
+    "# Manufacturer domains used for best-effort link resolution (no non-maker domains).\n",
+    "MAKER_DOMAINS = {\n",
+    "    \"CRADLEPOINT\": [\"cradlepoint.com\", \"ericsson.com\"],\n",
+    "    \"SIERRA\": [\"semtech.com\", \"airlink.com\"],\n",
+    "    \"FEENEY\": [\"inseego.com\"],\n",
+    "    \"DIGI\": [\"digi.com\"],\n",
+    "    \"CISCO_MERAKI\": [\"meraki.cisco.com\", \"cisco.com\"],\n",
+    "    \"CISCO\": [\"cisco.com\"],\n",
+    "    \"TELTONIKA\": [\"teltonika-networks.com\"],\n",
+    "    \"UNKNOWN\": [],\n",
+    "}\n",
+    "\n",
+    "HTTP_HEADERS = {\n",
+    "    \"User-Agent\": \"Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 \"\n",
+    "                  \"(KHTML, like Gecko) Chrome/120.0 Safari/537.36\"\n",
+    "}\n",
+    "HTTP_TIMEOUT = 12\n",
+    "\n",
+    "def _best_effort_manufacturer_url(model: str, canon_make: str) -> str:\n",
+    "    \"\"\"Try to find a manufacturer page or datasheet link using simple on-domain searches.\n",
+    "    If we can't confirm a page, return the manufacturer homepage for the maker family.\n",
+    "    \"\"\"\n",
+    "    model = str(model or \"\").strip()\n",
+    "    if not model or model in {\"Not listed\", \"Not applicable\"}:\n",
+    "        return \"\"\n",
+    "\n",
+    "    domains = MAKER_DOMAINS.get(canon_make, []) or []\n",
+    "    if not domains:\n",
+    "        return \"\"\n",
+    "\n",
+    "    # Candidate on-domain search URLs (common patterns across sites).\n",
+    "    # We keep these on the manufacturer domain (no Google/Bing).\n",
+    "    q = re.sub(r\"\\s+\", \"+\", model)\n",
+    "    url_candidates = []\n",
+    "    for d in domains:\n",
+    "        url_candidates += [\n",
+    "            f\"https://{d}/search?q={q}\",\n",
+    "            f\"https://{d}/search?query={q}\",\n",
+    "            f\"https://{d}/?s={q}\",\n",
+    "            f\"https://www.{d}/search?q={q}\",\n",
+    "            f\"https://www.{d}/search?query={q}\",\n",
+    "            f\"https://www.{d}/?s={q}\",\n",
+    "        ]\n",
+    "\n",
+    "    # Also try a few direct product patterns for known makers (best effort).\n",
+    "    if canon_make == \"TELTONIKA\":\n",
+    "        slug = model.lower()\n",
+    "        url_candidates += [\n",
+    "            f\"https://teltonika-networks.com/products/routers/{slug}\",\n",
+    "            f\"https://teltonika-networks.com/product/{slug}\",\n",
+    "            \"https://teltonika-networks.com/products/routers/\",\n",
+    "        ]\n",
+    "    if canon_make == \"DIGI\":\n",
+    "        url_candidates += [\n",
+    "            \"https://www.digi.com/products/networking/cellular-routers\",\n",
+    "            f\"https://www.digi.com/search?q={q}\",\n",
+    "        ]\n",
+    "    if canon_make == \"CRADLEPOINT\":\n",
+    "        url_candidates += [\n",
+    "            \"https://cradlepoint.com/products/\",\n",
+    "            f\"https://cradlepoint.com/?s={q}\",\n",
+    "        ]\n",
+    "    if canon_make in {\"CISCO\", \"CISCO_MERAKI\"}:\n",
+    "        url_candidates += [\n",
+    "            f\"https://www.cisco.com/c/en/us/search.html?q={q}\",\n",
+    "        ]\n",
+    "\n",
+    "    # Try to confirm a working page (HTTP 200 and model string somewhere in HTML).\n",
+    "    for u in url_candidates[:18]:\n",
+    "        try:\n",
+    "            import requests\n",
+    "            r = requests.get(u, headers=HTTP_HEADERS, timeout=HTTP_TIMEOUT, allow_redirects=True)\n",
+    "            if r.status_code != 200:\n",
+    "                continue\n",
+    "            html = (r.text or \"\").lower()\n",
+    "            if model.lower() in html or \"datasheet\" in html or \"data sheet\" in html:\n",
+    "                return r.url\n",
+    "        except Exception:\n",
+    "            continue\n",
+    "\n",
+    "    # Fallback: maker homepage\n",
+    "    d0 = domains[0]\n",
+    "    return f\"https://{d0}\"\n",
+    "\n",
+    "def _features_from_dec(model: str, canon_make: str) -> Dict[str, str]:\n",
+    "    \"\"\"Lookup a router model in dec2025routers.csv and return the key feature fields.\"\"\"\n",
+    "    if not model or model in {\"Not listed\", \"Not applicable\"}:\n",
+    "        return {k: \"Not listed\" for k in FEATURE_COLS[1:]}\n",
+    "\n",
+    "    pool = df_dec[df_dec[\"_canon_make\"] == canon_make].copy()\n",
+    "    if pool.empty:\n",
+    "        return {k: \"Not listed\" for k in FEATURE_COLS[1:]}\n",
+    "\n",
+    "    hit = process.extractOne(norm_text(model), pool[\"_norm_model\"].tolist(), scorer=fuzz.WRatio)\n",
+    "    if not hit or hit[1] < MATCH_OK:\n",
+    "        return {k: \"Not listed\" for k in FEATURE_COLS[1:]}\n",
+    "\n",
+    "    r = pool.iloc[int(hit[2])]\n",
+    "    ports = f\"WAN: {r.get('WAN ports and speed','')} | LAN: {r.get('LAN ports and speed','')}\"\n",
+    "    return {\n",
+    "        \"Modem technology\": str(r.get(\"Modem Type\",\"\")) or \"Not listed\",\n",
+    "        \"WiFi\": str(r.get(\"WiFi type\",\"\")) or \"Not listed\",\n",
+    "        \"Ports\": ports.strip() if ports.strip() else \"Not listed\",\n",
+    "        \"Antennas\": str(r.get(\"Antennas (internal/external/both)\",\"\")) or \"Not listed\",\n",
+    "        \"Ruggedness\": str(r.get(\"Ruggedization\",\"\")) or \"Not listed\",\n",
+    "        \"Use case\": str(r.get(\"Primary use case\",\"\")) or \"Not listed\",\n",
+    "    }\n",
+    "\n",
+    "def _gpt_fill_feature_row(device_label: str, model: str, canon_make: str, row: Dict[str, str]) -> Dict[str, str]:\n",
+    "    \"\"\"If dec can't supply values, ask GPT to fill missing ones (best guess).\"\"\"\n",
+    "    if client is None:\n",
+    "        return row\n",
+    "\n",
+    "    missing = [k for k,v in row.items() if (not v) or str(v).strip().lower() in {\"not listed\",\"nan\",\"\"}]\n",
+    "    if not missing:\n",
+    "        return row\n",
+    "\n",
+    "    sys = \"Fill missing router feature fields for a Verizon rep. Return strict JSON only.\"\n",
+    "    payload = {\n",
+    "        \"device_label\": device_label,\n",
+    "        \"model\": model,\n",
+    "        \"maker_family\": canon_make,\n",
+    "        \"known\": row,\n",
+    "        \"fill_only\": missing,\n",
+    "        \"rules\": [\n",
+    "            \"Fill only the requested fields.\",\n",
+    "            \"Best guess if needed. Short phrases only.\",\n",
+    "            \"Return JSON only.\"\n",
+    "        ],\n",
+    "        \"output_schema\": {k: \"string\" for k in missing}\n",
+    "    }\n",
+    "    out = gpt_json(sys, payload, max_tokens=260) or {}\n",
+    "    for k in missing:\n",
+    "        val = str(out.get(k, \"\") or \"\").strip()\n",
+    "        if val:\n",
+    "            row[k] = val\n",
+    "    return row\n",
+    "\n",
+    "def build_replacement_features_table(repl_4g: str, repl_5g: str, canon_make: str) -> pd.DataFrame:\n",
+    "    rows = []\n",
+    "\n",
+    "    # 4G\n",
+    "    row4 = _features_from_dec(repl_4g, canon_make)\n",
+    "    row4 = _gpt_fill_feature_row(\"4G alternative\", repl_4g, canon_make, row4)\n",
+    "    rows.append({\"Device\": \"4G alternative\", **row4})\n",
+    "\n",
+    "    # 5G\n",
+    "    row5 = _features_from_dec(repl_5g, canon_make)\n",
+    "    row5 = _gpt_fill_feature_row(\"5G replacement\", repl_5g, canon_make, row5)\n",
+    "    rows.append({\"Device\": \"5G replacement\", **row5})\n",
+    "\n",
+    "    df = pd.DataFrame(rows, columns=FEATURE_COLS)\n",
+    "    return df\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Verizon fit badges (small table) for recommended devices\n",
+    "# ============================\n",
+    "\n",
+    "FIT_COLS = [\"Device\", \"Fit badges\", \"Ethernet ports\", \"Battery\"]\n",
+    "\n",
+    "def _parse_ethernet_ports(wan_field: str, lan_field: str) -> str:\n",
+    "    \"\"\"Best-effort total ethernet ports based on WAN/LAN text.\"\"\"\n",
+    "    def _count(field: str) -> int:\n",
+    "        s = str(field or \"\")\n",
+    "        # Common forms: \"1x GbE\", \"2 x 10/100\", \"WAN: 1\", etc.\n",
+    "        nums = [int(x) for x in re.findall(r\"(\\\\d+)\\\\s*x\", s.lower())]\n",
+    "        if nums:\n",
+    "            return sum(nums)\n",
+    "        # Fallback: if it contains 'port' with a number\n",
+    "        m = re.search(r\"(\\\\d+)\\\\s*port\", s.lower())\n",
+    "        if m:\n",
+    "            return int(m.group(1))\n",
+    "        # If it contains '1' and 'wan' in short text, guess 1\n",
+    "        if \"wan\" in s.lower() and re.search(r\"\\\\b1\\\\b\", s):\n",
+    "            return 1\n",
+    "        return 0\n",
+    "\n",
+    "    total = _count(wan_field) + _count(lan_field)\n",
+    "    return str(total) if total > 0 else \"Not listed\"\n",
+    "\n",
+    "def _battery_badge(battery_field: str) -> str:\n",
+    "    s = str(battery_field or \"\").strip().lower()\n",
+    "    if not s or s in {\"none\", \"no\", \"n/a\", \"not listed\"}:\n",
+    "        return \"No\"\n",
+    "    return \"Yes\"\n",
+    "\n",
+    "def _bool_badge(flag: bool) -> str:\n",
+    "    return \"Yes\" if flag else \"No\"\n",
+    "\n",
+    "def _dual_sim_from_row_text(*fields: str) -> bool:\n",
+    "    txt = \" \".join([str(x or \"\") for x in fields]).lower()\n",
+    "    return (\"dual sim\" in txt) or (\"2 sim\" in txt) or (\"two sim\" in txt) or (\"dual-sim\" in txt)\n",
+    "\n",
+    "def _throughput_high(throughput_field: str) -> bool:\n",
+    "    t = str(throughput_field or \"\").lower()\n",
+    "    # Heuristic: anything mentioning gbps or >=1000 mbps\n",
+    "    if \"gbps\" in t:\n",
+    "        return True\n",
+    "    m = re.search(r\"(\\\\d+(?:\\\\.\\\\d+)?)\\\\s*mbps\", t)\n",
+    "    if m:\n",
+    "        try:\n",
+    "            return float(m.group(1)) >= 1000.0\n",
+    "        except Exception:\n",
+    "            pass\n",
+    "    return False\n",
+    "\n",
+    "def _gpt_fit_badges(model: str, canon_make: str, is_5g: bool, dec_row: Optional[pd.Series]) -> Tuple[str, str, str]:\n",
+    "    \"\"\"\n",
+    "    GPT-based fill for Fit badges / Ethernet ports / Battery, used when dec is missing or incomplete.\n",
+    "    Returns (badges_csv, ethernet_ports, battery_yesno).\n",
+    "    \"\"\"\n",
+    "    if client is None:\n",
+    "        return (\"Not listed\", \"Not listed\", \"Not listed\")\n",
+    "\n",
+    "    dec_ctx = {}\n",
+    "    if dec_row is not None:\n",
+    "        try:\n",
+    "            dec_ctx = {\n",
+    "                \"Model\": str(dec_row.get(\"Model\",\"\")),\n",
+    "                \"Modem Type\": str(dec_row.get(\"Modem Type\",\"\")),\n",
+    "                \"Ruggedization\": str(dec_row.get(\"Ruggedization\",\"\")),\n",
+    "                \"WAN ports and speed\": str(dec_row.get(\"WAN ports and speed\",\"\")),\n",
+    "                \"LAN ports and speed\": str(dec_row.get(\"LAN ports and speed\",\"\")),\n",
+    "                \"Antennas\": str(dec_row.get(\"Antennas (internal/external/both)\",\"\")),\n",
+    "                \"WiFi type\": str(dec_row.get(\"WiFi type\",\"\")),\n",
+    "                \"Primary use case\": str(dec_row.get(\"Primary use case\",\"\")),\n",
+    "                \"Serial port\": str(dec_row.get(\"Serial port (yes/no)\",\"\")),\n",
+    "                \"VPN\": str(dec_row.get(\"VPN capabilities\",\"\")),\n",
+    "                \"Throughput\": str(dec_row.get(\"Router throughput\",\"\")),\n",
+    "                \"Battery\": str(dec_row.get(\"Battery (internal/removable/none/optional)\",\"\")),\n",
+    "                \"Special notes\": str(dec_row.get(\"Special notes\",\"\")),\n",
+    "                \"Summary\": str(dec_row.get(\"summary and use case\",\"\")),\n",
+    "            }\n",
+    "        except Exception:\n",
+    "            dec_ctx = {}\n",
+    "\n",
+    "    sys = (\n",
+    "        \"You are helping a Verizon rep. Based on the provided router context, output fit badges and a couple quick traits.\\n\"\n",
+    "        \"Return STRICT JSON only.\\n\"\n",
+    "        \"Badges must be chosen from this set only:\\n\"\n",
+    "        \"['Vehicle','Fixed site','Wi‑Fi','Rugged','Dual‑SIM','4x4 MIMO','High throughput','Serial'].\\n\"\n",
+    "        \"Rules:\\n\"\n",
+    "        \"- If is_5g is true, ALWAYS include '4x4 MIMO'.\\n\"\n",
+    "        \"- Ethernet ports: return a single integer as a string if you can infer total ethernet ports, otherwise 'Not listed'.\\n\"\n",
+    "        \"- Battery: return 'Yes' or 'No' if you can infer, otherwise 'Not listed'.\\n\"\n",
+    "        \"- If uncertain between Vehicle vs Fixed site, pick the most likely based on use case/ruggedization.\\n\"\n",
+    "    )\n",
+    "\n",
+    "    payload = {\n",
+    "        \"model\": model,\n",
+    "        \"maker_family\": canon_make,\n",
+    "        \"is_5g\": bool(is_5g),\n",
+    "        \"dec_context\": dec_ctx,\n",
+    "        \"output_schema\": {\n",
+    "            \"badges\": [\"string\"],\n",
+    "            \"ethernet_ports\": \"string\",\n",
+    "            \"battery\": \"Yes|No|Not listed\"\n",
+    "        }\n",
+    "    }\n",
+    "\n",
+    "    out = gpt_json(sys, payload, max_tokens=260) or {}\n",
+    "\n",
+    "    badges = out.get(\"badges\", []) or []\n",
+    "    allowed = {\"Vehicle\",\"Fixed site\",\"Wi‑Fi\",\"Rugged\",\"Dual‑SIM\",\"4x4 MIMO\",\"High throughput\",\"Serial\"}\n",
+    "    clean = []\n",
+    "    for b in badges:\n",
+    "        bs = str(b).strip()\n",
+    "        if bs in allowed:\n",
+    "            clean.append(bs)\n",
+    "\n",
+    "    if is_5g and \"4x4 MIMO\" not in clean:\n",
+    "        clean.append(\"4x4 MIMO\")\n",
+    "\n",
+    "    eth = str(out.get(\"ethernet_ports\",\"\") or \"\").strip()\n",
+    "    if not eth or eth.lower() in {\"nan\",\"none\"}:\n",
+    "        eth = \"Not listed\"\n",
+    "    m = re.search(r\"\\d+\", eth)\n",
+    "    eth = m.group(0) if m else (\"Not listed\" if eth == \"Not listed\" else eth)\n",
+    "\n",
+    "    bat = str(out.get(\"battery\",\"\") or \"\").strip()\n",
+    "    if not bat:\n",
+    "        bat = \"Not listed\"\n",
+    "    if bat.lower().startswith(\"y\"):\n",
+    "        bat = \"Yes\"\n",
+    "    elif bat.lower().startswith(\"n\"):\n",
+    "        bat = \"No\"\n",
+    "    elif bat not in {\"Yes\",\"No\",\"Not listed\"}:\n",
+    "        bat = \"Not listed\"\n",
+    "\n",
+    "    dedup=[]\n",
+    "    seen=set()\n",
+    "    for b in clean:\n",
+    "        if b not in seen:\n",
+    "            seen.add(b); dedup.append(b)\n",
+    "    badges_csv = \", \".join(dedup) if dedup else \"Not listed\"\n",
+    "    return (badges_csv, eth, bat)\n",
+    "\n",
+    "\n",
+    "def _fit_badges_for_model(model: str, canon_make: str, is_5g: bool) -> Tuple[str, str, str]:\n",
+    "    \"\"\"Return (badges_csv, ethernet_ports, battery_yesno). Uses dec2025routers.csv first, then GPT fill.\"\"\"\n",
+    "    model = str(model or \"\").strip()\n",
+    "    if not model or model in {\"Not listed\", \"Not applicable\"}:\n",
+    "        return (\"Not listed\", \"Not listed\", \"Not listed\")\n",
+    "\n",
+    "    pool = df_dec[df_dec[\"_canon_make\"] == canon_make].copy()\n",
+    "    row = None\n",
+    "    if not pool.empty:\n",
+    "        hit = process.extractOne(norm_text(model), pool[\"_norm_model\"].tolist(), scorer=fuzz.WRatio)\n",
+    "        if hit and hit[1] >= MATCH_OK:\n",
+    "            row = pool.iloc[int(hit[2])]\n",
+    "\n",
+    "    badges = []\n",
+    "    eth = \"Not listed\"\n",
+    "    bat_yes = \"Not listed\"\n",
+    "\n",
+    "    if row is not None:\n",
+    "        use_case = str(row.get(\"Primary use case\",\"\") or \"\").lower()\n",
+    "        rugged = str(row.get(\"Ruggedization\",\"\") or \"\").lower()\n",
+    "\n",
+    "        if any(k in use_case for k in [\"vehicle\",\"mobile\",\"fleet\",\"in-vehicle\"]) or \"vehicle\" in rugged:\n",
+    "            badges.append(\"Vehicle\")\n",
+    "        else:\n",
+    "            badges.append(\"Fixed site\")\n",
+    "\n",
+    "        wifi = str(row.get(\"WiFi type\",\"\") or \"\").strip()\n",
+    "        if wifi and wifi.lower() not in {\"none\",\"no\",\"n/a\"}:\n",
+    "            badges.append(\"Wi‑Fi\")\n",
+    "\n",
+    "        if any(k in rugged for k in [\"rugged\",\"industrial\",\"ip\",\"harsh\"]):\n",
+    "            badges.append(\"Rugged\")\n",
+    "\n",
+    "        notes_blob = \" \".join([\n",
+    "            str(row.get(\"Special notes\",\"\") or \"\"),\n",
+    "            str(row.get(\"summary and use case\",\"\") or \"\"),\n",
+    "        ]).lower()\n",
+    "        if \"dual\" in notes_blob and \"sim\" in notes_blob:\n",
+    "            badges.append(\"Dual‑SIM\")\n",
+    "\n",
+    "        if is_5g:\n",
+    "            badges.append(\"4x4 MIMO\")\n",
+    "\n",
+    "        thr = str(row.get(\"Router throughput\",\"\") or \"\").lower()\n",
+    "        m = re.search(r\"(\\d+(\\.\\d+)?)\\s*gb\", thr)\n",
+    "        if m:\n",
+    "            try:\n",
+    "                if float(m.group(1)) >= 1.0:\n",
+    "                    badges.append(\"High throughput\")\n",
+    "            except Exception:\n",
+    "                pass\n",
+    "\n",
+    "        serial = str(row.get(\"Serial port (yes/no)\",\"\") or \"\").strip().lower()\n",
+    "        if serial in {\"yes\",\"y\",\"true\"}:\n",
+    "            badges.append(\"Serial\")\n",
+    "\n",
+    "        wan = str(row.get(\"WAN ports and speed\",\"\") or \"\")\n",
+    "        lan = str(row.get(\"LAN ports and speed\",\"\") or \"\")\n",
+    "        m1 = re.search(r\"(\\d+)\\s*x\", wan.lower())\n",
+    "        m2 = re.search(r\"(\\d+)\\s*x\", lan.lower())\n",
+    "        if m1 or m2:\n",
+    "            total = (int(m1.group(1)) if m1 else 0) + (int(m2.group(1)) if m2 else 0)\n",
+    "            eth = str(total) if total > 0 else \"Not listed\"\n",
+    "\n",
+    "        bat = str(row.get(\"Battery (internal/removable/none/optional)\",\"\") or \"\")\n",
+    "        bat_l = bat.lower().strip()\n",
+    "        if bat_l:\n",
+    "            if \"none\" in bat_l:\n",
+    "                bat_yes = \"No\"\n",
+    "            else:\n",
+    "                bat_yes = \"Yes\"\n",
+    "\n",
+    "    # Use GPT when anything is missing (instead of best-effort inference)\n",
+    "    if (row is None) or (eth == \"Not listed\") or (bat_yes == \"Not listed\") or (not badges):\n",
+    "        g_badges, g_eth, g_bat = _gpt_fit_badges(model, canon_make, is_5g, row)\n",
+    "\n",
+    "        if badges:\n",
+    "            if is_5g and \"4x4 MIMO\" not in badges:\n",
+    "                badges.append(\"4x4 MIMO\")\n",
+    "            dedup=[]\n",
+    "            seen=set()\n",
+    "            for b in badges:\n",
+    "                if b not in seen:\n",
+    "                    seen.add(b); dedup.append(b)\n",
+    "            badges_csv = \", \".join(dedup)\n",
+    "        else:\n",
+    "            badges_csv = g_badges\n",
+    "\n",
+    "        eth = eth if eth != \"Not listed\" else g_eth\n",
+    "        bat_yes = bat_yes if bat_yes != \"Not listed\" else g_bat\n",
+    "        return (badges_csv or \"Not listed\", eth or \"Not listed\", bat_yes or \"Not listed\")\n",
+    "\n",
+    "    dedup=[]\n",
+    "    seen=set()\n",
+    "    for b in badges:\n",
+    "        if b not in seen:\n",
+    "            seen.add(b); dedup.append(b)\n",
+    "    badges_csv = \", \".join(dedup) if dedup else \"Not listed\"\n",
+    "    return (badges_csv, eth, bat_yes)\n",
+    "\n",
+    "def build_fit_table(repl_4g: str, repl_5g: str, canon_make: str) -> pd.DataFrame:\n",
+    "    rows = []\n",
+    "    # 4G alt row (is_5g False)\n",
+    "    b4, eth4, bat4 = _fit_badges_for_model(repl_4g, canon_make, is_5g=False)\n",
+    "    rows.append({\"Device\": \"4G alternative\", \"Fit badges\": b4, \"Ethernet ports\": eth4, \"Battery\": bat4})\n",
+    "    # 5G row (is_5g True)\n",
+    "    b5, eth5, bat5 = _fit_badges_for_model(repl_5g, canon_make, is_5g=True)\n",
+    "    rows.append({\"Device\": \"5G replacement\", \"Fit badges\": b5, \"Ethernet ports\": eth5, \"Battery\": bat5})\n",
+    "    return pd.DataFrame(rows, columns=FIT_COLS)\n",
+    "\n",
+    "# ============================\n",
+    "# Output\n",
+    "# ============================\n",
+    "def assemble_output(life_row: pd.Series, status: str, eos: str, eol: str, repl: Dict[str,Any], ant: Dict[str,Any]) -> str:\n",
+    "    current_name = f\"{life_row.get('sku','')} — {life_row.get('description','')}\".strip(\" —\")\n",
+    "    st = ant.get(\"stationary_omni\", {})\n",
+    "    vh = ant.get(\"vehicle_omni\", {})\n",
+    "\n",
+    "    lines = []\n",
+    "    lines.append(f\"1. Current device: **{current_name}**\")\n",
+    "    lines.append(f\"2. Status: **{status}**\")\n",
+    "    lines.append(f\"3. End of Sale date: **{eos}**\")\n",
+    "    lines.append(f\"4. End of Life date: **{eol}**\")\n",
+    "    lines.append(f\"5. 4G alternative (lifecycle): **{repl.get('repl_4g','Not applicable')}**\")\n",
+    "    lines.append(f\"6. 5G replacement (lifecycle): **{repl.get('repl_5g','Not listed')}**\")\n",
+    "    lines.append(\"7. Antenna options (Parsec-only):\")\n",
+    "    conn_s = f\" | Conn: {st.get('connectors','')}\" if st.get(\"connectors\") else \"\"\n",
+    "    conn_v = f\" | Conn: {vh.get('connectors','')}\" if vh.get(\"connectors\") else \"\"\n",
+    "    lines.append(f\"   - Stationary (Omni): **{st.get('name','')}** (Part #: {st.get('part_number','')}) — {st.get('description','')} — MIMO: {st.get('mimo','')}{conn_s}\")\n",
+    "    lines.append(f\"   - Vehicle (Omni): **{vh.get('name','')}** (Part #: {vh.get('part_number','')}) — {vh.get('description','')} — MIMO: {vh.get('mimo','')}{conn_v}\")\n",
+    "\n",
+    "    lines.append(\"\\nSources (debug):\")\n",
+    "    for s in repl.get(\"sources\", []) if isinstance(repl.get(\"sources\"), list) else []:\n",
+    "        lines.append(f\"- {s}\")\n",
+    "    lines.append(\"- ParsecCatalog.pdf (local RAG)\")\n",
+    "    lines.append(\"- routers_eos_eol_by_sku.csv (replacements)\")\n",
+    "    return \"\\n\".join(lines)\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Customer-ready email summary (single lookup only)\n",
+    "# ============================\n",
+    "\n",
+    "def build_customer_email(life_row: pd.Series, status: str, eos: str, eol: str, repl: Dict[str,Any], ant: Dict[str,Any], link5: str) -> str:\n",
+    "    \"\"\"Email-style summary the rep can paste to a customer (lightly sales-y).\"\"\"\n",
+    "    current = f\"{life_row.get('sku','')} — {life_row.get('description','')}\".strip(\" —\")\n",
+    "    repl5 = str(repl.get(\"repl_5g\",\"\") or \"\").strip()\n",
+    "    repl4 = str(repl.get(\"repl_4g\",\"\") or \"\").strip()\n",
+    "\n",
+    "    st = ant.get(\"stationary_omni\", {}) or {}\n",
+    "    vh = ant.get(\"vehicle_omni\", {}) or {}\n",
+    "\n",
+    "    lines = []\n",
+    "    lines.append(\"Subject: Router replacement recommendation\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"Hi there,\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(f\"We reviewed your current router (**{current}**) and recommend the following path forward:\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(f\"- **Status:** {status}\")\n",
+    "    lines.append(f\"- **End of Sale:** {eos}\")\n",
+    "    lines.append(f\"- **End of Life:** {eol}\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"**Recommended replacement (5G):**\")\n",
+    "    lines.append(f\"- {repl5 if repl5 else 'Not listed'}\")\n",
+    "    if link5:\n",
+    "        lines.append(f\"- Manufacturer page (best effort): {link5}\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"**Optional 4G alternative (if needed):**\")\n",
+    "    lines.append(f\"- {repl4 if repl4 and repl4.lower() != 'not applicable' else 'Not applicable'}\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"**Antenna suggestions (Parsec):**\")\n",
+    "    lines.append(f\"- Stationary (Omni): {st.get('name','')} (PN {st.get('part_number','')})\")\n",
+    "    lines.append(f\"- Vehicle (Omni): {vh.get('name','')} (PN {vh.get('part_number','')})\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"If you’d like, we can confirm the best-fit option for your install environment and provide pricing.\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"Contact Peter Dunn @ 786.999.9127 or peter.dunn@masterstelecom.com for pricing.\")\n",
+    "    lines.append(\"\")\n",
+    "    lines.append(\"Thanks,\")\n",
+    "    lines.append(\"Peter Dunn\")\n",
+    "    return \"\\n\".join(lines)\n",
+    "\n",
+    "def generate_customer_email(st_json: str) -> str:\n",
+    "    st = state_load(st_json)\n",
+    "    if not st or \"row_idx\" not in st:\n",
+    "        return \"Run a lookup first.\"\n",
+    "    try:\n",
+    "        life_row = df_eos.iloc[int(st[\"row_idx\"])]\n",
+    "    except Exception:\n",
+    "        return \"Run a lookup first.\"\n",
+    "\n",
+    "    eos, eol, status = row_to_dates_and_status(life_row)\n",
+    "    repl = st.get(\"repl\", {}) or {}\n",
+    "    ant = st.get(\"ant\", {}) or {}\n",
+    "\n",
+    "    canon_make = str(life_row.get(\"_canon_make\",\"UNKNOWN\"))\n",
+    "    url5 = _best_effort_manufacturer_url(str(repl.get(\"repl_5g\",\"\") or \"\"), canon_make)\n",
+    "    return build_customer_email(life_row, status, eos, eol, repl, ant, url5)\n",
+    "\n",
+    "# ============================\n",
+    "# Gradio callbacks\n",
+    "# IMPORTANT: no dict state and ALL events have api_name=False (prevents api_info schema generation)\n",
+    "# ============================\n",
+    "def run_lookup(user_text: str, st_json: str):\n",
+    "    user_text = str(user_text or \"\").strip()\n",
+    "    if not user_text:\n",
+    "        return \"Enter a router SKU/model.\", \"\", None, None, \"\", gr.update(visible=False), gr.update(visible=False), \"{}\", \"\", \"\"\n",
+    "\n",
+    "    res = resolve_device(user_text)\n",
+    "\n",
+    "    if res.get(\"mode\") == \"pick\":\n",
+    "        opts = res.get(\"options\", [])\n",
+    "        choices = [o[\"label\"] for o in opts]\n",
+    "        st2 = {\"mode\":\"pick\",\"options\": opts, \"raw\": user_text}\n",
+    "        return \"Did you mean A or B? Pick one, then click Use selection.\", \"\", None, None, \"\", gr.update(choices=choices, value=None, visible=True), gr.update(visible=True), state_dump(st2), \"\", \"\"\n",
+    "\n",
+    "    if res.get(\"mode\") != \"ok\":\n",
+    "        return \"Not found.\", \"\", None, None, \"\", gr.update(visible=False), gr.update(visible=False), \"{}\", \"\", \"\"\n",
+    "\n",
+    "    life_row = df_eos.iloc[int(res[\"row_idx\"])]\n",
+    "    eos, eol, status = row_to_dates_and_status(life_row)\n",
+    "\n",
+    "    repl = pick_replacements_lifecycle(life_row, status, use_gpt=True)\n",
+    "    canon_make = str(life_row.get(\"_canon_make\",\"UNKNOWN\"))\n",
+    "    mimo = infer_mimo_for_5g(repl.get(\"repl_5g\",\"\"))\n",
+    "    tech = \"5G\" if repl.get(\"repl_5g\") and repl.get(\"repl_5g\") != \"Not listed\" else (\"4G\" if device_is_4g(life_row) else \"Unknown\")\n",
+    "    ant = antenna_options_for(repl.get(\"repl_5g\") or str(life_row.get(\"sku\",\"\")), tech, mimo)\n",
+    "\n",
+    "    output = assemble_output(life_row, status, eos, eol, repl, ant)\n",
+    "    st_out = {\"row_idx\": int(res[\"row_idx\"]), \"repl\": repl, \"ant\": ant, \"raw\": user_text}\n",
+    "    url5 = _best_effort_manufacturer_url(repl.get('repl_5g',''), canon_make)\n",
+    "    link = f\"**5G manufacturer page (best effort):** {url5}\" if url5 else \"\"\n",
+    "    feat_df = build_replacement_features_table(repl.get('repl_4g',''), repl.get('repl_5g',''), canon_make)\n",
+    "    fit = build_fit_table(repl.get('repl_4g',''), repl.get('repl_5g',''), canon_make)\n",
+    "    return output, link, feat_df, fit, \"\", gr.update(visible=False), gr.update(visible=False), state_dump(st_out), \"\", \"\"\n",
+    "\n",
+    "def use_selection(selected_label: str, st_json: str):\n",
+    "    st = state_load(st_json)\n",
+    "    if not st or st.get(\"mode\") != \"pick\":\n",
+    "        return \"Run a search first.\", \"\", None, None, \"\", gr.update(visible=False), gr.update(visible=False), \"{}\", \"\", \"\"\n",
+    "\n",
+    "    if not selected_label:\n",
+    "        return \"Pick A or B first.\", \"\", None, None, \"\", gr.update(visible=True), gr.update(visible=True), st_json, \"\", \"\"\n",
+    "\n",
+    "    chosen_row = None\n",
+    "    for o in st.get(\"options\", []):\n",
+    "        if o.get(\"label\") == selected_label:\n",
+    "            chosen_row = int(o[\"row_idx\"])\n",
+    "            break\n",
+    "    if chosen_row is None:\n",
+    "        return \"Pick a valid option.\", \"\", None, None, \"\", gr.update(visible=True), gr.update(visible=True), st_json, \"\", \"\"\n",
+    "\n",
+    "    life_row = df_eos.iloc[int(chosen_row)]\n",
+    "    eos, eol, status = row_to_dates_and_status(life_row)\n",
+    "\n",
+    "    repl = pick_replacements_lifecycle(life_row, status, use_gpt=True)\n",
+    "    canon_make = str(life_row.get(\"_canon_make\",\"UNKNOWN\"))\n",
+    "    mimo = infer_mimo_for_5g(repl.get(\"repl_5g\",\"\"))\n",
+    "    tech = \"5G\" if repl.get(\"repl_5g\") and repl.get(\"repl_5g\") != \"Not listed\" else (\"4G\" if device_is_4g(life_row) else \"Unknown\")\n",
+    "    ant = antenna_options_for(repl.get(\"repl_5g\") or str(life_row.get(\"sku\",\"\")), tech, mimo)\n",
+    "\n",
+    "    output = assemble_output(life_row, status, eos, eol, repl, ant)\n",
+    "    st_out = {\"row_idx\": int(chosen_row), \"repl\": repl, \"ant\": ant, \"raw\": st.get(\"raw\",\"\")}\n",
+    "    url5 = _best_effort_manufacturer_url(repl.get('repl_5g',''), canon_make)\n",
+    "    link = f\"**5G manufacturer page (best effort):** {url5}\" if url5 else \"\"\n",
+    "    feat_df = build_replacement_features_table(repl.get('repl_4g',''), repl.get('repl_5g',''), canon_make)\n",
+    "    fit = build_fit_table(repl.get('repl_4g',''), repl.get('repl_5g',''), canon_make)\n",
+    "    return output, link, feat_df, fit, \"\", gr.update(visible=False), gr.update(visible=False), state_dump(st_out), \"\", \"\"\n",
+    "\n",
+    "def make_install_ready(st_json: str):\n",
+    "    st = state_load(st_json)\n",
+    "    if not st or \"row_idx\" not in st:\n",
+    "        return \"Run a lookup first.\"\n",
+    "    life_row = df_eos.iloc[int(st[\"row_idx\"])]\n",
+    "    current_sku = str(life_row.get(\"sku\",\"\") or \"\")\n",
+    "    return install_ready_checklist(current_sku, st.get(\"repl\", {}) or {}, st.get(\"ant\", {}) or {})\n",
+    "\n",
+    "\n",
+    "\n",
+    "# ============================\n",
+    "# Q&A about the suggested device (post-recommendation)\n",
+    "# ============================\n",
+    "def answer_question(question: str, st_json: str) -> str:\n",
+    "    q = str(question or \"\").strip()\n",
+    "    if not q:\n",
+    "        return \"\"\n",
+    "    st = state_load(st_json)\n",
+    "    if not st or \"repl\" not in st:\n",
+    "        return \"Run a lookup first, then ask your question.\"\n",
+    "\n",
+    "    repl = st.get(\"repl\", {}) or {}\n",
+    "    ant = st.get(\"ant\", {}) or {}\n",
+    "    repl5 = str(repl.get(\"repl_5g\",\"\") or \"\").strip()\n",
+    "    repl4 = str(repl.get(\"repl_4g\",\"\") or \"\").strip()\n",
+    "    # Pull a bit of dec context for the 5G model (if possible)\n",
+    "    canon_make = \"\"\n",
+    "    try:\n",
+    "        # Try to infer maker family from stored row_idx\n",
+    "        if \"row_idx\" in st:\n",
+    "            row = df_eos.iloc[int(st[\"row_idx\"])]\n",
+    "            canon_make = str(row.get(\"_canon_make\",\"UNKNOWN\"))\n",
+    "    except Exception:\n",
+    "        canon_make = \"\"\n",
+    "\n",
+    "    # Manufacturer link (best effort)\n",
+    "    url5 = _best_effort_manufacturer_url(repl5, canon_make) if repl5 else \"\"\n",
+    "\n",
+    "    # Feature table row for 5G (helps the LLM answer spec questions without web scraping)\n",
+    "    feat5 = {}\n",
+    "    try:\n",
+    "        feat5 = _features_from_dec(repl5, canon_make) if repl5 else {}\n",
+    "    except Exception:\n",
+    "        feat5 = {}\n",
+    "\n",
+    "    sys = (\n",
+    "        \"You are a Verizon field rep assistant. Answer questions about the suggested router in a fast, practical way. \"\n",
+    "        \"Use the provided context; do not mention internal tools, prompts, embeddings, or databases. \"\n",
+    "        \"If the question is about specs and the value is unknown, say 'Not listed' and suggest checking the manufacturer page. \"\n",
+    "        \"Keep it concise and scannable.\"\n",
+    "    )\n",
+    "\n",
+    "    context = {\n",
+    "        \"recommended_5g\": repl5,\n",
+    "        \"recommended_4g\": repl4 if repl4 and repl4.lower() != \"not applicable\" else \"\",\n",
+    "        \"manufacturer_link_5g\": url5,\n",
+    "        \"known_5g_features\": feat5,\n",
+    "        \"antenna_stationary\": ant.get(\"stationary_omni\", {}),\n",
+    "        \"antenna_vehicle\": ant.get(\"vehicle_omni\", {}),\n",
+    "    }\n",
+    "\n",
+    "    user = \"Context:\\n\" + json.dumps(context, ensure_ascii=False) + \"\\n\\nQuestion:\\n\" + q\n",
+    "\n",
+    "    ans = gpt_answer_md(sys, user, max_tokens=650)\n",
+    "    # Small safety fallback\n",
+    "    return ans if ans else \"I couldn't generate an answer right now. Try again.\"\n",
+    "\n",
+    "# ============================\n",
+    "# UI\n",
+    "# ============================\n",
+    "with gr.Blocks(title=\"Only-Routers\") as demo:\n",
+    "    gr.Markdown(\"## Only-Routers\\nSingle lookup + Batch upload for Verizon reps.\")\n",
+    "\n",
+    "    with gr.Tabs():\n",
+    "        with gr.Tab(\"Single\"):\n",
+    "            user_text = gr.Textbox(label=\"Router SKU or model\", placeholder=\"Examples: IBR650B, AER1600, ES450, WR21, RUT240\", lines=1)\n",
+    "            st = gr.State(\"{}\")  # JSON string\n",
+    "\n",
+    "            check_btn = gr.Button(\"Check\", variant=\"primary\")\n",
+    "            pick_dd = gr.Dropdown(label=\"Pick A or B\", choices=[], visible=False)\n",
+    "            use_btn = gr.Button(\"Use selection\", visible=False)\n",
+    "\n",
+    "            output_md = gr.Markdown()\n",
+    "\n",
+    "            link_md = gr.Markdown()\n",
+    "            features_df = gr.Dataframe(headers=FEATURE_COLS, interactive=False, wrap=True)\n",
+    "            fit_df = gr.Dataframe(headers=FIT_COLS, interactive=False, wrap=True)\n",
+    "\n",
+    "\n",
+    "\n",
+    "gr.Markdown(\"### Questions about the suggested device?\")\n",
+    "question_box = gr.Textbox(label=\"Ask a question (optional)\", placeholder=\"Example: Does the 5G device support dual-SIM? What antenna ports does it have?\", lines=2)\n",
+    "ask_btn = gr.Button(\"Ask\", variant=\"secondary\")\n",
+    "qa_md = gr.Markdown()\n",
+    "\n",
+    "\n",
+    "            install_btn = gr.Button(\"Make install-ready checklist\")\n",
+    "            install_md = gr.Markdown()\n",
+    "\n",
+    "            gr.Markdown(\"### Customer-ready email (copy/paste)\")\n",
+    "            email_btn = gr.Button(\"Generate customer email\")\n",
+    "            customer_email_box = gr.Textbox(label=\"Email summary\", lines=14)\n",
+    "\n",
+    "\n",
+    "            check_btn.click(fn=run_lookup, inputs=[user_text, st], outputs=[output_md, link_md, features_df, fit_df, qa_md, pick_dd, use_btn, st, install_md, customer_email_box], api_name=False)\n",
+    "            use_btn.click(fn=use_selection, inputs=[pick_dd, st], outputs=[output_md, link_md, features_df, fit_df, qa_md, pick_dd, use_btn, st, install_md, customer_email_box], api_name=False)\n",
+    "            install_btn.click(fn=make_install_ready, inputs=[st], outputs=[install_md], api_name=False)\n",
+    "            email_btn.click(fn=generate_customer_email, inputs=[st], outputs=[customer_email_box], api_name=False)\n",
+    "            ask_btn.click(fn=answer_question, inputs=[question_box, st], outputs=[qa_md], api_name=False)\n",
+    "\n",
+    "        with gr.Tab(\"Batch\"):\n",
+    "            gr.Markdown(\"Paste one per line or upload a CSV (first column). Batch runs fast (no GPT).\")\n",
+    "            batch_text = gr.Textbox(label=\"Paste devices (one per line)\", lines=8, placeholder=\"WR21\\nRUT240\\nIBR650B\")\n",
+    "            batch_file = gr.File(label=\"Upload CSV\", file_types=[\".csv\"])\n",
+    "            include_ant = gr.Checkbox(label=\"Include antenna picks (slower)\", value=False)\n",
+    "            run_btn = gr.Button(\"Run batch\", variant=\"primary\")\n",
+    "\n",
+    "            summary_md = gr.Markdown()\n",
+    "            rollup_md = gr.Markdown()\n",
+    "            table = gr.Dataframe(interactive=False, wrap=True)\n",
+    "            dl = gr.File(label=\"Download results CSV\")\n",
+    "\n",
+    "            run_btn.click(fn=run_batch, inputs=[batch_text, batch_file, include_ant], outputs=[summary_md, table, dl, rollup_md], api_name=False)\n",
+    "\n",
+    "# IMPORTANT: On Spaces, demo.launch() is correct; do NOT use share=True.\n",
+    "demo.launch(show_api=False)\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

app.py

@@ -902,9 +902,9 @@ HTTP_HEADERS = {
 HTTP_TIMEOUT = 12
 
 def _best_effort_manufacturer_url(model: str, canon_make: str) -> str:
-    \"\"\"Try to find a manufacturer page or datasheet link using simple on-domain searches.
+    """Try to find a manufacturer page or datasheet link using simple on-domain searches.
     If we can't confirm a page, return the manufacturer homepage for the maker family.
-    \"\"\"
+    """
     model = str(model or "").strip()
     if not model or model in {"Not listed", "Not applicable"}:
         return ""
@@ -968,7 +968,7 @@ def _best_effort_manufacturer_url(model: str, canon_make: str) -> str:
     return f"https://{d0}"
 
 def _features_from_dec(model: str, canon_make: str) -> Dict[str, str]:
-    \"\"\"Lookup a router model in dec2025routers.csv and return the key feature fields.\"\"\"
+    """Lookup a router model in dec2025routers.csv and return the key feature fields."""
     if not model or model in {"Not listed", "Not applicable"}:
         return {k: "Not listed" for k in FEATURE_COLS[1:]}
 
@@ -992,7 +992,7 @@ def _features_from_dec(model: str, canon_make: str) -> Dict[str, str]:
     }
 
 def _gpt_fill_feature_row(device_label: str, model: str, canon_make: str, row: Dict[str, str]) -> Dict[str, str]:
-    \"\"\"If dec can't supply values, ask GPT to fill missing ones (best guess).\"\"\"
+    """If dec can't supply values, ask GPT to fill missing ones (best guess)."""
     if client is None:
         return row
 
@@ -1045,7 +1045,7 @@ def build_replacement_features_table(repl_4g: str, repl_5g: str, canon_make: str
 FIT_COLS = ["Device", "Fit badges", "Ethernet ports", "Battery"]
 
 def _parse_ethernet_ports(wan_field: str, lan_field: str) -> str:
-    \"\"\"Best-effort total ethernet ports based on WAN/LAN text.\"\"\"
+    """Best-effort total ethernet ports based on WAN/LAN text."""
     def _count(field: str) -> int:
         s = str(field or "")
         # Common forms: "1x GbE", "2 x 10/100", "WAN: 1", etc.
@@ -1326,7 +1326,7 @@ def assemble_output(life_row: pd.Series, status: str, eos: str, eol: str, repl:
 # ============================
 
 def build_customer_email(life_row: pd.Series, status: str, eos: str, eol: str, repl: Dict[str,Any], ant: Dict[str,Any], link5: str) -> str:
-    \"\"\"Email-style summary the rep can paste to a customer (lightly sales-y).\"\"\"
+    """Email-style summary the rep can paste to a customer (lightly sales-y)."""
     current = f"{life_row.get('sku','')} — {life_row.get('description','')}".strip(" —")
     repl5 = str(repl.get("repl_5g","") or "").strip()
     repl4 = str(repl.get("repl_4g","") or "").strip()