Upload 7 files

app.py

@@ -10,7 +10,7 @@ import gradio as gr
 import yaml
 from openai import OpenAI
 
-from schemas import FinancialExtract, ExtractedPeriod, MultipleSuggestion
+from schemas import FinancialExtract, ExtractedPeriod, MultipleSuggestion, MarketOutlook
 from finance_core import (
     compute_ratios,
     credit_decision,
@@ -23,6 +23,7 @@ from llm_extract import (
     upload_file_to_openai,
     extract_financials_from_files,
     suggest_multiples_with_llm,
+    suggest_market_outlook_with_llm,
 )
 
 HF_SPACE = os.environ.get("HF_SPACE_NAME", "hf-credit-loan-investment-app")
@@ -107,7 +108,7 @@ def _read_file_input(f):
         pass
     raise ValueError(f"Unsupported file input type: {type(f)}")
 
-# --- 単位検出＆換算ヘルパー（PDF本文を走査して「単位：百万円」等を検出） ---
+# --- 単位検出＆換算（PDFの「単位：百万円」等を検出） ---
 def _concat_pdf_text(paths: List[str], max_chars: int = 180_000) -> str:
     try:
         from pypdf import PdfReader
@@ -132,19 +133,13 @@ def _concat_pdf_text(paths: List[str], max_chars: int = 180_000) -> str:
     return "\n\n".join(out)[:max_chars]
 
 def detect_unit_multiplier_from_paths(paths: List[str]) -> Tuple[float, str]:
-    """
-    PDF本文から単位を推定して (乗数, ラベル) を返す。
-    例: ('百万円'→1_000_000, '千円'→1_000, '万円'→10_000, '円'→1,
-         'millions'→1_000_000, 'thousands'→1_000)
-    見つからなければ (1, '不明')
-    """
     text = _concat_pdf_text(paths)
     if not text:
         return 1.0, "不明"
 
     lower = text.lower()
 
-    # 日本語パターン（優先度：百万円→千円→万円→円）
+    # 日本語パターン
     if re.search(r"単位[：:]\s*百万円", text) or re.search(r"（百万円）", text):
         return 1_000_000.0, "百万円"
     if re.search(r"単位[：:]\s*千円", text) or re.search(r"（千円）", text):
@@ -160,7 +155,6 @@ def detect_unit_multiplier_from_paths(paths: List[str]) -> Tuple[float, str]:
     if re.search(r"in\s+thousands\s+of\s+(yen|jpy|usd|dollars?)", lower) or re.search(r"\b(jpy|¥|\$|usd)\s*\(\s*thousands?\s*\)", lower):
         return 1_000.0, "thousands"
 
-    # コンテキストで単独出現
     if re.search(r"百万円", text):
         return 1_000_000.0, "百万円"
 
@@ -174,7 +168,6 @@ _NUM_FIELDS = [
 ]
 
 def scale_extract_inplace(extract: FinancialExtract, multiplier: float) -> None:
-    """抽出済みオブジェクトの数値を指定乗数でインプレース換算する（Noneは無視）。"""
     if not multiplier or multiplier == 1:
         return
     for period in extract.periods:
@@ -186,12 +179,102 @@ def scale_extract_inplace(extract: FinancialExtract, multiplier: float) -> None:
                 except Exception:
                     pass
 
+# --- サマリ生成（見やすい要約） ---
+def build_human_summary(extract: FinancialExtract, ratios: Dict[str, Any], decisions: Dict[str, Any],
+                        unit_info: Dict[str, Any], market: Optional[MarketOutlook]) -> str:
+    pieces = []
+    pieces.append(f"### 企業名\n{extract.company_name or '（不明）'}")
+    if extract.industry:
+        pieces.append(f"### 業種（推定/指定）\n{extract.industry}")
+    if extract.currency:
+        pieces.append(f"### 通貨\n{extract.currency}")
+    if extract.fiscal_year_end:
+        pieces.append(f"### 決算期末\n{extract.fiscal_year_end}")
+
+    # 評価サマリ（トップに集約）
+    badge = []
+    if "credit" in decisions:
+        badge.append(f"与信: **{decisions['credit']['rating']}**")
+    if "investment" in decisions:
+        badge.append(f"投資魅力度: **{decisions['investment']['attractiveness']}/5**")
+    if market:
+        badge.append(f"市場期待: **{market.expectation_label}**（{market.expected_market_cagr:.1f}%）")
+    if ratios.get("revenue_growth_pct"):
+        badge.append(f"売上成長率: **{ratios['revenue_growth_pct']}**")
+    pieces.append("### 評価サマリ\n" + " / ".join(badge))
+
+    # 単位
+    pieces.append("### 単位（検出結果）")
+    pieces.append(
+        f"- ソース表記: {unit_info['source_label']} / 乗数: x{unit_info['multiplier']:,}"
+        + ("（数値は換算済み）" if unit_info["multiplier"] != 1 else "")
+    )
+
+    # 主要指標
+    pieces.append("### 指標（主要）")
+    pieces.append(
+        f"- 売上高: {ratios.get('revenue')}\n"
+        f"- 営業利益（EBIT）: {ratios.get('ebit')}\n"
+        f"- EBITDA: {ratios.get('ebitda')}\n"
+        f"- 当期純利益: {ratios.get('net_income')}\n"
+        f"- 流動比率: {ratios.get('current_ratio')}\n"
+        f"- 当座比率: {ratios.get('quick_ratio')}\n"
+        f"- D/Eレシオ: {ratios.get('debt_to_equity')}\n"
+        f"- インタレストカバレッジ: {ratios.get('interest_coverage')}\n"
+        f"- 売上成長率: {ratios.get('revenue_growth_pct')}"
+    )
+
+    # 与信
+    if "credit" in decisions:
+        c = decisions["credit"]
+        pieces.append("### 与信判断（提案）")
+        pieces.append(
+            f"- 与信ランク: **{c['rating']}**（スコア {c['risk_score']}/100）\n"
+            f"- 取引サイト: **{c['site_days']}日**\n"
+            f"- 取引可能上限: **{c['transaction_limit_display']}**\n"
+            f"- 見直しタイミング: **{c['review_cycle']}**"
+        )
+
+    # 融資
+    if "loan" in decisions:
+        l = decisions["loan"]
+        pieces.append("### 融資判断（提案）")
+        pieces.append(
+            f"- 融資上限額（概算）: **{l['max_principal_display']}**\n"
+            f"- 期間案: **{l['term_years']}年**\n"
+            f"- 参考金利: **{l['interest_rate_pct']}%**\n"
+            f"- 目標DSCR: **{l['target_dscr']}**"
+        )
+
+    # 投資＋市場期待
+    if "investment" in decisions:
+        inv = decisions["investment"]
+        pieces.append("### 投資判断（提案）")
+        pieces.append(
+            f"- 推定企業価値（EV）: **{inv['ev_display']}**\n"
+            f"- 推定時価総額: **{inv['market_cap_display']}**\n"
+            f"- 想定投資レンジ: **{inv['recommended_check_size_display']}**\n"
+            f"- 魅力度: **{inv['attractiveness']}/5**（成長性: {inv['growth_label']}）"
+        )
+        if inv.get("market_factor"):
+            pieces.append(f"- 市場期待補正: x{inv['market_factor']:.2f}（投資レンジへ反映）")
+    if market:
+        pieces.append("### 市場規模拡大の期待（LLM評価）")
+        pieces.append(
+            f"- 期待度: **{market.expectation_label}**（スコア {market.expectation_score}/5）\n"
+            f"- 想定市場CAGR（3-5年）: **{market.expected_market_cagr:.1f}%**\n"
+            f"- 根拠要約: {market.rationale or '—'}"
+        )
+
+    return "\n\n".join(pieces)
+
 
 def analyze(
     files: List,
     company_name: str,
     industry_hint: str,
     currency_hint: str,
+    market_notes: str,
     base_rate: float,
     want_credit: bool,
     want_loan: bool,
@@ -207,7 +290,7 @@ def analyze(
     if not files or len(files) == 0:
         raise gr.Error("決算書ファイル（PDF/画像）を1つ以上アップロードしてください。")
 
-    # 1) Upload files to OpenAI and extract structured financials via vision
+    # 1) Upload to OpenAI
     try:
         file_ids = []
         for f in files:
@@ -216,14 +299,14 @@ def analyze(
     except Exception as e:
         raise gr.Error(f"ファイルのアップロードに失敗しました: {e}")
 
-    # Local paths for text & unit fallback
+    # Local paths for text/unit fallback
     local_paths = []
     for f in files:
         if isinstance(f, (str, bytes)) or hasattr(f, "__fspath__"):
             local_paths.append(os.fspath(f))
 
+    # 2) Vision抽出（失敗時はテキストフォールバック）
     try:
-        # Prefer version that accepts local_paths; fallback if not supported
         try:
             extract = extract_financials_from_files(
                 client=client,
@@ -246,13 +329,13 @@ def analyze(
     except Exception as e:
         raise gr.Error(f"LLM抽出に失敗しました: {e}")
 
-    # allow override company name / industry if provided
+    # allow overrides
     if company_name:
         extract.company_name = company_name
     if industry_hint:
         extract.industry = industry_hint
 
-    # --- 単位検出＆換算（円/ドル等の素単位に正規化）---
+    # 3) 単位検出→換算（素単位化）
     unit_info = {"source_label": "不明", "multiplier": 1}
     try:
         if local_paths:
@@ -264,15 +347,17 @@ def analyze(
         if debug:
             print(f"[unit-detect] warning: {e}")
 
-    # 2) Compute derived ratios and risk score
+    # 4) 指標計算
     ratios = compute_ratios(extract)
 
-    # 3) Decisions
+    # 5) 意思決定
     decisions: Dict[str, Any] = {}
     if want_credit:
         decisions["credit"] = credit_decision(extract, ratios, POLICIES)
     if want_loan:
         decisions["loan"] = loan_decision(extract, ratios, base_rate or BASE_RATE, POLICIES)
+
+    market_outlook: Optional[MarketOutlook] = None
     if want_invest:
         multiples: Optional[MultipleSuggestion] = None
         try:
@@ -285,13 +370,27 @@ def analyze(
             )
         except Exception:
             multiples = None
-        decisions["investment"] = investment_decision(extract, ratios, POLICIES, multiples)
+        try:
+            market_outlook = suggest_market_outlook_with_llm(
+                client=client,
+                text_model=TEXT_MODEL,
+                industry=extract.industry or industry_hint or "",
+                market_notes=market_notes or "",
+                region="JP",
+                debug=debug,
+            )
+        except Exception:
+            market_outlook = None
 
-    # 4) Build a combined report (dict) and return displays
-    report = build_report_dict(extract, ratios, decisions, unit_info=unit_info)
+        decisions["investment"] = investment_decision(
+            extract, ratios, POLICIES, multiples, market_outlook=market_outlook
+        )
+
+    # 6) レポート構築
+    report = build_report_dict(extract, ratios, decisions, unit_info=unit_info, market_outlook=market_outlook)
     report_json = json.dumps(report, ensure_ascii=False, indent=2)
 
-    # Save a downloadable JSON (ensure directory exists; Gradio v5 は /tmp を推奨)
+    # 保存（Gradio v5は /tmp が安全）
     ts = datetime.utcnow().strftime("%Y%m%d-%H%M%S")
     data_dir = os.environ.get("HF_DATA_DIR", "/tmp")
     os.makedirs(data_dir, exist_ok=True)
@@ -299,93 +398,52 @@ def analyze(
     with open(out_path, "w", encoding="utf-8") as f:
         f.write(report_json)
 
-    # Pretty sections for Gradio
-    summary_md = []
-    summary_md.append(f"### 企業名\n{extract.company_name or '（不明）'}")
-    if extract.industry:
-        summary_md.append(f"### 業種（推定/指定）\n{extract.industry}")
-    if extract.currency:
-        summary_md.append(f"### 通貨\n{extract.currency}")
-    if extract.fiscal_year_end:
-        summary_md.append(f"### 決算期末\n{extract.fiscal_year_end}")
-
-    summary_md.append("### 単位（検出結果）")
-    summary_md.append(f"- ソース表記: {unit_info['source_label']} / 乗数: x{unit_info['multiplier']:,}" + ("（数値は換算済み）" if unit_info["multiplier"] != 1 else ""))
-
-    summary_md.append("### 指標（主要）")
-    summary_md.append(
-        f"- 売上高: {ratios.get('revenue')}\n"
-        f"- 営業利益（EBIT）: {ratios.get('ebit')}\n"
-        f"- EBITDA: {ratios.get('ebitda')}\n"
-        f"- 当期純利益: {ratios.get('net_income')}\n"
-        f"- 流動比率: {ratios.get('current_ratio')}\n"
-        f"- 当座比率: {ratios.get('quick_ratio')}\n"
-        f"- D/Eレシオ: {ratios.get('debt_to_equity')}\n"
-        f"- インタレストカバレッジ: {ratios.get('interest_coverage')}\n"
-        f"- 売上成長率: {ratios.get('revenue_growth_pct')}"
-    )
-
-    if "credit" in decisions:
-        c = decisions["credit"]
-        summary_md.append("### 与信判断（提案）")
-        summary_md.append(
-            f"- 与信ランク: **{c['rating']}**（スコア {c['risk_score']}/100）\n"
-            f"- 取引サイト: **{c['site_days']}日**\n"
-            f"- 取引可能上限: **{c['transaction_limit_display']}**\n"
-            f"- 見直しタイミング: **{c['review_cycle']}**"
-        )
-
-    if "loan" in decisions:
-        l = decisions["loan"]
-        summary_md.append("### 融資判断（提案）")
-        summary_md.append(
-            f"- 融資上限額（概算）: **{l['max_principal_display']}**\n"
-            f"- 期間案: **{l['term_years']}年**\n"
-            f"- 参考金利: **{l['interest_rate_pct']}%**\n"
-            f"- 目標DSCR: **{l['target_dscr']}**"
-        )
-
-    if "investment" in decisions:
-        inv = decisions["investment"]
-        summary_md.append("### 投資判断（提案）")
-        summary_md.append(
-            f"- 推定企業価値（EV）: **{inv['ev_display']}**\n"
-            f"- 推定時価総額: **{inv['market_cap_display']}**\n"
-            f"- 想定投資レンジ: **{inv['recommended_check_size_display']}**\n"
-            f"- 魅力度: **{inv['attractiveness']}/5**（成長性: {inv['growth_label']}）"
-        )
+    # 7) UI 向けサマリ
+    summary_md = build_human_summary(extract, ratios, decisions, unit_info, market_outlook)
 
-    return "\n\n".join(summary_md), json.loads(report_json), out_path
+    return summary_md, json.loads(report_json), out_path
 
 
 def build_ui():
-    with gr.Blocks(theme=gr.themes.Soft(), css="footer {visibility: hidden}") as demo:
-        gr.Markdown("# 決算書→与信・融資・投資判断（HF+OpenAI）")
+    with gr.Blocks(theme=gr.themes.Soft(), css="""
+    footer {visibility: hidden}
+    .badge {display:inline-block;padding:4px 8px;border-radius:9999px;background:#eef; margin-right:6px}
+    """) as demo:
+        gr.Markdown("## 決算書→与信・融資・投資判断（HF + OpenAI）")
         with gr.Row():
-            with gr.Column():
+            with gr.Column(scale=1):
                 files = gr.File(
                     label="決算書ファイル（PDF/JPG/PNG, 複数可）",
                     file_types=[".pdf", ".png", ".jpg", ".jpeg"],
                     file_count="multiple",
-                    type="filepath",  # return string paths (stable across Gradio versions)
+                    type="filepath",
                 )
-                company_name = gr.Textbox(label="会社名（任意）")
-                industry_hint = gr.Textbox(label="業種（任意）")
+                company_name = gr.Textbox(label="会社名（任意）", placeholder="例：株式会社サンプル")
+                industry_hint = gr.Textbox(label="業種（任意）", placeholder="例：ソフトウェア、食品、物流 等")
                 currency_hint = gr.Textbox(label="通貨（任意, 例: JPY, USD）")
+                market_notes = gr.Textbox(
+                    label="市場拡大の期待（自由記述）",
+                    placeholder="例：生成AI関連の新製品投入、医療×SaaS拡販、EV向け部材、海外展開 等"
+                )
                 base_rate = gr.Number(label="ベース金利（%/年）", value=BASE_RATE)
-                want_credit = gr.Checkbox(label="与信判断", value=True)
-                want_loan = gr.Checkbox(label="融資判断", value=True)
-                want_invest = gr.Checkbox(label="投資判断", value=True)
-                debug = gr.Checkbox(label="デバッグモード（LLM抽出JSONを厳密化）", value=False)
+                with gr.Row():
+                    want_credit = gr.Checkbox(label="与信判断", value=True)
+                    want_loan = gr.Checkbox(label="融資判断", value=True)
+                    want_invest = gr.Checkbox(label="投資判断", value=True)
+                debug = gr.Checkbox(label="デバッグモード（厳格JSON/ログ）", value=False)
                 run_btn = gr.Button("分析する", variant="primary")
-            with gr.Column():
-                summary = gr.Markdown()
-                report = gr.JSON(label="詳細レポート（JSON）")
-                download = gr.File(label="レポート（JSONダウンロード）")
+            with gr.Column(scale=1):
+                with gr.Tabs():
+                    with gr.TabItem("概要（評価サマリ）"):
+                        summary = gr.Markdown()
+                    with gr.TabItem("詳細JSON"):
+                        report = gr.JSON(label="詳細レポート（JSON）")
+                    with gr.TabItem("ダウンロード"):
+                        download = gr.File(label="レポート（JSON）")
 
         run_btn.click(
             analyze,
-            inputs=[files, company_name, industry_hint, currency_hint, base_rate, want_credit, want_loan, want_invest, debug],
+            inputs=[files, company_name, industry_hint, currency_hint, market_notes, base_rate, want_credit, want_loan, want_invest, debug],
             outputs=[summary, report, download],
         )
     return demo
@@ -393,4 +451,4 @@ def build_ui():
 
 if __name__ == "__main__":
     demo = build_ui()
-    demo.launch(allowed_paths=["/tmp", "/mnt/data"])  # /tmp を既定保存先にしつつ、必要なら /mnt/data も許可
+    demo.launch(allowed_paths=["/tmp", "/mnt/data"])

finance_core.py

@@ -2,9 +2,9 @@
 from __future__ import annotations
 from typing import Dict, Any, Optional
 
-from schemas import FinancialExtract, ExtractedPeriod, MultipleSuggestion
+from schemas import FinancialExtract, ExtractedPeriod, MultipleSuggestion, MarketOutlook
 
-def _fmt_currency(x: Optional[float], currency: str) -> str:
+def _fmt_currency(x, currency: str) -> str:
     if x is None:
         return "—"
     try:
@@ -199,8 +199,13 @@ def loan_decision(extract: FinancialExtract, ratios: Dict[str, Any], base_rate_p
     return {"max_principal": max_principal, "max_principal_display": _fmt_currency(max_principal, currency),
             "term_years": term_years, "interest_rate_pct": round(interest_rate_pct, 2), "target_dscr": target_dscr}
 
-def investment_decision(extract: FinancialExtract, ratios: Dict[str, Any], policies: Dict[str, Any],
-                        multiples: Optional[MultipleSuggestion]) -> Dict[str, Any]:
+def investment_decision(
+    extract: FinancialExtract,
+    ratios: Dict[str, Any],
+    policies: Dict[str, Any],
+    multiples: Optional[MultipleSuggestion],
+    market_outlook: Optional[MarketOutlook] = None,
+) -> Dict[str, Any]:
     latest = extract.latest()
     currency = extract.currency or "JPY"
     if not latest:
@@ -236,6 +241,7 @@ def investment_decision(extract: FinancialExtract, ratios: Dict[str, Any], polic
     pct = {"A": 0.15, "B": 0.12, "C": 0.08, "D": 0.04, "E": 0.0}[rating]
     check = market_cap * pct
 
+    # ベース魅力度（内部健全性＋成長性）
     attractiveness = 1
     if rating == "A":
         attractiveness = 5 if glabel == "High" else 4
@@ -246,16 +252,33 @@ def investment_decision(extract: FinancialExtract, ratios: Dict[str, Any], polic
     elif rating == "D":
         attractiveness = 1
 
-    return {"ev": ev, "ev_display": _fmt_currency(ev, currency),
-            "market_cap": market_cap, "market_cap_display": _fmt_currency(market_cap, currency),
-            "recommended_check_size": check, "recommended_check_size_display": _fmt_currency(check, currency),
-            "attractiveness": attractiveness, "growth_label": glabel}
+    market_factor = 1.0
+    if market_outlook:
+        # 期待スコアに応じて投資レンジを±15%調整
+        factor_map = {1: 0.85, 2: 0.93, 3: 1.00, 4: 1.07, 5: 1.15}
+        market_factor = factor_map.get(int(market_outlook.expectation_score), 1.0)
+        # 魅力度にも±1段階の微調整（1〜5にクランプ）
+        if market_outlook.expectation_score >= 4:
+            attractiveness = min(5, attractiveness + 1)
+        elif market_outlook.expectation_score <= 2:
+            attractiveness = max(1, attractiveness - 1)
+
+    check *= market_factor
+
+    return {
+        "ev": ev, "ev_display": _fmt_currency(ev, currency),
+        "market_cap": market_cap, "market_cap_display": _fmt_currency(market_cap, currency),
+        "recommended_check_size": check, "recommended_check_size_display": _fmt_currency(check, currency),
+        "attractiveness": attractiveness, "growth_label": glabel,
+        "market_factor": round(market_factor, 2)
+    }
 
 def build_report_dict(
     extract: FinancialExtract,
     ratios: Dict[str, Any],
     decisions: Dict[str, Any],
-    unit_info: Optional[Dict[str, Any]] = None,   # ← 追加
+    unit_info: Optional[Dict[str, Any]] = None,
+    market_outlook: Optional[MarketOutlook] = None,
 ) -> Dict[str, Any]:
     out = {
         "metadata": {
@@ -271,4 +294,6 @@ def build_report_dict(
     }
     if unit_info:
         out["unit_detection"] = unit_info
+    if market_outlook:
+        out["market_outlook"] = market_outlook.model_dump()
     return out

llm_extract.py

@@ -4,7 +4,7 @@ import os, json
 from typing import List, Optional
 from openai import OpenAI
 from pydantic import ValidationError
-from schemas import FinancialExtract, MultipleSuggestion
+from schemas import FinancialExtract, MultipleSuggestion, MarketOutlook
 
 VISION_MODEL = os.environ.get("OPENAI_VISION_MODEL", "gpt-4o-mini")
 TEXT_MODEL  = os.environ.get("OPENAI_TEXT_MODEL",   "gpt-4o-mini")
@@ -77,7 +77,7 @@ def extract_financials_from_files(
     currency_hint: Optional[str],
     model: str = VISION_MODEL,
     debug: bool = False,
-    local_paths: Optional[List[str]] = None,  # ← フォールバック用
+    local_paths: Optional[List[str]] = None,
 ) -> FinancialExtract:
 
     schema = FinancialExtract.model_json_schema()
@@ -95,7 +95,7 @@ def extract_financials_from_files(
     if currency_hint:
         base_user += f"\nCurrency hint: {currency_hint}"
 
-    # 1) Vision + file_id で試す（response_format 未使用）
+    # 1) Vision + file_id
     try:
         resp = client.responses.create(
             model=model,
@@ -115,7 +115,7 @@ def extract_financials_from_files(
         return FinancialExtract.model_validate(data)
 
     except Exception as e_vision:
-        # 2) テキスト抽出 → TEXT_MODEL で構造化
+        # 2) Fallback: PDFテキスト→TEXTモデルで構造化
         if not local_paths:
             raise RuntimeError(f"Vision抽出に失敗し、かつローカルPDFテキストがありません: {e_vision}")
 
@@ -168,3 +168,46 @@ def suggest_multiples_with_llm(client: OpenAI, text_model: str, industry: str, r
         if debug:
             raise
         return None
+
+def suggest_market_outlook_with_llm(
+    client: OpenAI,
+    text_model: str,
+    industry: str,
+    market_notes: str,
+    region: str = "JP",
+    debug: bool = False,
+) -> Optional[MarketOutlook]:
+    """取扱商品・新製品・展開計画などに基づく、3-5年の市場拡大期待を定性的にスコア化。"""
+    if not (industry or market_notes):
+        return None
+
+    system = "You are a market analyst. Rate market expansion expectations for the product portfolio in the next 3-5 years."
+    user = (
+        f"Industry: {industry or 'Unknown'}\nRegion: {region}\n"
+        f"Notes: {market_notes or 'None'}\n"
+        "Return STRICT JSON with keys:\n"
+        "  expectation_label (one of: Very Low, Low, Medium, High, Very High),\n"
+        "  expectation_score (integer 1-5),\n"
+        "  expected_market_cagr (float, %),\n"
+        "  rationale (<=60 words).\n"
+        "No prose outside JSON."
+    )
+
+    try:
+        resp = client.responses.create(
+            model=text_model,
+            input=[
+                {"role": "system", "content": [{"type": "input_text", "text": system}]},
+                {"role": "user", "content": [{"type": "input_text", "text": user}]},
+            ],
+            max_output_tokens=250,
+        )
+        raw = _safe_output_text(resp)
+        if not raw:
+            return None
+        data = _json_loads_strict(raw)
+        return MarketOutlook.model_validate(data)
+    except Exception:
+        if debug:
+            raise
+        return None

schemas.py

@@ -44,3 +44,9 @@ class FinancialExtract(BaseModel):
 class MultipleSuggestion(BaseModel):
     revenue_multiple: float = 1.5
     ebitda_multiple: float = 8.0
+
+class MarketOutlook(BaseModel):
+    expectation_label: str  # Very Low / Low / Medium / High / Very High
+    expectation_score: int  # 1..5
+    expected_market_cagr: float  # percentage
+    rationale: Optional[str] = None