Update app.py

app.py

@@ -1,37 +1,31 @@
 # -*- coding: utf-8 -*-
 """
-Text2Receipt — Hugging Face Gradio Space
-=========================================
+Text2Receipt — HF Space  (final clean version)
+================================================
 Pipeline:
-  raw Hebrew note
-    → fine-tuned parser (gemma-2-2b-it + LoRA)
-    → agentic clarifier  (+5% bonus)
-    → deterministic complete()  (VAT · serial · allocation number)
-    → fiscal document display
-    → FAISS recommender  (3 similar past receipts)
-Repos:
-  Dataset : yonilev/Text2Receipt
-  Model   : yonilev/Text2Receipt-parser
-  Space   : yonilev/Text2Receipt
+  free-text Hebrew income note
+  → gemma-2-2b fine-tuned parser  (extracts client / items)
+  → fiscal decision layer          (VAT cues, doc-type, allocation)
+  → t2r_core.complete()            (deterministic arithmetic)
+  → Gradio UI                      (render document + FAISS recommendations)
 """
-import os, json, random
-# ── Disable SSR BEFORE importing gradio ──────────────────────────────────────
-# On HF Spaces, Gradio runs in SSR mode by default. In Gradio 6.17 this breaks
-# value-passing from Textbox inputs (raw_text arrives empty). Force CSR.
-os.environ["GRADIO_SSR_MODE"] = "false"
+
+import json, random, re as _re
 import numpy as np
 import gradio as gr
+import t2r_core as core
+import datetime as _dt
 
-DATASET_REPO = "yonilev/Text2Receipt"
-MODEL_REPO   = "yonilev/Text2Receipt-parser"
-BASE_MODEL   = os.environ.get("BASE_MODEL", "unsloth/gemma-2-2b-it")
+# ─────────────────────────────────────────────────────────────────────────────
+# 1. Lazy model + FAISS state
+# ─────────────────────────────────────────────────────────────────────────────
+BASE_MODEL  = "unsloth/gemma-2-2b-it"
+MODEL_REPO  = "yonilev/Text2Receipt-parser"
 
-import t2r_core as core
+_STATE = dict(ready=False, tok=None, model=None, enc=None,
+              index=None, store=None, e5_family=False,
+              rng=random.Random(42), device="cpu")
 
-# ═════════════════════════════════════════════════════════════════════════════
-# 1. Lazy init
-# ═════════════════════════════════════════════════════════════════════════════
-_STATE: dict = {"ready": False}
 
 def _lazy_init():
     if _STATE["ready"]:
@@ -98,9 +92,9 @@ def _lazy_init():
     print(f"✅ FAISS ready — {emb.shape[0]} vectors")
 
 
-# ═════════════════════════════════════════════════════════════════════════════
-# 2. Parser
-# ═════════════════════════════════════════════════════════════════════════════
+# ─────────────────────────────────────────────────────────────────────────────
+# 2. Parser + post-processing cleaner
+# ─────────────────────────────────────────────────────────────────────────────
 INSTRUCTION = (
     "אתה ממיר הערת הכנסה חופשית בעברית למבנה JSON. "
     "חלץ אך ורק את מה שכתוב בהערה: שם הלקוח (client_name), "
@@ -109,9 +103,18 @@ INSTRUCTION = (
     "החזר JSON תקין בלבד, ללא טקסט נוסף."
 )
 
+_VAT_NOISE = [
+    "לפני מע\"מ", "כולל מע\"מ", "לפני מע״מ", "כולל מע״מ",
+    "בתוספת מע\"מ", "+ מע\"מ", "+מע\"מ", "מע\"מ", "מע״מ", "מעמ",
+]
+_NOISE_SET = {"לפני", "כולל", "נטו", "ברוטו", "מס", "+", "פלוס",
+              "מע\"מ", "מע״מ", "מעמ"}
+
+
 def _build_prompt(raw_text):
     return f"{INSTRUCTION}\n\nהערה: {raw_text}\n\nJSON:"
 
+
 def _extract_json(text):
     s = text.find("{")
     if s < 0: return None
@@ -125,6 +128,37 @@ def _extract_json(text):
                 except: return None
     return None
 
+
+def _clean_name(name):
+    """Strip VAT-related noise the model may accidentally grab as a client name."""
+    name = (name or "").strip()
+    for tok in _VAT_NOISE:
+        name = name.replace(tok, "").strip(" ,.-+")
+    return "" if (name in _NOISE_SET or len(name) <= 1) else name
+
+
+def _clean_parse(parse):
+    """Post-process raw model output: remove noise, validate structure."""
+    if not isinstance(parse, dict):
+        return parse
+    # Clean client name
+    parse["client_name"] = _clean_name(parse.get("client_name", ""))
+    # Clean items
+    clean_items = []
+    for it in (parse.get("items") or []):
+        desc = (it.get("description") or "").strip()
+        for tok in _VAT_NOISE:
+            desc = desc.replace(tok, "").strip(" ,.-+")
+        if desc in _NOISE_SET or len(desc) <= 0:
+            desc = "שירות"
+        it = dict(it)
+        it["description"] = desc
+        if it.get("unit_price"):
+            clean_items.append(it)
+    parse["items"] = clean_items
+    return parse
+
+
 def model_parse(raw_text):
     tok, model = _STATE["tok"], _STATE["model"]
     msgs   = [{"role": "user", "content": _build_prompt(raw_text)}]
@@ -135,15 +169,9 @@ def model_parse(raw_text):
         out = model.generate(**enc, max_new_tokens=128, do_sample=False,
                              pad_token_id=tok.pad_token_id)
     decoded = tok.decode(out[0, enc["input_ids"].shape[1]:], skip_special_tokens=True)
-    return _extract_json(decoded)
-
+    raw = _extract_json(decoded)
+    return _clean_parse(raw) if raw else None
 
-# ═════════════════════════════════════════════════════════════════════════════
-# 3. Agentic clarification
-# ═════════════════════════════════════════════════════════════════════════════
-import datetime as _dt
-
-def _today(): return _dt.date.today()
 
 def _missing_fields(parse):
     if parse is None: return ["parse_failed"]
@@ -152,50 +180,10 @@ def _missing_fields(parse):
     if not parse.get("items"):                   m.append("items")
     return m
 
-def _needs_allocation_clarification(parse):
-    if not parse: return False
-    subtotal  = sum(it.get("unit_price", 0) * it.get("quantity", 1)
-                    for it in parse.get("items", []))
-    threshold = core.allocation_threshold_for_date(_today())
-    return bool(parse.get("client_is_business")) and subtotal >= threshold * 0.8
-
-def clarification_questions(parse, answers):
-    missing = _missing_fields(parse)
-    if "parse_failed" in missing:
-        return ["לא הצלחתי לחלץ פרטים. נסח מחדש — לדוגמה: 'קיבלתי 500₪ מ[שם] על [שירות]'"]
-    qs = []
-    if "client_name" in missing and "client_name" not in answers:
-        qs.append("מה שם הלקוח?")
-    if "items" in missing and "items" not in answers:
-        qs.append("מה השירות/המוצר ובאיזה מחיר?")
-    if "doc_type" not in answers:
-        if parse and _needs_allocation_clarification(parse):
-            qs.append('האם להנפיק חשבונית מס/קבלה (לעסק עם מע"מ) או קבלה פשוטה?')
-        elif parse and not parse.get("doc_type"):
-            qs.append("איזה מסמך? קבלה / חשבונית מס / חשבונית מס וקבלה")
-    return qs[:3]
-
-def apply_answers(parse, answers):
-    p = dict(parse or {})
-    if "client_name" in answers: p["client_name"] = answers["client_name"]
-    if "doc_type" in answers:
-        raw = answers["doc_type"].strip()
-        p["doc_type"] = ("tax_invoice_receipt" if "מס/קבלה" in raw or "מס וקבלה" in raw
-                         else "tax_invoice" if "מס" in raw else "receipt")
-    p.setdefault("doc_type",           "receipt")
-    p.setdefault("date",               _today().isoformat())
-    p.setdefault("payment_method",     "bank_transfer")
-    p.setdefault("amount_basis",       "net")
-    p.setdefault("currency",           "ILS")
-    p.setdefault("client_tax_id",      None)
-    p.setdefault("client_is_business", False)
-    p.setdefault("items",              [])
-    return p
-
-
-# ═════════════════════════════════════════════════════════════════════════════
-# 4. FAISS recommender
-# ═════════════════════════════════════════════════════════════════════════════
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 3. FAISS recommender
+# ────────────────────────────���────────────────────────────────────────────────
 def recommend(query_text, k=3):
     pref  = "query: " if _STATE["e5_family"] else ""
     q_emb = _STATE["enc"].encode([pref + query_text],
@@ -210,9 +198,186 @@ def recommend(query_text, k=3):
     return out
 
 
-# ═════════════════════════════════════════════════════════════════════════════
+# ─────────────────────────────────────────────────────────────────────────────
+# 4. Fiscal decision layer  (app-level, never touches frozen t2r_core.py)
+# ─────────────────────────────────────────────────────────────────────────────
+DEMO_ISSUER_BASE = {"name": "מים שקטים", "tax_id": "962569844"}
+
+_GROSS_PATTERNS = [r"כולל\s*מע[\"״]?מ", r"כולל\s*מס", r"גרוס", r"ברוטו", r"מחיר\s*כולל"]
+_NET_PATTERNS   = [r"לפני\s*מע[\"״]?מ", r"\+\s*מע[\"״]?מ", r"בתוספת\s*מע[\"״]?מ",
+                   r"פלוס\s*מע[\"״]?מ", r"נטו", r"לא\s*כולל\s*מע[\"״]?מ",
+                   r"בלי\s*מע[\"״]?מ", r"לפני\s*מס"]
+
+def _text_has(text, patterns):
+    t = text.lower()
+    return any(_re.search(p, t) for p in patterns)
+
+def _detect_date(text):
+    for pat, order in [(r"(\d{1,2})[/\-\.](\d{1,2})[/\-\.](\d{2,4})", "dmy"),
+                       (r"(\d{4})[/\-\.](\d{1,2})[/\-\.](\d{1,2})", "ymd")]:
+        m = _re.search(pat, text)
+        if m:
+            try:
+                g = [int(x) for x in m.groups()]
+                d, mo, y = (g[0], g[1], g[2]) if order == "dmy" else (g[2], g[1], g[0])
+                if y < 100: y += 2000
+                return f"{y:04d}-{mo:02d}-{d:02d}"
+            except Exception:
+                pass
+    return None
+
+def _today_str():
+    return _dt.date.today().isoformat()
+
+
+def decide_fiscal(note, issuer_status, parse, doc_radio,
+                  clarify_vat=None, clarify_tax_id=None):
+    """
+    Returns (issuer_dict, final_parse, questions_dict).
+    questions_dict empty → ready to generate.
+    """
+    # ── 1. Issuer required — early return if missing ──────────────────────
+    if not issuer_status:
+        return None, None, {"issuer_status": "בחר סטטוס מנפיק למעלה (עוסק פטור / עוסק מורשה)."}
+
+    questions = {}
+    exempt = (issuer_status == "פטור")
+    issuer = {**DEMO_ISSUER_BASE,
+              "status": "exempt_dealer" if exempt else "authorized_dealer",
+              "is_company": False}
+
+    # ── 2. Mandatory content ──────────────────────────────────────────────
+    if not parse.get("client_name", "").strip():
+        questions["client_name"] = "שם הלקוח / המשלם?"
+    if not parse.get("items"):
+        questions["items"] = "מה השירות/המוצר ובאיזה סכום? (למשל: ייעוץ ₪500)"
+
+    # ── 3. Date: detect from note → today ────────────────────────────────
+    parse = dict(parse)
+    parse["date"] = _detect_date(note) or _today_str()
+
+    # ── 4. Doc type + VAT basis ───────────────────────────────────────────
+    is_gross    = _text_has(note, _GROSS_PATTERNS) or clarify_vat == "כולל מע״מ"
+    is_net      = _text_has(note, _NET_PATTERNS)   or clarify_vat == "לפני מע״מ"
+    has_vat_cue = is_gross or is_net
+
+    if exempt:
+        doc_type, amount_basis = "receipt", "net"
+
+    elif doc_radio and doc_radio != "זיהוי אוטומטי":
+        doc_type = {"קבלה": "receipt",
+                    "חשבונית מס": "tax_invoice",
+                    "חשבונית מס וקבלה": "tax_invoice_receipt"}.get(doc_radio, "receipt")
+        if doc_type != "receipt":
+            if is_gross:
+                amount_basis = "gross"
+            elif is_net:
+                amount_basis = "net"
+            elif clarify_vat in (None, "לא צוין"):
+                total = sum(i.get("unit_price", 0) * i.get("quantity", 1)
+                            for i in parse.get("items", []))
+                questions["vat_basis"] = f"הסכום (₪{total:,.0f}) — כולל מע״מ או לפני מע״מ?"
+                amount_basis = "net"
+            else:
+                amount_basis = "net"
+        else:
+            amount_basis = "net"
+
+    else:
+        # Auto: VAT cue in note → tax_invoice_receipt; otherwise → receipt (no VAT line)
+        if has_vat_cue:
+            doc_type, amount_basis = "tax_invoice_receipt", "gross" if is_gross else "net"
+        else:
+            doc_type, amount_basis = "receipt", "net"
+
+    # ── 5. Allocation: only בע"מ / חברה + amount ≥ threshold ─────────────
+    if doc_type in ("tax_invoice", "tax_invoice_receipt") and not exempt:
+        subtotal  = sum(it.get("unit_price", 0) * it.get("quantity", 1)
+                        for it in parse.get("items", []))
+        threshold = core.allocation_threshold_for_date(
+                        _dt.date.fromisoformat(parse["date"]))
+        name      = parse.get("client_name", "")
+        is_corp   = any(kw in name for kw in ['בע"מ', "בעמ", "חברה", "בע'מ", "בע.מ"])
+        if is_corp and subtotal >= threshold:
+            if clarify_tax_id and clarify_tax_id.strip():
+                parse["client_tax_id"] = clarify_tax_id.strip()
+            elif not parse.get("client_tax_id"):
+                questions["client_tax_id"] = f"ח.פ. הלקוח (נדרש להקצאה — עסקה ≥ ₪{threshold:,})"
+
+    # ── 6. Finalise parse ─────────────────────────────────────────────────
+    parse.update({"doc_type": doc_type, "amount_basis": amount_basis})
+    parse.setdefault("payment_method",     "bank_transfer")
+    parse.setdefault("currency",           "ILS")
+    parse.setdefault("client_tax_id",      None)
+    parse.setdefault("client_is_business", False)
+    parse.setdefault("client_name",        "—")
+    parse.setdefault("items",              [])
+
+    return issuer, parse, questions
+
+
+def _build_clarification_message(questions):
+    icons = {"issuer_status": "👤", "client_name": "👤",
+             "items": "📦", "vat_basis": "💰", "client_tax_id": "🔖"}
+    lines = "\n".join(f"{icons.get(k,'•')} {v}" for k, v in questions.items())
+    return f"❓ **כדי להפיק את המסמך, חסר מידע:**\n\n{lines}\n\nהוסף את הפרטים לפתק ולחץ **הפק מסמך** שוב."
+
+
+# ─────────────────────────────────────────────────────────────────────────────
 # 5. HTML renderers
-# ═════════════════════════════════════════════════════════════════════════════
+# ─────────────────────────────────────────────────────────────────────────────
+def render_parse_panel(parse, completed, show_vat):
+    items = parse.get("items", [])
+    item_rows = "".join(
+        f'<div style="display:flex;justify-content:space-between;font-size:12px;'
+        f'color:#334155;padding:3px 0;"><span>{it.get("description","")}</span>'
+        f'<span style="color:#64748b;">{it.get("quantity",1)} × ₪{it.get("unit_price",0):,.0f}</span></div>'
+        for it in items)
+
+    chips = (
+        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
+        f'border-radius:999px;padding:2px 9px;font-size:11px;">לקוח: {parse.get("client_name","—")}</span> '
+        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
+        f'border-radius:999px;padding:2px 9px;font-size:11px;">'
+        f'{"עסק" if parse.get("client_is_business") else "פרטי"}</span> '
+        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
+        f'border-radius:999px;padding:2px 9px;font-size:11px;">{len(items)} פריטים</span>')
+
+    alloc_line = (
+        '<div style="font-size:12px;color:#334155;padding:3px 0;">'
+        'מספר הקצאה <span style="color:#0d9488;">✓ נדרש</span></div>'
+        if completed.get("allocation_required") else "")
+
+    vat_line = (
+        f'<div style="font-size:12px;color:#334155;padding:3px 0;">'
+        f'מע"מ {int(round(completed.get("vat_rate",0)*100))}% · ₪{completed.get("vat_amount",0):,.2f}</div>'
+        if show_vat else
+        '<div style="font-size:12px;color:#94a3b8;padding:3px 0;">ללא פירוק מע"מ (קבלה)</div>')
+
+    return f"""
+<div dir="rtl" style="font-family:'Segoe UI',Arial,sans-serif;max-width:660px;
+     margin:14px auto 0;display:grid;grid-template-columns:1fr 1fr;gap:12px;">
+  <div style="background:#f8fafc;border:1px solid #e2e8f0;border-radius:12px;padding:14px 16px;">
+    <div style="font-size:11px;font-weight:700;color:#0d9488;letter-spacing:.05em;
+         text-transform:uppercase;margin-bottom:10px;">🧠 המודל חילץ</div>
+    <div style="display:flex;flex-wrap:wrap;gap:5px;margin-bottom:10px;">{chips}</div>
+    {item_rows}
+    <div style="font-size:10px;color:#94a3b8;margin-top:10px;border-top:1px dashed #e2e8f0;
+         padding-top:8px;">gemma-2-2b + LoRA · חוזה רק <code>parse</code></div>
+  </div>
+  <div style="background:#f8fafc;border:1px solid #e2e8f0;border-radius:12px;padding:14px 16px;">
+    <div style="font-size:11px;font-weight:700;color:#7c3aed;letter-spacing:.05em;
+         text-transform:uppercase;margin-bottom:10px;">⚙️ המנוע חישב</div>
+    {vat_line}
+    <div style="font-size:12px;color:#334155;padding:3px 0;">סה"כ ₪{completed.get("total",0):,.2f}</div>
+    <div style="font-size:12px;color:#334155;padding:3px 0;">מס׳ מסמך {completed.get("serial_number","")}</div>
+    {alloc_line}
+    <div style="font-size:10px;color:#94a3b8;margin-top:10px;border-top:1px dashed #e2e8f0;
+         padding-top:8px;">דטרמיניסטי · אפס הזיות בחשבון</div>
+  </div>
+</div>"""
+
+
 def render_document(c, show_vat=True):
     issuer  = c.get("issuer", {})
     client  = c.get("client", {})
@@ -233,6 +398,17 @@ def render_document(c, show_vat=True):
         f'</tr>'
         for ln in lines)
 
+    vat_block = "" if not show_vat else f"""
+    <div style="display:flex;justify-content:space-between;font-size:13px;color:#64748b;margin-bottom:8px;">
+      <span>סכום לפני מע&quot;מ</span>
+      <span style="color:#94a3b8;">₪{c.get("subtotal",0):,.2f}</span>
+    </div>
+    <div style="display:flex;justify-content:space-between;font-size:13px;color:#64748b;
+         padding-bottom:12px;border-bottom:1px solid #334155;margin-bottom:12px;">
+      <span>מע&quot;מ {vat_pct}%</span>
+      <span style="color:#94a3b8;">₪{c.get("vat_amount",0):,.2f}</span>
+    </div>"""
+
     return f"""
 <div dir="rtl" style="font-family:'Segoe UI','Arial Hebrew',Arial,sans-serif;
      background:#0f172a;color:#e2e8f0;border-radius:16px;
@@ -289,16 +465,7 @@ def render_document(c, show_vat=True):
   </table>
 
   <div style="background:#1e293b;border-radius:10px;padding:16px 18px;">
-    {"" if not show_vat else f'''
-    <div style="display:flex;justify-content:space-between;font-size:13px;color:#64748b;margin-bottom:8px;">
-      <span>סכום לפני מע&quot;מ</span>
-      <span style="color:#94a3b8;">₪{c.get("subtotal",0):,.2f}</span>
-    </div>
-    <div style="display:flex;justify-content:space-between;font-size:13px;color:#64748b;
-         padding-bottom:12px;border-bottom:1px solid #334155;margin-bottom:12px;">
-      <span>מע&quot;מ {vat_pct}%</span>
-      <span style="color:#94a3b8;">₪{c.get("vat_amount",0):,.2f}</span>
-    </div>'''}
+    {vat_block}
     <div style="display:flex;justify-content:space-between;align-items:baseline;">
       <span style="font-size:15px;font-weight:600;color:#e2e8f0;">סה"כ לתשלום</span>
       <span style="font-size:24px;font-weight:700;color:#2dd4bf;">₪{c.get("total",0):,.2f}</span>
@@ -333,253 +500,60 @@ def render_recommendations(recs):
 </div>"""
 
 
-# ═════════════════════════════════════════════════════════════════════════════
-# 6. Pipeline
-# ═════════════════════════════════════════════════════════════════════════════
-DEMO_ISSUER_BASE = {"name": "מים שקטים", "tax_id": "962569844"}
-
-# ── VAT cue patterns ─────────────────────────────────────────────────────────
-import re as _re
-
-_GROSS_PATTERNS = [r"כולל\s*מע[\"״]?מ", r"כולל\s*מס", r"גרוס", r"ברוטו",
-                   r"מחיר\s*כולל"]
-_NET_PATTERNS   = [r"לפני\s*מע[\"״]?מ", r"\+\s*מע[\"״]?מ", r"בתוספת\s*מע[\"״]?מ",
-                   r"פלוס\s*מע[\"״]?מ", r"נטו", r"לא\s*כולל\s*מע[\"״]?מ",
-                   r"בלי\s*מע[\"״]?מ", r"לפני\s*מס"]
-_EXEMPT_PATTERNS= [r"עוסק\s*פטור", r"פטור\s*ממע[\"״]?מ", r"ללא\s*מע[\"״]?מ"]
-_PAYMENT_RECV   = [r"קיבלתי", r"שילמ", r"העביר", r"שולם", r"נכנס", r"הועבר",
-                   r"ביט", r"פיי", r"מזומן"]
-
-def _text_has(text, patterns):
-    t = text.lower()
-    return any(_re.search(p, t) for p in patterns)
-
-def _detect_date_from_text(text):
-    """Try to extract a date from free text. Returns ISO string or None."""
-    patterns = [
-        (r"(\d{1,2})[/\-\.](\d{1,2})[/\-\.](\d{2,4})", "dmy"),
-        (r"(\d{4})[/\-\.](\d{1,2})[/\-\.](\d{1,2})", "ymd"),
-    ]
-    for pat, order in patterns:
-        m = _re.search(pat, text)
-        if m:
-            try:
-                g = [int(x) for x in m.groups()]
-                if order == "dmy":
-                    d, mo, y = g
-                    if y < 100: y += 2000
-                else:
-                    y, mo, d = g
-                return f"{y:04d}-{mo:02d}-{d:02d}"
-            except Exception:
-                pass
-    return None
-
-def render_parse_panel(parse, completed, show_vat=True):
-    """'Behind the scenes' panel — model extraction vs deterministic computation."""
-    items = parse.get("items", [])
-    item_rows = "".join(
-        f'<div style="display:flex;justify-content:space-between;font-size:12px;'
-        f'color:#334155;padding:3px 0;"><span>{it.get("description","")}</span>'
-        f'<span style="color:#64748b;">{it.get("quantity",1)} × ₪{it.get("unit_price",0):,.0f}</span></div>'
-        for it in items)
-
-    chips = (
-        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
-        f'border-radius:999px;padding:2px 9px;font-size:11px;">לקוח: {parse.get("client_name","—")}</span>'
-        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
-        f'border-radius:999px;padding:2px 9px;font-size:11px;">'
-        f'{"עסק" if parse.get("client_is_business") else "פרטי"}</span>'
-        f'<span style="background:#ecfdf5;color:#047857;border:1px solid #a7f3d0;'
-        f'border-radius:999px;padding:2px 9px;font-size:11px;">{len(items)} פריטים</span>')
-
-    alloc_line = ""
-    if completed.get("allocation_required"):
-        alloc_line = ('<div style="font-size:12px;color:#334155;padding:3px 0;">'
-                      'מספר הקצאה <span style="color:#0d9488;">✓ נדרש</span></div>')
-
-    vat_line = (f'<div style="font-size:12px;color:#334155;padding:3px 0;">'
-                f'מע"מ {int(round(completed.get("vat_rate",0)*100))}% · '
-                f'₪{completed.get("vat_amount",0):,.2f}</div>'
-                if show_vat else
-                '<div style="font-size:12px;color:#94a3b8;padding:3px 0;">ללא פירוק מע"מ (קבלה)</div>')
-
-    return f"""
-<div dir="rtl" style="font-family:'Segoe UI',Arial,sans-serif;max-width:660px;
-     margin:14px auto 0;display:grid;grid-template-columns:1fr 1fr;gap:12px;">
-  <div style="background:#f8fafc;border:1px solid #e2e8f0;border-radius:12px;padding:14px 16px;">
-    <div style="font-size:11px;font-weight:700;color:#0d9488;letter-spacing:.05em;
-         text-transform:uppercase;margin-bottom:10px;">🧠 המודל חילץ</div>
-    <div style="display:flex;flex-wrap:wrap;gap:5px;margin-bottom:10px;">{chips}</div>
-    {item_rows}
-    <div style="font-size:10px;color:#94a3b8;margin-top:10px;border-top:1px dashed #e2e8f0;
-         padding-top:8px;">gemma-2-2b + LoRA · חוזה רק <code>parse</code></div>
-  </div>
-  <div style="background:#f8fafc;border:1px solid #e2e8f0;border-radius:12px;padding:14px 16px;">
-    <div style="font-size:11px;font-weight:700;color:#7c3aed;letter-spacing:.05em;
-         text-transform:uppercase;margin-bottom:10px;">⚙️ המנוע חישב</div>
-    {vat_line}
-    <div style="font-size:12px;color:#334155;padding:3px 0;">סה"כ ₪{completed.get("total",0):,.2f}</div>
-    <div style="font-size:12px;color:#334155;padding:3px 0;">מס׳ מסמך {completed.get("serial_number","")}</div>
-    {alloc_line}
-    <div style="font-size:10px;color:#94a3b8;margin-top:10px;border-top:1px dashed #e2e8f0;
-         padding-top:8px;">דטרמיניסטי · אפס הזיות בחשבון</div>
-  </div>
-</div>"""
-
-
-def decide_fiscal(text, issuer_status, parse, explicit_doc_choice,
-                  clarify_vat_basis=None, clarify_client_tax_id=None):
-    """Returns (issuer_dict, final_parse, questions_dict)."""
-    today = _today()
-
-    # ── 1. Issuer required first — no model run needed if missing ────────────
-    if not issuer_status:
-        return None, None, {"issuer_status": "בחר סטטוס מנפיק למעלה (עוסק פטור / עוסק מורשה)."}
-
-    questions = {}
-    exempt = (issuer_status == "פטור")
-    issuer = {**DEMO_ISSUER_BASE,
-              "status": "exempt_dealer" if exempt else "authorized_dealer",
-              "is_company": False}
-
-    # ── 2. Mandatory content ──────────────────────────────────────────────────
-    if not parse.get("client_name", "").strip():
-        questions["client_name"] = "שם הלקוח / המשלם?"
-    if not parse.get("items"):
-        questions["items"] = "מה השירות/המוצר ובאיזה סכום? (למשל: ייעוץ עסקי ₪500)"
-
-    # ── 3. Date ───────────────────────────────────────────────────────────────
-    parse = dict(parse)
-    parse.setdefault("date", _detect_date_from_text(text) or today.isoformat())
-
-    # ── 4. Doc type + VAT basis ───────────────────────────────────��───────────
-    is_gross    = _text_has(text, _GROSS_PATTERNS) or clarify_vat_basis == "כולל מע״מ"
-    is_net      = _text_has(text, _NET_PATTERNS)   or clarify_vat_basis == "לפני מע״מ"
-    has_vat_cue = is_gross or is_net
-
-    if exempt:
-        doc_type, amount_basis = "receipt", "net"
-
-    elif explicit_doc_choice and explicit_doc_choice != "זיהוי אוטומטי":
-        doc_type = {"קבלה": "receipt", "חשבונית מס": "tax_invoice",
-                    "חשבונית מס וקבלה": "tax_invoice_receipt"}.get(explicit_doc_choice, "receipt")
-        if doc_type != "receipt":
-            if is_gross:
-                amount_basis = "gross"
-            elif is_net:
-                amount_basis = "net"
-            elif clarify_vat_basis in (None, "לא צוין"):
-                total = sum(i.get("unit_price",0)*i.get("quantity",1) for i in parse.get("items",[]))
-                questions["vat_basis"] = f"הסכום (₪{total:,.0f}) — כולל מע״מ או לפני מע״מ?"
-                amount_basis = "net"
-            else:
-                amount_basis = "net"
-        else:
-            amount_basis = "net"
-    else:
-        # Auto: VAT cue → tax_invoice_receipt; no cue → receipt (no VAT)
-        if has_vat_cue:
-            doc_type, amount_basis = "tax_invoice_receipt", "gross" if is_gross else "net"
-        else:
-            doc_type, amount_basis = "receipt", "net"
-
-    # ── 5. Allocation: ח.פ. only for בע"מ/חברה clients + large amounts ───────
-    if doc_type in ("tax_invoice", "tax_invoice_receipt") and not exempt:
-        subtotal  = sum(it.get("unit_price",0)*it.get("quantity",1) for it in parse.get("items",[]))
-        threshold = core.allocation_threshold_for_date(_dt.date.fromisoformat(parse["date"]))
-        name      = parse.get("client_name", "")
-        is_corp   = any(kw in name for kw in ['בע"מ', "בעמ", "חברה", "בע'מ", "בע.מ"])
-        if is_corp and subtotal >= threshold:
-            if clarify_client_tax_id and clarify_client_tax_id.strip():
-                parse["client_tax_id"] = clarify_client_tax_id.strip()
-            elif not parse.get("client_tax_id"):
-                questions["client_tax_id"] = f"ח.פ. הלקוח (נדרש להקצאה — עסקה ≥ ₪{threshold:,})"
-
-    parse.update({"doc_type": doc_type, "amount_basis": amount_basis})
-    parse.setdefault("payment_method",     "bank_transfer")
-    parse.setdefault("currency",           "ILS")
-    parse.setdefault("client_tax_id",      None)
-    parse.setdefault("client_is_business", False)
-    parse.setdefault("client_name",        "—")
-    parse.setdefault("items",              [])
-
-    return issuer, parse, questions
-
-
-def _build_clarification_message(questions: dict, note_text: str) -> str:
-    """Build ONE unified clarification message listing all missing fields."""
-    lines = []
-    for key, q in questions.items():
-        if key == "issuer_status":
-            lines.append(f"👤 {q}")
-        elif key == "client_name":
-            lines.append(f"👤 **שם לקוח:** {q}")
-        elif key == "items":
-            lines.append(f"📦 **פריטים:** {q}")
-        elif key == "vat_basis":
-            lines.append(f"💰 **בסיס מע״מ:** {q}")
-        elif key == "client_tax_id":
-            lines.append(f"🔖 **הקצאה:** {q}")
-        else:
-            lines.append(f"• {q}")
-
-    fields_text = "\n".join(lines)
-    return (f"❓ **כדי להפיק את המסמך, חסר מידע:**\n\n"
-            f"{fields_text}\n\n"
-            f"הוסף את הפרטים לפתק ולחץ **הפק מסמך** שוב.")
-
-
-def generate(note, issuer_radio, doc_choice,
-             clarify_vat, clarify_tax_id):
-    """Single-shot, stateless pipeline."""
+# ─────────────────────────────────────────────────────────────────────────────
+# 6. Main pipeline
+# ─────────────────────────────────────────────────────────────────────────────
+def generate(note, issuer_radio, doc_choice, clarify_vat, clarify_tax_id):
+    """
+    Stateless single handler.
+    Returns: (status_md, parse_html, doc_html, recs_html, vat_group_visible, tax_group_visible)
+    """
     note = (note or "").strip()
     if not note:
-        return ("✏️ כתוב פתק הכנסה ולחץ הפק מסמך.",
-                "", "", "", "", False)
+        return "✏️ כתוב פתק הכנסה ולחץ הפק מסמך.", "", "", "", False, False
 
     issuer_status = issuer_radio if issuer_radio in ("פטור", "מורשה") else None
 
-    # ── Check issuer BEFORE running the model (saves 90s on CPU) ─────────────
+    # Issuer check BEFORE model (saves ~90s on CPU)
     if not issuer_status:
-        return ("👤 בחר תחילה: **עוסק פטור** או **עוסק מורשה** למעלה.",
-                "", "", "", "", False)
+        return "👤 בחר תחילה: **עוסק פטור** או **עוסק מורשה** למעלה.", "", "", "", False, False
 
     _lazy_init()
     parse = model_parse(note)
 
     if parse is None or _missing_fields(parse) == ["parse_failed"]:
-        return ("🤔 לא הצלחתי לחלץ פרטים. נסח מחדש — לדוגמה: "
-                "'קיבלתי 500₪ ממשה על ייעוץ'.",
-                "", "", "", "", False)
+        return ("🤔 לא הצלחתי לחלץ פרטים. נסח מחדש — לדוגמה: 'קיבלתי 500₪ ממשה על ייעוץ'.",
+                "", "", "", False, False)
 
-    cvat = clarify_vat   if clarify_vat   not in (None, "לא צוין") else None
+    cvat = clarify_vat    if clarify_vat    not in (None, "לא צוין") else None
     ctax = clarify_tax_id.strip() if clarify_tax_id and clarify_tax_id.strip() else None
 
     issuer, final, questions = decide_fiscal(
         note, issuer_status, parse, doc_choice, cvat, ctax)
 
     if questions:
-        msg = _build_clarification_message(questions, note)
-        return (msg, "", "", "", "", True)
+        show_vat_q   = "vat_basis"    in questions
+        show_taxid_q = "client_tax_id" in questions
+        return (_build_clarification_message(questions),
+                "", "", "", show_vat_q, show_taxid_q)
 
     try:
         completed = core.complete(issuer, final, _STATE["rng"])
     except Exception as e:
-        return (f"⚠️ שגיאה: {e}", "", "", "", "", False)
+        return f"⚠️ שגיאה בעיבוד: {e}", "", "", "", False, False
 
     show_vat = (final["doc_type"] != "receipt")
 
     return ("✅ המסמך הופק בהצלחה.",
-            "",
             render_parse_panel(final, completed, show_vat),
             render_document(completed, show_vat),
             render_recommendations(recommend(note)),
-            False)
+            False, False)
+
 
-# ═════════════════════════════════════════════════════════════════════════════
-# 7. UI — Gradio 6.17 compatible, ChatGPT/Claude style
-# ═════════════════════════════════════════════════════════════════════════════
+# ─────────────────────────────────────────────────────────────────────────────
+# 7. Gradio UI
+# ─────────────────────────────────────────────────────────────────────────────
 CSS = """
 .gradio-container { max-width: 880px !important; margin: 0 auto !important; }
 footer { display: none !important; }
@@ -590,13 +564,12 @@ footer { display: none !important; }
   color: #0f172a !important;
 }
 .qs-btn:hover { background: #e2e8f0 !important; }
-.btn-send { font-weight: 600 !important; }
 """
 
 QUICK_STARTERS = [
-    ("💼 עוסק מורשה + לקוח עסקי",  'חברת מעגל בע"מ שילמה 36,500 ש"ח על פרויקט אתר. העברה בנקאית.'),
+    ("💼 עוסק מורשה + לקוח עסקי",  'חברת מעגל בע"מ שילמה 36,500 כולל מע"מ על פרויקט אתר. העברה בנקאית.'),
     ("🧾 עוסק פטור — שיעור פרטי",   "קיבלתי 220 שקל מדביר על שיעור שחייה פרטי, מזומן"),
-    ("📦 כמה פריטים",                'נועה כהן שילמה 2×350 ש"ח ייעוץ עסקי + 180 ש"ח הכנת מצגת'),
+    ("📦 כמה פריטים",                'נועה כהן שילמה 2×350 ש"ח ייעוץ עסקי + 180 ש"ח הכנת מצגת, לפני מע"מ'),
 ]
 
 with gr.Blocks(title="Text2Receipt", theme=gr.themes.Soft(
@@ -611,17 +584,15 @@ with gr.Blocks(title="Text2Receipt", theme=gr.themes.Soft(
       </div>
     </div>""")
 
-    # ── Who am I (required, no default) ─────────────────────────────────────
     issuer_radio = gr.Radio(
         ["פטור", "מורשה"], value=None,
         label="מי אני (סטטוס מנפיק)",
-        info="עוסק פטור = ללא מע\"מ · עוסק מורשה = עם מע\"מ",
+        info='עוסק פטור = ללא מע"מ · עוסק מורשה = עם מע"מ',
     )
 
     note_input = gr.Textbox(
-        label="פתק / תוכן ההכנסה", lines=2,
+        label="פתק / תוכן ההכנסה", lines=2, rtl=True, autofocus=True,
         placeholder='לדוגמה: קיבלתי 1,200 ש"ח ממשה כהן על ייעוץ עסקי',
-        rtl=True, autofocus=True,
     )
 
     doc_choice = gr.Radio(
@@ -636,18 +607,20 @@ with gr.Blocks(title="Text2Receipt", theme=gr.themes.Soft(
 
     status_output = gr.Markdown("")
 
-    # ── Clarification fields — visible only when needed ──────────────────────
-    with gr.Group(visible=False) as clarify_group:
+    # Clarification fields — appear only when needed
+    with gr.Group(visible=False) as clarify_vat_group:
         gr.HTML('<div dir="rtl" style="font-size:12px;font-weight:600;color:#7c3aed;'
-                'margin-bottom:8px;">📋 פרטים נוספים נדרשים</div>')
+                'margin-bottom:6px;">💰 בסיס מע״מ נדרש</div>')
         clarify_vat = gr.Radio(
             ["כולל מע״מ", "לפני מע״מ", "לא צוין"],
             value="לא צוין", label="הסכום המצוין בפתק",
         )
+
+    with gr.Group(visible=False) as clarify_tax_group:
+        gr.HTML('<div dir="rtl" style="font-size:12px;font-weight:600;color:#7c3aed;'
+                'margin-bottom:6px;">🔖 ח.פ. לקוח נדרש (הקצאה)</div>')
         clarify_tax_id = gr.Textbox(
-            label="ח.פ. / ת.ז. לקוח (נדרש למספר הקצאה)",
-            placeholder="למשל: 514123458",
-            value="",
+            label="ח.פ. / ת.ז. לקוח", placeholder="למשל: 514123458", value="",
         )
 
     gr.HTML('<div dir="rtl" style="font-size:11px;color:#94a3b8;margin:10px 0 6px;'
@@ -661,153 +634,34 @@ with gr.Blocks(title="Text2Receipt", theme=gr.themes.Soft(
     doc_output   = gr.HTML()
     recs_output  = gr.HTML()
 
-    # ── Handlers ─────────────────────────────────────────────────────────────
+    # Handlers
     def _on_generate(note, issuer_r, doc_c, cvat, ctax):
-        status, chtml, phtml, dhtml, rhtml, show_clarify = generate(
+        status, phtml, dhtml, rhtml, show_vat_q, show_taxid_q = generate(
             note, issuer_r, doc_c, cvat, ctax)
         return (status, phtml, dhtml, rhtml,
-                gr.update(visible=show_clarify))
+                gr.update(visible=show_vat_q),
+                gr.update(visible=show_taxid_q))
 
     submit_btn.click(
         fn=_on_generate,
         inputs=[note_input, issuer_radio, doc_choice, clarify_vat, clarify_tax_id],
-        outputs=[status_output, parse_output, doc_output, recs_output, clarify_group],
+        outputs=[status_output, parse_output, doc_output, recs_output,
+                 clarify_vat_group, clarify_tax_group],
     )
 
     def _on_clear():
         return ("", None, "זיהוי אוטומטי", "לא צוין", "", "", "", "", "",
-                gr.update(visible=False))
+                gr.update(visible=False), gr.update(visible=False))
 
     clear_btn.click(
         fn=_on_clear,
         outputs=[note_input, issuer_radio, doc_choice, clarify_vat, clarify_tax_id,
-                 status_output, parse_output, doc_output, recs_output, clarify_group],
-    )
-
-# ═════════════════════════════════════════════════════════════════════════════
-# 8.  Discord bot — runs as a background thread inside the Space  (+5% bonus)
-# ═════════════════════════════════════════════════════════════════════════════
-# The bot shares the SAME model loaded by _lazy_init(), so no extra memory.
-# Token is read from the DISCORD_TOKEN secret (Space → Settings → Secrets).
-# If the secret is absent, the bot is silently skipped and the UI still works.
-import threading, asyncio
-
-def _format_doc_for_discord(completed: dict) -> str:
-    """Render a completed fiscal document as Discord-friendly monospace text."""
-    c       = completed
-    issuer  = c.get("issuer", {})
-    client  = c.get("client", {})
-    lines   = c.get("lines",  [])
-    vat_pct = int(round(c.get("vat_rate", 0) * 100))
-
-    line_rows = "\n".join(
-        f"  • {ln['description']} × {ln['quantity']} → ₪{ln['line_total']:,.2f}"
-        for ln in lines
+                 status_output, parse_output, doc_output, recs_output,
+                 clarify_vat_group, clarify_tax_group],
     )
-    alloc = (f"\n🔖 מספר הקצאה: {c.get('allocation_number')}"
-             if c.get("allocation_required") else "")
-
-    return (
-        f"```\n"
-        f"{'━'*40}\n"
-        f"  {c.get('doc_type_he','מסמך')}   |   מס׳ {c.get('serial_number','')}\n"
-        f"  {c.get('issue_date','')}\n"
-        f"{'━'*40}\n"
-        f"  מנפיק : {issuer.get('name','')}  (ח.פ. {issuer.get('tax_id','')})\n"
-        f"  לקוח   : {client.get('name','')}"
-        f"{'  [עסק]' if client.get('is_business') else ''}\n"
-        f"{'━'*40}\n"
-        f"{line_rows}\n"
-        f"{'━'*40}\n"
-        f"  סכום נטו : ₪{c.get('subtotal',0):>12,.2f}\n"
-        f"  מע\"מ {vat_pct:2d}% : ₪{c.get('vat_amount',0):>12,.2f}\n"
-        f"  סה\"כ     : ₪{c.get('total',0):>12,.2f}\n"
-        f"  תשלום : {core.PAYMENT_HE.get(c.get('payment_method',''),'—')}\n"
-        f"{'━'*40}\n"
-        f"```"
-        f"{alloc}"
-    )
-
-
-def _discord_pipeline(raw_text: str) -> str:
-    """Full pipeline for Discord: raw Hebrew note → formatted document string.
-    Uses safe defaults (no agentic loop in chat — Discord is single-shot)."""
-    _lazy_init()
-    parse = model_parse(raw_text)
-    if parse is None:
-        return "❌ לא הצלחתי לחלץ את הפרטים. נסה לנסח מחדש."
-    final = apply_answers(parse, {})           # apply_answers fills safe defaults
-    try:
-        completed = core.complete(
-            {**DEMO_ISSUER_BASE, "status": "authorized_dealer", "is_company": False},
-            final, _STATE["rng"])
-    except Exception as e:
-        return f"❌ שגיאה בעיבוד: {e}"
-    return _format_doc_for_discord(completed)
-
-
-def _start_discord_bot():
-    """Launch the Discord bot in a background thread if DISCORD_TOKEN is set."""
-    token = os.environ.get("DISCORD_TOKEN", "").strip()
-    if not token:
-        print("ℹ DISCORD_TOKEN not set — Discord bot skipped (UI still runs).")
-        return
-
-    try:
-        import discord
-        from discord.ext import commands
-    except ImportError:
-        print("⚠ discord.py not installed — add 'discord.py' to requirements.txt")
-        return
-
-    intents = discord.Intents.default()
-    intents.message_content = True
-    bot = commands.Bot(command_prefix="!", intents=intents)
-
-    @bot.event
-    async def on_ready():
-        print(f"✅ Discord bot online: {bot.user}")
-
-    @bot.command(name="receipt")
-    async def receipt_cmd(ctx, *, note: str):
-        await ctx.message.add_reaction("⏳")
-        try:
-            loop   = asyncio.get_event_loop()
-            result = await loop.run_in_executor(None, _discord_pipeline, note)
-            await ctx.reply(result)
-        except Exception as e:
-            await ctx.reply(f"❌ שגיאה: {e}")
-        finally:
-            try:
-                await ctx.message.remove_reaction("⏳", bot.user)
-                await ctx.message.add_reaction("✅")
-            except Exception:
-                pass
-
-    @bot.command(name="help_t2r")
-    async def help_cmd(ctx):
-        await ctx.reply(
-            "**Text2Receipt Bot** 🧾\n"
-            "המר הערת הכנסה עברית למסמך פיסקלי ישראלי.\n\n"
-            "שימוש: `!receipt <הערה בעברית>`\n"
-            "דוגמה: `!receipt קיבלתי 500 שקל ממשה על ייעוץ`"
-        )
-
-    def _run():
-        try:
-            asyncio.run(bot.start(token))
-        except Exception as e:
-            print(f"ℹ Discord bot could not connect "
-                  f"(expected on HF Spaces — outbound to discord.com is blocked): "
-                  f"{type(e).__name__}. UI is unaffected.", flush=True)
-
-    threading.Thread(target=_run, daemon=True).start()
-    print("🤖 Discord bot thread started.")
-
-# ═════════════════════════════════════════════════════════════════════════════
-# 9.  Launch
-# ═════════════════════════════════════════════════════════════════════════════
-_start_discord_bot()
 
+# ─────────────────────────────────────────────────────────────────────────────
+# 8. Launch
+# ─────────────────────────────────────────────────────────────────────────────
 if __name__ == "__main__":
     demo.launch(css=CSS)