baseline: Clover 独立仓库首次基线提交

将 Clover 从上层产品包旧仓库中独立出来,建立专属版本控制。当前状态=纵切片端到端已打通(登录→选品→出文出图→审核→下载包), M1文案质量去套路化已验收。此提交作为后续按核销清单逐条修复的基线。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-16 11:30:22 +08:00
commit 6a2632da70
253 changed files with 27467 additions and 0 deletions
--- a/backend/tests/test_ai_engine.py
+++ b/backend/tests/test_ai_engine.py
@@ -0,0 +1,253 @@
+"""
+AI 引擎单元测试（核心逻辑覆盖）
+对照 JS 版逻辑验证 Python 重写正确性
+"""
+import sys
+import os
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), ".."))
+
+import pytest
+from app.services.ai_engine.text_scoring import score_copy, is_similar_copy, dedupe_copies
+from app.services.ai_engine.banned_word_checker import check_and_fix, BannedWordEntry
+from app.services.ai_engine.storyboard import get_narrative_roles, plan_image_set, clamp_count
+from app.services.ai_engine.preference_aggregator import aggregate_preference_context, collect_preference_event
+from app.services.ai_engine.prompt_composer import (
+    compose_variants, compose_preference_context, compose_image_prompt, parse_model_output
+)
+
+# ── 测试用 product ─────────────────────────────────────────
+PRODUCT = {
+    "name": "倍分子素颜霜",
+    "category": "美妆护肤",
+    "selling_points": ["轻薄不厚重", "水润自然", "不卡粉"],
+    "keywords": ["素颜霜", "日常通勤"],
+    "style_tone": "素人分享风",
+    "text_angles": ["痛点切入", "场景型", "避坑型"],
+    "custom_prompt": "",
+    "target_audience": "上班族",
+}
+
+GOOD_COPY = {
+    "title": "早八素颜也能有点气色！✨",
+    "content": "姐妹们，最近实测下来，轻薄不厚重是我最在意的。✅ 通勤路上随手一抹，水润自然有气色。不卡粉这点太加分了。整体就是那种上班族能马上代入的日常好物。",
+    "tags": ["#素颜霜", "#日常通勤", "#好物分享", "#真实测评"],
+    "angle": "场景型",
+    "buyingPoint": "轻薄不厚重，适合上班族",
+    "coverTitle": "早八素颜也能有点气色",
+    "imageBrief": "封面自然光上脸局部，内页质地推开+软性转化。",
+    "source": "ai",
+}
+
+
+# ── 五维打分 ───────────────────────────────────────────────
+class TestScoreCopy:
+    def test_good_copy_passes(self):
+        result = score_copy(GOOD_COPY, PRODUCT)
+        assert result["passed"] is True
+        assert result["score"] >= 90
+
+    def test_banned_word_kills_compliance(self):
+        bad = {**GOOD_COPY, "title": "美白素颜霜强推！", "content": GOOD_COPY["content"] + "美白效果显著"}
+        result = score_copy(bad, PRODUCT)
+        # 含美白 → 合规0分 + 总分不过
+        assert result["passed"] is False
+        compliance = next(d for d in result["score_detail"] if d["item"] == "合规性")
+        assert compliance["score"] == 0
+
+    def test_score_detail_has_five_dims(self):
+        result = score_copy(GOOD_COPY, PRODUCT)
+        assert len(result["score_detail"]) == 5
+        items = {d["item"] for d in result["score_detail"]}
+        assert "标题吸引力" in items
+        assert "情绪共鸣" in items
+        assert "买点表达" in items
+        assert "关键词覆盖" in items
+        assert "合规性" in items
+
+
+# ── 去重 ───────────────────────────────────────────────────
+class TestDedup:
+    def test_identical_title_deduped(self):
+        a = {**GOOD_COPY}
+        b = {**GOOD_COPY, "content": "略有不同的内容，但标题一样"}
+        result = dedupe_copies([a, b])
+        assert len(result) == 1
+
+    def test_different_angle_both_kept(self):
+        a = {**GOOD_COPY, "angle": "痛点切入"}
+        b = {**GOOD_COPY, "title": "通勤懒人必囤！换季不再干脸", "angle": "避坑型"}
+        result = dedupe_copies([a, b])
+        assert len(result) == 2
+
+    def test_similar_body_deduped(self):
+        a = {**GOOD_COPY}
+        b = {**GOOD_COPY, "title": "不一样的标题", "content": GOOD_COPY["content"][:150] + "稍有变动"}
+        assert is_similar_copy(a, b) or len(dedupe_copies([a, b])) <= 2
+
+
+# ── 违禁词三级 ──────────────────────────────────────────────
+class TestBannedWordChecker:
+    def test_hard_block(self):
+        # "美白"已改为 auto_fix（提亮肤色感），hard_block 用速效/医用等词
+        result = check_and_fix("速效医用护肤品")
+        assert result.status == "hard_block"
+
+    def test_meibaibanned_now_auto_fix(self):
+        # Q5对齐：美白→auto_fix 提亮肤色感（不再 hard_block）
+        result = check_and_fix("这款美白效果很好")
+        assert result.status == "auto_fixed"
+        assert result.fixed_text is not None
+        assert "美白" not in result.fixed_text
+
+    def test_auto_fix(self):
+        result = check_and_fix("这款神器真的好用")
+        assert result.status == "auto_fixed"
+        assert result.fixed_text is not None
+        assert "神器" not in result.fixed_text
+
+    def test_soft_warn(self):
+        result = check_and_fix("这绝对是最好的护肤品")
+        assert result.status == "soft_warn"
+
+    def test_clean_text_passes(self):
+        result = check_and_fix("这款素颜霜轻薄水润，通勤必备。")
+        assert result.status == "pass"
+
+    def test_custom_entries_override(self):
+        entries = [BannedWordEntry("必备", "hard_block")]
+        result = check_and_fix("通勤必备好物", entries)
+        assert result.status == "hard_block"
+
+
+# ── storyboard ─────────────────────────────────────────────
+class TestStoryboard:
+    def test_clamp_count(self):
+        assert clamp_count(0) == 1
+        assert clamp_count(9) == 8
+        assert clamp_count(3) == 3
+
+    def test_narrative_roles_3(self):
+        roles = get_narrative_roles(3)
+        assert len(roles) == 3
+        assert roles[0]["role"] == "hook"
+        assert roles[-1]["role"] == "closer"
+
+    def test_narrative_roles_6(self):
+        roles = get_narrative_roles(6)
+        assert len(roles) == 6
+        # Q6对齐北哥套路：①hook ②product_closeup(单品特写) ③ingredient ④texture ⑤applied_proof ⑥closer
+        assert roles[0]["role"] == "hook"
+        assert roles[1]["role"] == "product_closeup"
+        assert roles[2]["role"] == "ingredient"
+        assert roles[4]["role"] == "applied_proof"
+        assert roles[5]["role"] == "closer"
+
+    def test_narrative_roles_8(self):
+        roles = get_narrative_roles(8)
+        assert len(roles) == 8
+
+    def test_plan_image_set_structure(self):
+        plan = plan_image_set(GOOD_COPY, PRODUCT, image_count=3)
+        assert "storyboard" in plan
+        assert "base_prompt" in plan
+        assert len(plan["storyboard"]) == 3
+        # 产品图锚点说明必须在 base_prompt 中
+        assert "不可修改" in plan["base_prompt"]
+
+
+# ── 飞轮聚合 ───────────────────────────────────────────────
+class TestPreferenceAggregator:
+    def test_cold_start_when_few_events(self):
+        result = aggregate_preference_context([], PRODUCT, workspace_id=1, product_id=1)
+        assert result["injected_count"] == 0
+        assert "冷启动" in result["recent_preference"]
+
+    def test_aggregate_top_angles(self):
+        events = [
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1, "angle_label": "痛点切入", "signal_weight": 3, "reason": ""},
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1, "angle_label": "痛点切入", "signal_weight": 3, "reason": ""},
+            {"signal_type": "approve",     "workspace_id": 1, "product_id": 1, "angle_label": "场景型",  "signal_weight": 5, "reason": ""},
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1, "angle_label": "避坑型",  "signal_weight": 3, "reason": ""},
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1, "angle_label": "痛点切入", "signal_weight": 3, "reason": ""},
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1, "angle_label": "避坑型",  "signal_weight": 3, "reason": ""},
+        ]
+        result = aggregate_preference_context(events, PRODUCT, workspace_id=1, product_id=1)
+        assert result["injected_count"] == 6
+        assert "痛点切入" in result["recent_preference"]
+        assert "偏好角度" in result["prompt_fragment"]
+
+    def test_reject_reason_in_prompt(self):
+        events = [
+            {"signal_type": "reject_with_reason", "workspace_id": 1, "product_id": 1,
+             "angle_label": "", "signal_weight": -3, "reason": "标题太硬广"},
+            *[{"signal_type": "text_select", "workspace_id": 1, "product_id": 1,
+               "angle_label": "痛点切入", "signal_weight": 3, "reason": ""} for _ in range(5)],
+        ]
+        result = aggregate_preference_context(events, PRODUCT, workspace_id=1, product_id=1)
+        assert "标题太硬广" in result["prompt_fragment"]
+
+    def test_product_isolation(self):
+        """不同 product_id 的事件不会混进来"""
+        events = [
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 2,
+             "angle_label": "成分党", "signal_weight": 3, "reason": ""},
+            *[{"signal_type": "text_select", "workspace_id": 1, "product_id": 1,
+               "angle_label": "场景型", "signal_weight": 3, "reason": ""} for _ in range(5)],
+        ]
+        result = aggregate_preference_context(events, PRODUCT, workspace_id=1, product_id=1)
+        assert "成分党" not in result["prompt_fragment"]
+
+    def test_collect_event_structure(self):
+        event = collect_preference_event("text_select", user_id=1, workspace_id=1, product_id=1, angle_label="痛点切入")
+        assert event["signal_weight"] == 3
+        assert event["data_ownership"] == "client_data"
+
+
+# ── prompt_composer ────────────────────────────────────────
+class TestPromptComposer:
+    def test_compose_variants_returns_two_strings(self):
+        sys_p, user_p = compose_variants(PRODUCT, count=5)
+        assert isinstance(sys_p, str) and len(sys_p) > 0
+        assert isinstance(user_p, str) and len(user_p) > 0
+
+    def test_compose_variants_includes_product_name(self):
+        _, user_p = compose_variants(PRODUCT, count=3)
+        assert "倍分子素颜霜" in user_p
+
+    def test_compose_variants_includes_count(self):
+        _, user_p = compose_variants(PRODUCT, count=7)
+        assert "7" in user_p
+
+    def test_compose_variants_flywheel_injected(self):
+        fragment = "偏好角度参考：场景型、痛点切入"
+        _, user_p = compose_variants(PRODUCT, count=3, flywheel_context=fragment)
+        assert "场景型" in user_p
+
+    def test_compose_preference_context_delegates(self):
+        # compose_preference_context 应委托给 aggregate_preference_context
+        events = [
+            {"signal_type": "text_select", "workspace_id": 1, "product_id": 1,
+             "angle_label": "场景型", "signal_weight": 3, "reason": ""}
+            for _ in range(6)
+        ]
+        result = compose_preference_context(events, PRODUCT, workspace_id=1, product_id=1)
+        assert "injected_count" in result
+        assert result["injected_count"] == 6
+        assert "prompt_fragment" in result
+
+    def test_parse_model_output_valid_json(self):
+        raw = '[{"title":"标题1","content":"正文","tags":[],"angle":"场景型"}]'
+        parsed = parse_model_output(raw)
+        assert len(parsed) == 1 and parsed[0]["title"] == "标题1"
+
+    def test_parse_model_output_markdown_wrapped(self):
+        raw = "```json\n[{\"title\":\"a\",\"content\":\"b\",\"tags\":[]}]\n```"
+        parsed = parse_model_output(raw)
+        assert len(parsed) == 1
+
+    def test_compose_image_prompt_includes_role_and_product(self):
+        vs = {"style": "ins摆拍风", "color_palette": "米白+杏色", "base_prompt": "产品近景"}
+        prompt = compose_image_prompt("hook", vs, PRODUCT)
+        assert "hook" in prompt
+        assert "倍分子素颜霜" in prompt
+        assert "ins摆拍风" in prompt