@hupan56/wlkj 3.3.13 → 3.3.15

package/templates/qoder/scripts/validation/eval/prototype_fidelity_flywheel.py

@@ -1,166 +1,166 @@
-"""prototype_fidelity_flywheel.py — FW9 原型保真度飞轮（vs 真实 fywl-ui 设计系统）。
-
-用户根因例子："原型保真不保真"。本尺子量：生成原型的颜色/组件是不是 fywl-ui 真设计系统的
-(get_design_system 返回的真色板 hsl(214,86%,59%)/真组件/真layout)，而非泛泛 #409eff 蓝或编造组件。
-bare(无设计系统接地→通用色/编造组件) vs +真设计系统接地(用真 hsl 色板/真组件)。
-color保真率 = 原型颜色∈真色板比例；component保真率 = 真组件占比。
-
-跑法: docker exec -i wlinkj-workspace-backend-1 python /app/scripts/validation/eval/prototype_fidelity_flywheel.py
-"""
-import os
-import re
-import sys
-import json
-import colorsys
-import time
-import urllib.request
-import urllib.error
-
-for _p in ("/app", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "..", ".."))):
-    if os.path.isdir(_p) and _p not in sys.path:
-        sys.path.insert(0, _p)
-
-from app.routers import mcp_handlers as H  # noqa: E402
-try:
-    H._init_handlers()
-except Exception:
-    pass
-
-CHAT_URL = os.environ.get("ALI_CHAT_BASE", "").rstrip("/") + "/chat/completions"
-CHAT_KEY = os.environ.get("DASHSCOPE_API_KEY", "")
-MODEL = "qwen-turbo"
-PAGES = ["车辆保险列表页", "异常记录表单页", "合同维护详情页"]
-_NEUTRAL = {(255, 255, 255), (0, 0, 0), (51, 51, 51), (102, 102, 102), (153, 153, 153),
-            (240, 240, 240), (250, 250, 250), (217, 217, 217), (232, 232, 232), (245, 245, 245)}
-
-
-def chat(prompt, max_tokens=1200, timeout=90):
-    body = json.dumps({"model": MODEL, "messages": [{"role": "user", "content": prompt}],
-        "temperature": 0.2, "max_tokens": max_tokens}).encode()
-    for a in range(3):
-        try:
-            req = urllib.request.Request(CHAT_URL, data=body,
-                headers={"Authorization": f"Bearer {CHAT_KEY}", "Content-Type": "application/json"})
-            return json.loads(urllib.request.urlopen(req, timeout=timeout).read())["choices"][0]["message"]["content"]
-        except (urllib.error.URLError, TimeoutError, OSError, KeyError) as e:
-            if a < 2: time.sleep(1.0*(a+1))
-            else: return ""
-
-
-def real_design():
-    """fywl-ui 真设计系统：色板(rgb 集) + 真组件名 + layout。"""
-    r = H.dispatch("get_design_system", {}) or {}
-    proj = next((p for p in r.get("projects", []) if p.get("project") == "fywl-ui"), {})
-    colors = proj.get("colors", {}) or {}
-    palette_rgb = set()
-    for v in colors.values():
-        if isinstance(v, str) and v.startswith("hsl"):
-            m = re.match(r"hsl\(\s*([\d.]+)[,\s]+([\d.]+)%[,\s]+([\d.]+)%", v)
-            if m:
-                h, s, l = float(m.group(1)) / 360.0, float(m.group(2)) / 100.0, float(m.group(3)) / 100.0
-                r_, g_, b_ = colorsys.hls_to_rgb(h, l, s)
-                palette_rgb.add((round(r_ * 255), round(g_ * 255), round(b_ * 255)))
-    comps = set()
-    for d in (proj.get("commonComponents", {}), proj.get("formComponents", {})):
-        if isinstance(d, dict):
-            comps.update(k.lower() for k in d.keys())
-    return palette_rgb, comps, proj
-
-
-def extract_colors(html):
-    """提取 HTML 里的颜色(#hex/hsl/rgb)→ rgb 集合。"""
-    out = set()
-    for m in re.findall(r"#([0-9a-fA-F]{6}|[0-9a-fA-F]{3})\b", html):
-        h = m if len(m) == 6 else m * 2
-        out.add((int(h[0:2], 16), int(h[2:4], 16), int(h[4:6], 16)))
-    for m in re.findall(r"hsl\(\s*([\d.]+)[,\s]+([\d.]+)%[,\s]+([\d.]+)%", html):
-        h, s, l = float(m[0]) / 360.0, float(m[1]) / 100.0, float(m[2]) / 100.0
-        r_, g_, b_ = colorsys.hls_to_rgb(h, l, s)
-        out.add((round(r_ * 255), round(g_ * 255), round(b_ * 255)))
-    for m in re.findall(r"rgb\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)", html):
-        out.add((int(m[0]), int(m[1]), int(m[2])))
-    return out
-
-
-def extract_components(html):
-    """提取 HTML 里的组件标签(el-xxx/van-xxx/a-xxx 等)。"""
-    return set(m.lower() for m in re.findall(r"<(?:el|van|a|n|iview|ant)-([\w-]+)", html))
-
-
-def near(c1, c2, tol=40):
-    return all(abs(a - b) <= tol for a, b in zip(c1, c2))
-
-
-def color_fidelity(html, palette_rgb):
-    cols = extract_colors(html)
-    if not cols:
-        return 0.0, 0
-    real = sum(1 for c in cols if any(near(c, p) for p in palette_rgb) or c in _NEUTRAL)
-    return round(real / len(cols), 3), len(cols)
-
-
-def comp_fidelity(html, real_comps):
-    comps = extract_components(html)
-    if not comps:
-        return 0.0, 0
-    if not real_comps:
-        return 1.0, len(comps)  # 无真组件集则不罚
-    real = sum(1 for c in comps if c in real_comps)
-    return round(real / len(comps), 3), len(comps)
-
-
-def run():
-    palette, comps, proj = real_design()
-    print("  真色板 rgb:", list(palette)[:4], "| 真组件:", list(comps)[:8])
-    rep = {"pages": []}
-    for pg in PAGES:
-        bare = chat(f"画一个 {pg} 的 HTML 原型（单文件，含表格/表单/按钮，内联样式）。直接输出 HTML。")
-        ds = ("fywl-ui 真设计系统：主色 hsl(214,86%,59%) 成功 hsl(100,54%,49%) 警告 hsl(36,100%,64%) "
-              "危险 hsl(0,100%,68%) 圆角 0.25rem 布局 mixed-nav 侧栏 160px；"
-              "组件用 ElementPlus(el-table/el-form/el-button/el-input/el-select/el-pagination)。")
-        # BF4 杠杆：CSS 变量注入（:root 定义真色板 + 强制 var() 用色 → 颜色全是色板色）
-        cssvar = ("<style>:root{--primary:hsl(214,86%,59%);--success:hsl(100,54%,49%);"
-                  "--warning:hsl(36,100%,64%);--danger:hsl(0,100%,68%);--text:#333;--border:#d9d9d9;--bg:#f5f5f5}</style>")
-        scaff = chat(f"画一个 {pg} 的 HTML 原型（单文件，内联样式）。\n"
-                     f"【设计系统接地，必须保真】：{ds}\n"
-                     f"⚠️ 颜色铁律（CSS 变量）：把这段 {cssvar} 放在 <head>，"
-                     f"所有 color/background/border 【必须用 var(--primary)/var(--success)/var(--warning)/var(--danger)/var(--text)/var(--border)/var(--bg)】，"
-                     f"【禁止】写任何裸 hex/hsl 色值（#409eff 等一律不许）；组件用 el-xxx。直接输出 HTML。")
-        bP, bN = color_fidelity(bare, palette)
-        sP, sN = color_fidelity(scaff, palette)
-        bC, _ = comp_fidelity(bare, comps)
-        sC, _ = comp_fidelity(scaff, comps)
-        rep["pages"].append({"page": pg, "bare_color": bP, "scaffold_color": sP,
-                             "bare_comp": bC, "scaffold_comp": sC})
-        print("  %s: 颜色保真 bare=%.0f%% scaff=%.0f%% | 组件保真 bare=%.0f%% scaff=%.0f%%" % (
-            pg, bP*100, sP*100, bC*100, sC*100))
-    n = len(PAGES) or 1
-    rep["agg"] = {
-        "bare_color": round(sum(p["bare_color"] for p in rep["pages"])/n, 3),
-        "scaffold_color": round(sum(p["scaffold_color"] for p in rep["pages"])/n, 3),
-        "bare_comp": round(sum(p["bare_comp"] for p in rep["pages"])/n, 3),
-        "scaffold_comp": round(sum(p["scaffold_comp"] for p in rep["pages"])/n, 3),
-    }
-    return rep
-
-
-def main():
-    import argparse
-    ap = argparse.ArgumentParser()
-    ap.add_argument("--json", action="store_true")
-    args = ap.parse_args()
-    r = run()
-    a = r["agg"]
-    if args.json:
-        print(json.dumps(r, ensure_ascii=False, indent=2))
-        return
-    print("\n## FW9 原型保真度飞轮（vs 真实 fywl-ui 设计系统）")
-    print("| 维度 | bare(无设计接地) | +真设计系统接地 |")
-    print("|---|---|---|")
-    print("| 颜色保真(色∈真fywl-ui色板) | %.0f%% | %.0f%% |" % (a["bare_color"]*100, a["scaffold_color"]*100))
-    print("| 组件保真(用真el-组件) | %.0f%% | %.0f%% |" % (a["bare_comp"]*100, a["scaffold_comp"]*100))
-    print("\n保真=原型是 fywl-ui 真样子(真色板/真组件/真layout),非通用蓝/编造组件。bare 通用→低保真;scaffold 接真→高保真。")
-
-
-if __name__ == "__main__":
-    main()
+"""prototype_fidelity_flywheel.py — FW9 原型保真度飞轮（vs 真实 fywl-ui 设计系统）。
+
+用户根因例子："原型保真不保真"。本尺子量：生成原型的颜色/组件是不是 fywl-ui 真设计系统的
+(get_design_system 返回的真色板 hsl(214,86%,59%)/真组件/真layout)，而非泛泛 #409eff 蓝或编造组件。
+bare(无设计系统接地→通用色/编造组件) vs +真设计系统接地(用真 hsl 色板/真组件)。
+color保真率 = 原型颜色∈真色板比例；component保真率 = 真组件占比。
+
+跑法: docker exec -i wlinkj-workspace-backend-1 python /app/scripts/validation/eval/prototype_fidelity_flywheel.py
+"""
+import os
+import re
+import sys
+import json
+import colorsys
+import time
+import urllib.request
+import urllib.error
+
+for _p in ("/app", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "..", ".."))):
+    if os.path.isdir(_p) and _p not in sys.path:
+        sys.path.insert(0, _p)
+
+from app.routers import mcp_handlers as H  # noqa: E402
+try:
+    H._init_handlers()
+except Exception:
+    pass
+
+CHAT_URL = os.environ.get("ALI_CHAT_BASE", "").rstrip("/") + "/chat/completions"
+CHAT_KEY = os.environ.get("DASHSCOPE_API_KEY", "")
+MODEL = "qwen-turbo"
+PAGES = ["车辆保险列表页", "异常记录表单页", "合同维护详情页"]
+_NEUTRAL = {(255, 255, 255), (0, 0, 0), (51, 51, 51), (102, 102, 102), (153, 153, 153),
+            (240, 240, 240), (250, 250, 250), (217, 217, 217), (232, 232, 232), (245, 245, 245)}
+
+
+def chat(prompt, max_tokens=1200, timeout=90):
+    body = json.dumps({"model": MODEL, "messages": [{"role": "user", "content": prompt}],
+        "temperature": 0.2, "max_tokens": max_tokens}).encode()
+    for a in range(3):
+        try:
+            req = urllib.request.Request(CHAT_URL, data=body,
+                headers={"Authorization": f"Bearer {CHAT_KEY}", "Content-Type": "application/json"})
+            return json.loads(urllib.request.urlopen(req, timeout=timeout).read())["choices"][0]["message"]["content"]
+        except (urllib.error.URLError, TimeoutError, OSError, KeyError) as e:
+            if a < 2: time.sleep(1.0*(a+1))
+            else: return ""
+
+
+def real_design():
+    """fywl-ui 真设计系统：色板(rgb 集) + 真组件名 + layout。"""
+    r = H.dispatch("get_design_system", {}) or {}
+    proj = next((p for p in r.get("projects", []) if p.get("project") == "fywl-ui"), {})
+    colors = proj.get("colors", {}) or {}
+    palette_rgb = set()
+    for v in colors.values():
+        if isinstance(v, str) and v.startswith("hsl"):
+            m = re.match(r"hsl\(\s*([\d.]+)[,\s]+([\d.]+)%[,\s]+([\d.]+)%", v)
+            if m:
+                h, s, l = float(m.group(1)) / 360.0, float(m.group(2)) / 100.0, float(m.group(3)) / 100.0
+                r_, g_, b_ = colorsys.hls_to_rgb(h, l, s)
+                palette_rgb.add((round(r_ * 255), round(g_ * 255), round(b_ * 255)))
+    comps = set()
+    for d in (proj.get("commonComponents", {}), proj.get("formComponents", {})):
+        if isinstance(d, dict):
+            comps.update(k.lower() for k in d.keys())
+    return palette_rgb, comps, proj
+
+
+def extract_colors(html):
+    """提取 HTML 里的颜色(#hex/hsl/rgb)→ rgb 集合。"""
+    out = set()
+    for m in re.findall(r"#([0-9a-fA-F]{6}|[0-9a-fA-F]{3})\b", html):
+        h = m if len(m) == 6 else m * 2
+        out.add((int(h[0:2], 16), int(h[2:4], 16), int(h[4:6], 16)))
+    for m in re.findall(r"hsl\(\s*([\d.]+)[,\s]+([\d.]+)%[,\s]+([\d.]+)%", html):
+        h, s, l = float(m[0]) / 360.0, float(m[1]) / 100.0, float(m[2]) / 100.0
+        r_, g_, b_ = colorsys.hls_to_rgb(h, l, s)
+        out.add((round(r_ * 255), round(g_ * 255), round(b_ * 255)))
+    for m in re.findall(r"rgb\(\s*(\d+)\s*,\s*(\d+)\s*,\s*(\d+)", html):
+        out.add((int(m[0]), int(m[1]), int(m[2])))
+    return out
+
+
+def extract_components(html):
+    """提取 HTML 里的组件标签(el-xxx/van-xxx/a-xxx 等)。"""
+    return set(m.lower() for m in re.findall(r"<(?:el|van|a|n|iview|ant)-([\w-]+)", html))
+
+
+def near(c1, c2, tol=40):
+    return all(abs(a - b) <= tol for a, b in zip(c1, c2))
+
+
+def color_fidelity(html, palette_rgb):
+    cols = extract_colors(html)
+    if not cols:
+        return 0.0, 0
+    real = sum(1 for c in cols if any(near(c, p) for p in palette_rgb) or c in _NEUTRAL)
+    return round(real / len(cols), 3), len(cols)
+
+
+def comp_fidelity(html, real_comps):
+    comps = extract_components(html)
+    if not comps:
+        return 0.0, 0
+    if not real_comps:
+        return 1.0, len(comps)  # 无真组件集则不罚
+    real = sum(1 for c in comps if c in real_comps)
+    return round(real / len(comps), 3), len(comps)
+
+
+def run():
+    palette, comps, proj = real_design()
+    print("  真色板 rgb:", list(palette)[:4], "| 真组件:", list(comps)[:8])
+    rep = {"pages": []}
+    for pg in PAGES:
+        bare = chat(f"画一个 {pg} 的 HTML 原型（单文件，含表格/表单/按钮，内联样式）。直接输出 HTML。")
+        ds = ("fywl-ui 真设计系统：主色 hsl(214,86%,59%) 成功 hsl(100,54%,49%) 警告 hsl(36,100%,64%) "
+              "危险 hsl(0,100%,68%) 圆角 0.25rem 布局 mixed-nav 侧栏 160px；"
+              "组件用 ElementPlus(el-table/el-form/el-button/el-input/el-select/el-pagination)。")
+        # BF4 杠杆：CSS 变量注入（:root 定义真色板 + 强制 var() 用色 → 颜色全是色板色）
+        cssvar = ("<style>:root{--primary:hsl(214,86%,59%);--success:hsl(100,54%,49%);"
+                  "--warning:hsl(36,100%,64%);--danger:hsl(0,100%,68%);--text:#333;--border:#d9d9d9;--bg:#f5f5f5}</style>")
+        scaff = chat(f"画一个 {pg} 的 HTML 原型（单文件，内联样式）。\n"
+                     f"【设计系统接地，必须保真】：{ds}\n"
+                     f"⚠️ 颜色铁律（CSS 变量）：把这段 {cssvar} 放在 <head>，"
+                     f"所有 color/background/border 【必须用 var(--primary)/var(--success)/var(--warning)/var(--danger)/var(--text)/var(--border)/var(--bg)】，"
+                     f"【禁止】写任何裸 hex/hsl 色值（#409eff 等一律不许）；组件用 el-xxx。直接输出 HTML。")
+        bP, bN = color_fidelity(bare, palette)
+        sP, sN = color_fidelity(scaff, palette)
+        bC, _ = comp_fidelity(bare, comps)
+        sC, _ = comp_fidelity(scaff, comps)
+        rep["pages"].append({"page": pg, "bare_color": bP, "scaffold_color": sP,
+                             "bare_comp": bC, "scaffold_comp": sC})
+        print("  %s: 颜色保真 bare=%.0f%% scaff=%.0f%% | 组件保真 bare=%.0f%% scaff=%.0f%%" % (
+            pg, bP*100, sP*100, bC*100, sC*100))
+    n = len(PAGES) or 1
+    rep["agg"] = {
+        "bare_color": round(sum(p["bare_color"] for p in rep["pages"])/n, 3),
+        "scaffold_color": round(sum(p["scaffold_color"] for p in rep["pages"])/n, 3),
+        "bare_comp": round(sum(p["bare_comp"] for p in rep["pages"])/n, 3),
+        "scaffold_comp": round(sum(p["scaffold_comp"] for p in rep["pages"])/n, 3),
+    }
+    return rep
+
+
+def main():
+    import argparse
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--json", action="store_true")
+    args = ap.parse_args()
+    r = run()
+    a = r["agg"]
+    if args.json:
+        print(json.dumps(r, ensure_ascii=False, indent=2))
+        return
+    print("\n## FW9 原型保真度飞轮（vs 真实 fywl-ui 设计系统）")
+    print("| 维度 | bare(无设计接地) | +真设计系统接地 |")
+    print("|---|---|---|")
+    print("| 颜色保真(色∈真fywl-ui色板) | %.0f%% | %.0f%% |" % (a["bare_color"]*100, a["scaffold_color"]*100))
+    print("| 组件保真(用真el-组件) | %.0f%% | %.0f%% |" % (a["bare_comp"]*100, a["scaffold_comp"]*100))
+    print("\n保真=原型是 fywl-ui 真样子(真色板/真组件/真layout),非通用蓝/编造组件。bare 通用→低保真;scaffold 接真→高保真。")
+
+
+if __name__ == "__main__":
+    main()

package/templates/qoder/scripts/validation/eval/recall_flywheel.py

@@ -1,148 +1,148 @@
-"""recall_flywheel.py — FW1 检索类召回飞轮（非循环 held-out gold）。
-
-T1.0 的召回 R~80% 用 ILIKE-gold 是循环的（改非ILIKE检索提不了R）。本飞轮用**非循环 gold**：
-held-out = 图邻居中【不含主题关键词】的实体（只有 graph/semantic 检索能找到，ILIKE 找不到）。
-量各检索策略对此 gold 的召回 → 改检索（加图扩展→语义）→ 重量 → 直到 ≥0.85。
-纯检索质量（无 LLM），快，可反复转飞轮。
-
-跑法: docker exec -i wlinkj-workspace-backend-1 python /app/scripts/validation/eval/recall_flywheel.py
-"""
-import os
-import sys
-import json
-
-for _p in ("/app", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "..", ".."))):
-    if os.path.isdir(_p) and _p not in sys.path:
-        sys.path.insert(0, _p)
-
-from sqlalchemy import text, bindparam  # noqa: E402
-from app.db import SessionLocal  # noqa: E402
-
-THEMES = ["考勤", "排班", "加班", "车辆", "报警", "保险", "能耗", "预算", "导出", "审批"]
-TARGET = 0.85
-
-
-def _q(s, sql, params=None):
-    try:
-        return s.execute(text(sql), params or {}).fetchall()
-    except Exception:
-        try:
-            s.rollback()
-        except Exception:
-            pass
-        return []
-
-
-def ilike_ids(s, theme, limit):
-    """name/cn 关键词命中（T1.0 现状检索口径）。"""
-    return [r[0] for r in _q(s,
-        "SELECT id FROM entities WHERE (cn ILIKE :p OR canonical ILIKE :p) LIMIT :n",
-        {"p": f"%{theme}%", "n": limit})]
-
-
-def heldout_gold(s, theme, max_n=25):
-    """非循环 gold：cn_desc(描述)提及主题 但 name/cn 不含 = 只有描述/语义/图能找到，
-    关键词搜 name/cn 找不到。cn_desc 与 name/cn 是独立信号 → 公平。"""
-    rows = _q(s,
-        "SELECT id FROM entities WHERE cn_desc ILIKE :p "
-        "AND NOT (canonical ILIKE :p OR cn ILIKE :p) LIMIT :n",
-        {"p": f"%{theme}%", "n": max_n * 3})
-    return [r[0] for r in rows][:max_n]
-
-
-def strat_ilike(s, theme, limit=80):
-    """基线：name/cn 关键词（T1.0 现状）。"""
-    return set(ilike_ids(s, theme, limit))
-
-
-def strat_ilike_desc(s, theme, limit=80):
-    """杠杆1：name/cn + cn_desc（多搜描述列，廉价高杠杆）。"""
-    return set(r[0] for r in _q(s,
-        "SELECT id FROM entities WHERE (canonical ILIKE :p OR cn ILIKE :p OR cn_desc ILIKE :p) LIMIT :n",
-        {"p": f"%{theme}%", "n": limit}))
-
-
-def strat_ilike_graph(s, theme, seed_limit=25):
-    """杠杆2：name/cn 种子 + 1跳图扩展。"""
-    seeds = ilike_ids(s, theme, seed_limit)
-    res = set(seeds)
-    if seeds:
-        stmt = text(
-            "SELECT DISTINCT e.to_id FROM edges e WHERE e.from_id IN :ids"
-        ).bindparams(bindparam("ids", expanding=True))
-        try:
-            for r in s.execute(stmt, {"ids": seeds}).fetchall():
-                if r[0]:
-                    res.add(r[0])
-        except Exception:
-            try:
-                s.rollback()
-            except Exception:
-                pass
-    return res
-
-
-def recall(strategy_ids, gold):
-    if not gold:
-        return None
-    return len(set(gold) & strategy_ids) / len(gold)
-
-
-def run(verbose=True):
-    s = SessionLocal()
-    per_theme = []
-    accA = accB = accC = 0.0
-    cnt = 0
-    for th in THEMES:
-        gold = heldout_gold(s, th, 25)
-        if not gold:
-            continue
-        A = strat_ilike(s, th, 80)
-        B = strat_ilike_desc(s, th, 80)
-        C = strat_ilike_graph(s, th, 25)
-        rA, rB, rC = recall(A, gold), recall(B, gold), recall(C, gold)
-        per_theme.append({"theme": th, "gold_n": len(gold),
-                          "ilike": round(rA, 3), "ilike_desc": round(rB, 3), "ilike_graph": round(rC, 3)})
-        accA += rA
-        accB += rB
-        accC += rC
-        cnt += 1
-        if verbose:
-            print("  %s: gold=%d | ILIKE=%.2f | +desc=%.2f | +图=%.2f" % (th, len(gold), rA, rB, rC))
-    s.close()
-    n = cnt or 1
-    return {"per_theme": per_theme, "n_themes": cnt,
-            "ilike": round(accA / n, 3), "ilike_desc": round(accB / n, 3), "ilike_graph": round(accC / n, 3),
-            "desc_gain_pp": round((accB - accA) / n * 100, 1),
-            "graph_gain_pp": round((accC - accA) / n * 100, 1),
-            "target": TARGET}
-
-
-def main():
-    import argparse
-    ap = argparse.ArgumentParser()
-    ap.add_argument("--json", action="store_true")
-    args = ap.parse_args()
-    r = run()
-    if args.json:
-        print(json.dumps(r, ensure_ascii=False, indent=2))
-        return
-    print("\n## FW1 召回飞轮（held-out = cn_desc 提及但 name/cn 不含，非循环独立 gold）")
-    print("| 检索策略 | 召回 R | 达标(≥85%%) |")
-    print("|---|---|---|")
-    print("| ILIKE name/cn（T1.0 现状） | %.1f%% | %s |" % (r["ilike"] * 100, "✓" if r["ilike"] >= TARGET else "✗"))
-    print("| ILIKE +cn_desc（杠杆1 廉价） | %.1f%% | %s |" % (r["ilike_desc"] * 100, "✓" if r["ilike_desc"] >= TARGET else "✗"))
-    print("| ILIKE +图扩展（杠杆2） | %.1f%% | %s |" % (r["ilike_graph"] * 100, "✓" if r["ilike_graph"] >= TARGET else "✗"))
-    print("| +cn_desc 增益 | %+.1fpp | — |" % r["desc_gain_pp"])
-    print("| +图扩展 增益 | %+.1fpp | — |" % r["graph_gain_pp"])
-    best = max(r["ilike_desc"], r["ilike_graph"])
-    met = best >= TARGET
-    winner = "+cn_desc" if r["ilike_desc"] >= r["ilike_graph"] else "+图扩展"
-    print("\n飞轮状态: 最佳策略(%s)召回 %.1f%% %s 85%% 目标。%s" % (
-        winner, best * 100, "达标" if met else "未达",
-        "→ 胜出杠杆该接进 wl-search 检索；召回飞轮此轮达标，转 FW2" if met
-        else "→ 下轮：组合杠杆(+desc+图) / 语义 rag_search / 提种子数，继续转"))
-
-
-if __name__ == "__main__":
-    main()
+"""recall_flywheel.py — FW1 检索类召回飞轮（非循环 held-out gold）。
+
+T1.0 的召回 R~80% 用 ILIKE-gold 是循环的（改非ILIKE检索提不了R）。本飞轮用**非循环 gold**：
+held-out = 图邻居中【不含主题关键词】的实体（只有 graph/semantic 检索能找到，ILIKE 找不到）。
+量各检索策略对此 gold 的召回 → 改检索（加图扩展→语义）→ 重量 → 直到 ≥0.85。
+纯检索质量（无 LLM），快，可反复转飞轮。
+
+跑法: docker exec -i wlinkj-workspace-backend-1 python /app/scripts/validation/eval/recall_flywheel.py
+"""
+import os
+import sys
+import json
+
+for _p in ("/app", os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..", "..", ".."))):
+    if os.path.isdir(_p) and _p not in sys.path:
+        sys.path.insert(0, _p)
+
+from sqlalchemy import text, bindparam  # noqa: E402
+from app.db import SessionLocal  # noqa: E402
+
+THEMES = ["考勤", "排班", "加班", "车辆", "报警", "保险", "能耗", "预算", "导出", "审批"]
+TARGET = 0.85
+
+
+def _q(s, sql, params=None):
+    try:
+        return s.execute(text(sql), params or {}).fetchall()
+    except Exception:
+        try:
+            s.rollback()
+        except Exception:
+            pass
+        return []
+
+
+def ilike_ids(s, theme, limit):
+    """name/cn 关键词命中（T1.0 现状检索口径）。"""
+    return [r[0] for r in _q(s,
+        "SELECT id FROM entities WHERE (cn ILIKE :p OR canonical ILIKE :p) LIMIT :n",
+        {"p": f"%{theme}%", "n": limit})]
+
+
+def heldout_gold(s, theme, max_n=25):
+    """非循环 gold：cn_desc(描述)提及主题 但 name/cn 不含 = 只有描述/语义/图能找到，
+    关键词搜 name/cn 找不到。cn_desc 与 name/cn 是独立信号 → 公平。"""
+    rows = _q(s,
+        "SELECT id FROM entities WHERE cn_desc ILIKE :p "
+        "AND NOT (canonical ILIKE :p OR cn ILIKE :p) LIMIT :n",
+        {"p": f"%{theme}%", "n": max_n * 3})
+    return [r[0] for r in rows][:max_n]
+
+
+def strat_ilike(s, theme, limit=80):
+    """基线：name/cn 关键词（T1.0 现状）。"""
+    return set(ilike_ids(s, theme, limit))
+
+
+def strat_ilike_desc(s, theme, limit=80):
+    """杠杆1：name/cn + cn_desc（多搜描述列，廉价高杠杆）。"""
+    return set(r[0] for r in _q(s,
+        "SELECT id FROM entities WHERE (canonical ILIKE :p OR cn ILIKE :p OR cn_desc ILIKE :p) LIMIT :n",
+        {"p": f"%{theme}%", "n": limit}))
+
+
+def strat_ilike_graph(s, theme, seed_limit=25):
+    """杠杆2：name/cn 种子 + 1跳图扩展。"""
+    seeds = ilike_ids(s, theme, seed_limit)
+    res = set(seeds)
+    if seeds:
+        stmt = text(
+            "SELECT DISTINCT e.to_id FROM edges e WHERE e.from_id IN :ids"
+        ).bindparams(bindparam("ids", expanding=True))
+        try:
+            for r in s.execute(stmt, {"ids": seeds}).fetchall():
+                if r[0]:
+                    res.add(r[0])
+        except Exception:
+            try:
+                s.rollback()
+            except Exception:
+                pass
+    return res
+
+
+def recall(strategy_ids, gold):
+    if not gold:
+        return None
+    return len(set(gold) & strategy_ids) / len(gold)
+
+
+def run(verbose=True):
+    s = SessionLocal()
+    per_theme = []
+    accA = accB = accC = 0.0
+    cnt = 0
+    for th in THEMES:
+        gold = heldout_gold(s, th, 25)
+        if not gold:
+            continue
+        A = strat_ilike(s, th, 80)
+        B = strat_ilike_desc(s, th, 80)
+        C = strat_ilike_graph(s, th, 25)
+        rA, rB, rC = recall(A, gold), recall(B, gold), recall(C, gold)
+        per_theme.append({"theme": th, "gold_n": len(gold),
+                          "ilike": round(rA, 3), "ilike_desc": round(rB, 3), "ilike_graph": round(rC, 3)})
+        accA += rA
+        accB += rB
+        accC += rC
+        cnt += 1
+        if verbose:
+            print("  %s: gold=%d | ILIKE=%.2f | +desc=%.2f | +图=%.2f" % (th, len(gold), rA, rB, rC))
+    s.close()
+    n = cnt or 1
+    return {"per_theme": per_theme, "n_themes": cnt,
+            "ilike": round(accA / n, 3), "ilike_desc": round(accB / n, 3), "ilike_graph": round(accC / n, 3),
+            "desc_gain_pp": round((accB - accA) / n * 100, 1),
+            "graph_gain_pp": round((accC - accA) / n * 100, 1),
+            "target": TARGET}
+
+
+def main():
+    import argparse
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--json", action="store_true")
+    args = ap.parse_args()
+    r = run()
+    if args.json:
+        print(json.dumps(r, ensure_ascii=False, indent=2))
+        return
+    print("\n## FW1 召回飞轮（held-out = cn_desc 提及但 name/cn 不含，非循环独立 gold）")
+    print("| 检索策略 | 召回 R | 达标(≥85%%) |")
+    print("|---|---|---|")
+    print("| ILIKE name/cn（T1.0 现状） | %.1f%% | %s |" % (r["ilike"] * 100, "✓" if r["ilike"] >= TARGET else "✗"))
+    print("| ILIKE +cn_desc（杠杆1 廉价） | %.1f%% | %s |" % (r["ilike_desc"] * 100, "✓" if r["ilike_desc"] >= TARGET else "✗"))
+    print("| ILIKE +图扩展（杠杆2） | %.1f%% | %s |" % (r["ilike_graph"] * 100, "✓" if r["ilike_graph"] >= TARGET else "✗"))
+    print("| +cn_desc 增益 | %+.1fpp | — |" % r["desc_gain_pp"])
+    print("| +图扩展 增益 | %+.1fpp | — |" % r["graph_gain_pp"])
+    best = max(r["ilike_desc"], r["ilike_graph"])
+    met = best >= TARGET
+    winner = "+cn_desc" if r["ilike_desc"] >= r["ilike_graph"] else "+图扩展"
+    print("\n飞轮状态: 最佳策略(%s)召回 %.1f%% %s 85%% 目标。%s" % (
+        winner, best * 100, "达标" if met else "未达",
+        "→ 胜出杠杆该接进 wl-search 检索；召回飞轮此轮达标，转 FW2" if met
+        else "→ 下轮：组合杠杆(+desc+图) / 语义 rag_search / 提种子数，继续转"))
+
+
+if __name__ == "__main__":
+    main()