kba-cli 0.3.0rc4__py3-none-any.whl

kba/tags.py

@@ -0,0 +1,108 @@
+"""Three-layer tag/domain inference: frontmatter + path/filename + content keywords."""
+from __future__ import annotations
+
+import re
+
+# Conservative subset of packs/generic/tags.md; only infer a tag when a path/name
+# keyword matches or at least two distinct content keywords are present.
+TAG_RULES = {
+    "行业研究": ("行业", "研报", "市场", "竞品", "公司研究", "半导体", "新能源", "消费", "医药", "策略"),
+    "财务": ("财务", "估值", "收入", "利润", "现金流", "预算", "成本", "模型", "三张表"),
+    "合同": ("合同", "协议", "条款", "法务", "甲方", "乙方"),
+    "合规": ("合规", "法规", "审计", "隐私", "监管"),
+    "安全": ("安全", "权限", "风控", "密钥", "漏洞"),
+    "AI": ("AI", "大模型", "LLM", "RAG", "Agent", "提示词", "机器学习"),
+    "数据治理": ("数据治理", "主数据", "数据质量", "数据标准", "数据口径"),
+    "指标体系": ("指标", "KPI", "口径", "度量", "报表"),
+    "知识库": ("知识库", "知识管理", "检索", "标签体系", "语料"),
+    "方案": ("方案", "建设", "规划", "实施"),
+    "架构": ("架构", "技术选型", "接口", "部署", "微服务"),
+    "需求": ("需求", "PRD", "用户故事", "验收标准"),
+    "复盘": ("复盘", "回顾", "总结", "经验"),
+    "教程": ("教程", "指南", "操作手册", "步骤"),
+    "培训": ("培训", "课程", "内训", "教材"),
+    "营销": ("营销", "品牌", "活动", "增长", "投放"),
+    "销售": ("销售", "线索", "客户", "转化", "商机"),
+    "客户案例": ("案例", "标杆", "客户故事"),
+    "项目管理": ("项目", "排期", "里程碑", "风险"),
+    "模板": ("模板", "清单", "可复用"),
+}
+
+DOMAIN_NAMES = (
+    "业务方案", "技术架构", "产品设计", "数据与指标",
+    "学习与考试", "内容创作", "个人事务", "参考资料",
+)
+
+DOMAIN_BY_TAG = {
+    "架构": "技术架构", "接口": "技术架构", "研发": "技术架构", "安全": "技术架构",
+    "运维": "技术架构", "AI": "技术架构", "效率工具": "技术架构",
+    "产品设计": "产品设计", "用户研究": "产品设计", "需求": "产品设计",
+    "数据治理": "数据与指标", "指标体系": "数据与指标", "报表": "数据与指标",
+    "教程": "学习与考试", "考试": "学习与考试", "学习笔记": "学习与考试", "培训": "学习与考试",
+    "内容创作": "内容创作", "自媒体": "内容创作", "品牌": "内容创作", "营销": "内容创作",
+    "求职": "个人事务", "职业发展": "个人事务", "个人效率": "个人事务",
+    "方案": "业务方案", "项目管理": "业务方案", "复盘": "业务方案", "方法论": "业务方案",
+    "客户案例": "业务方案", "销售": "业务方案", "客户成功": "业务方案",
+    "合同": "业务方案", "合规": "业务方案",
+    "行业研究": "参考资料", "财务": "参考资料", "参考资料": "参考资料",
+}
+
+
+def _contains(keyword: str, text: str) -> bool:
+    """ASCII keywords need word boundaries; Chinese uses substring."""
+    if not keyword:
+        return False
+    if keyword.isascii():
+        return re.search(rf"(?<![a-z0-9]){re.escape(keyword)}(?![a-z0-9])", text, re.I) is not None
+    return keyword in text
+
+
+def infer_metadata(rel: str, name: str, text: str, fm_tags, fm_domain: str = "",
+                   max_tags: int = 6):
+    """Return (tags, sources, inferred_tags, domain, domain_source)."""
+    fm_tags = [str(t).strip() for t in (fm_tags or []) if str(t).strip()]
+    tags: list[str] = list(dict.fromkeys(fm_tags))
+    sources: set[str] = {"frontmatter"} if fm_tags else set()
+    inferred: list[str] = []
+
+    haystack = f"{rel or ''} {name or ''}".lower()
+    for tag, keywords in TAG_RULES.items():
+        if tag in tags:
+            continue
+        if any(_contains(k.lower(), haystack) for k in keywords):
+            tags.append(tag)
+            inferred.append(tag)
+            sources.add("path")
+
+    sample = (text or "")[:60_000].lower()
+    for tag, keywords in TAG_RULES.items():
+        if tag in tags:
+            continue
+        hits = sum(1 for k in keywords if _contains(k.lower(), sample))
+        if hits >= 2:
+            tags.append(tag)
+            inferred.append(tag)
+            sources.add("content")
+
+    if len(tags) > max_tags:
+        front = [t for t in tags if t in fm_tags]
+        extra = [t for t in tags if t not in fm_tags][:max(0, max_tags - len(front))]
+        tags = front + extra
+        inferred = [t for t in inferred if t in tags]
+
+    domain = (fm_domain or "").strip()
+    domain_source = "frontmatter" if domain else ""
+    if not domain:
+        for segment in (rel or "").split("/"):
+            if segment in DOMAIN_NAMES:
+                domain = segment
+                domain_source = "path"
+                break
+    if not domain:
+        for tag in tags:
+            mapped = DOMAIN_BY_TAG.get(tag)
+            if mapped:
+                domain = mapped
+                domain_source = "tags"
+                break
+    return tags, sorted(sources), inferred, domain, domain_source

kba/telemetry.py

@@ -0,0 +1,78 @@
+"""Opt-in, aggregated telemetry. No network code in v0.1 — preview only."""
+from __future__ import annotations
+
+import json
+import uuid
+from pathlib import Path
+
+from . import util
+from .jobs import kba_home
+
+
+def _cfg_path() -> Path:
+    return kba_home() / "config.json"
+
+
+def _load() -> dict:
+    return util.read_json(_cfg_path(), {}) or {}
+
+
+def enabled() -> bool:
+    return bool(_load().get("telemetry", False))
+
+
+def set_enabled(value: bool) -> dict:
+    cfg = _load()
+    cfg["telemetry"] = bool(value)
+    cfg.setdefault("anon_id", str(uuid.uuid4()))
+    util.write_json(_cfg_path(), cfg)
+    return cfg
+
+
+def anon_id() -> str:
+    cfg = _load()
+    if not cfg.get("anon_id"):
+        cfg["anon_id"] = str(uuid.uuid4())
+        util.write_json(_cfg_path(), cfg)
+    return cfg["anon_id"]
+
+
+def build_payload(job: Path) -> dict:
+    job = Path(job)
+    fp = util.read_json(job / "fp.json") or {}
+    analysis = util.read_json(job / "analysis.json") or {}
+    m = analysis.get("metrics", {})
+    s = fp.get("summary", {})
+    return {
+        "schema": "kba.telemetry/1",
+        "anon_id": anon_id(),
+        "event": "audit_completed",
+        "ts": util.now_iso(),
+        "props": {
+            "files": s.get("files", 0),
+            "bytes": s.get("bytes", 0),
+            "by_ext": s.get("by_ext", {}),
+            "dup_rate": m.get("dup_rate"),
+            "stale_rate": m.get("stale_rate"),
+            "tag_coverage": m.get("tag_coverage"),
+            "valid_tag_coverage": m.get("valid_tag_coverage"),
+            "top5_hit": m.get("top5_hit"),
+            "extract_fail": m.get("extract_fail"),
+            "health_score": analysis.get("health_score"),
+            "kba_version": __import__("kba").__version__,
+        },
+    }
+
+
+def record(job: Path) -> None:
+    if not enabled():
+        return
+    payload = build_payload(job)
+    q = kba_home() / "telemetry_queue.jsonl"
+    q.parent.mkdir(parents=True, exist_ok=True)
+    with open(q, "a", encoding="utf-8") as f:
+        f.write(json.dumps(payload, ensure_ascii=False) + "\n")
+
+
+def preview(job: Path) -> dict:
+    return build_payload(job)

kba/util.py

@@ -0,0 +1,204 @@
+"""Shared helpers: encoding, frontmatter, file walking, hashing."""
+from __future__ import annotations
+
+import hashlib
+import json
+import os
+import re
+import time
+import unicodedata
+from datetime import datetime
+from pathlib import Path
+
+TEXT_EXTS = {".md", ".markdown", ".txt", ".rst", ".org"}
+# Binaries with a local text extractor (via `kba extract` sidecars).
+EXTRACTABLE_EXTS = {".pdf", ".docx", ".pptx", ".xlsx"}
+# Legacy binaries that are counted but not content-analysed yet.
+# LibreOffice conversion is a later option; do not silently drop them.
+UNSUPPORTED_TEXT_EXTS = {".ppt", ".xls", ".doc"}
+BINARY_EXTS = EXTRACTABLE_EXTS | UNSUPPORTED_TEXT_EXTS
+SUPPORTED_EXTS = TEXT_EXTS | BINARY_EXTS
+DEFAULT_EXCLUDE = {
+    "_index", "_extract", "_hub", "_organize", ".git", "__pycache__",
+    "node_modules", ".venv", "venv", "dist", "build",
+}
+DEFAULT_TEXT_TTL_DAYS = 7
+
+
+def now_iso() -> str:
+    return datetime.now().astimezone().isoformat(timespec="seconds")
+
+
+def text_ttl_days() -> int:
+    """Days to keep job/text/ content; 0 disables retention (purge on maintenance)."""
+    raw = os.environ.get("KBA_TEXT_TTL_DAYS")
+    try:
+        return max(0, int(raw)) if raw not in (None, "") else DEFAULT_TEXT_TTL_DAYS
+    except (TypeError, ValueError):
+        return DEFAULT_TEXT_TTL_DAYS
+
+
+def ext_kind(ext: str) -> str:
+    """text | extractable | unsupported | other."""
+    ext = (ext or "").lower()
+    if ext in TEXT_EXTS:
+        return "text"
+    if ext in EXTRACTABLE_EXTS:
+        return "extractable"
+    if ext in UNSUPPORTED_TEXT_EXTS:
+        return "unsupported"
+    return "other"
+
+
+FINAL_MARK_RE = re.compile(r"终稿|定稿|终版|发布版|已发布|final(?:ized)?", re.I)
+DRAFT_MARK_RE = re.compile(r"底稿|草稿|初稿|过程稿|中间稿|draft|wip|working", re.I)
+VERSION_NUM_RE = re.compile(r"(?:^|[\s_\-（(])v(?:er)?\.?\s?(\d+(?:\.\d+)*)", re.I)
+
+
+def version_stage(name: str, rel: str = "") -> str:
+    """final | draft | normal, inferred from filename/path markers only."""
+    s = f"{rel or ''} {name or ''}"
+    if DRAFT_MARK_RE.search(s):
+        return "draft"
+    if FINAL_MARK_RE.search(s):
+        return "final"
+    return "normal"
+
+
+def version_number(name: str, rel: str = "") -> str:
+    m = VERSION_NUM_RE.search(f"{rel or ''} {name or ''}")
+    return m.group(1) if m else ""
+
+
+def text_status_of(rec: dict, extraction_enabled: bool) -> str:
+    """Per-file extraction state, tolerating fingerprints written before P0."""
+    st = rec.get("text_status")
+    if st:
+        return st
+    kind = ext_kind(rec.get("ext", ""))
+    if kind == "text":
+        return "text"
+    if kind == "unsupported":
+        return "unsupported"
+    if rec.get("has_text"):
+        return "extracted"
+    return "not_enabled" if not extraction_enabled else "failed"
+
+
+def read_text(path: Path) -> str:
+    """BOM-aware UTF-8/UTF-16, then strict gb18030/big5/shift_jis; else empty."""
+    try:
+        raw = Path(path).read_bytes()
+    except OSError:
+        return ""
+    if raw.startswith((b"\xff\xfe", b"\xfe\xff")):
+        try:
+            return raw.decode("utf-16")
+        except UnicodeDecodeError:
+            pass
+    for enc in ("utf-8-sig", "gb18030", "big5", "shift_jis"):
+        try:
+            return raw.decode(enc)
+        except (UnicodeDecodeError, UnicodeError):
+            continue
+    return ""
+
+
+FM_RE = re.compile(r"^--- *\r?\n(.*?)\r?\n?--- *\r?\n?(.*)$", re.S)
+
+
+def parse_frontmatter(text: str):
+    m = FM_RE.match(text)
+    if not m:
+        return {}, text
+    raw, body = m.group(1), m.group(2)
+    fm = {}
+    for line in raw.splitlines():
+        if not line.strip() or line.strip().startswith("#") or ":" not in line:
+            continue
+        k, v = line.split(":", 1)
+        k, v = k.strip(), v.strip().strip("'\"").strip()
+        if v.startswith("[") and v.endswith("]"):
+            v = [x.strip().strip("'\"") for x in v[1:-1].split(",") if x.strip()]
+        fm[k] = v
+    return fm, body
+
+
+def iter_files(root: Path, exclude=None, exts=None):
+    root = Path(root).resolve()
+    exclude = set(exclude or DEFAULT_EXCLUDE)
+    exts = {e.lower() for e in (exts or SUPPORTED_EXTS)}
+    for dirpath, dirnames, filenames in os.walk(root):
+        dirnames[:] = sorted(d for d in dirnames if d not in exclude)
+        for fn in sorted(filenames):
+            # Office lock/temp files (~$deck.pptx) are not real documents.
+            if fn.startswith("~$"):
+                continue
+            p = Path(dirpath) / fn
+            if p.suffix.lower() not in exts:
+                continue
+            if p.is_symlink():
+                try:
+                    if root not in p.resolve().parents and p.resolve() != root:
+                        continue
+                except OSError:
+                    continue
+            yield p
+
+
+def sha256_file(path: Path) -> str:
+    h = hashlib.sha256()
+    with open(path, "rb") as f:
+        for chunk in iter(lambda: f.read(1 << 20), b""):
+            h.update(chunk)
+    return h.hexdigest()
+
+
+def human_bytes(n: float) -> str:
+    for unit in ("B", "KB", "MB", "GB", "TB"):
+        if abs(n) < 1024.0:
+            return f"{n:.1f}{unit}"
+        n /= 1024.0
+    return f"{n:.1f}PB"
+
+
+def normalize_name(name: str) -> str:
+    s = unicodedata.normalize("NFKC", name.lower())
+    s = re.sub(r"__dup\d+", "", s)
+    s = re.sub(r"[\s\-_()（）\[\]【】]+", "", s)
+    return s
+
+
+def write_json(path: Path, obj, indent: int = 2) -> None:
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+    tmp = path.with_name(f".{path.name}.{os.getpid()}.{time.time_ns()}.tmp")
+    tmp.write_text(json.dumps(obj, ensure_ascii=False, indent=indent), encoding="utf-8")
+    # Windows can briefly lock a JSON target (AV/indexer); retry the atomic replace.
+    for attempt in range(8):
+        try:
+            tmp.replace(path)
+            break
+        except PermissionError:
+            if attempt == 7:
+                raise
+            time.sleep(0.02)
+
+
+def read_json(path: Path, default=None):
+    try:
+        return json.loads(Path(path).read_text(encoding="utf-8"))
+    except (OSError, ValueError):
+        return default
+
+
+def rel_posix(path: Path, root: Path) -> str:
+    return Path(path).resolve().relative_to(Path(root).resolve()).as_posix()
+
+
+def safe_name(rel: str) -> str:
+    return hashlib.sha256(rel.encode("utf-8")).hexdigest()[:12]
+
+
+def mtime_days_ago(ts: float) -> float:
+    return max(0.0, (datetime.now().timestamp() - ts) / 86400.0)

kba_cli-0.3.0rc4.dist-info/METADATA

@@ -0,0 +1,135 @@
+Metadata-Version: 2.4
+Name: kba-cli
+Version: 0.3.0rc4
+Summary: KBA - read-only knowledge base auditor (local-first)
+Author: KBA
+License-Expression: MIT
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Provides-Extra: extract
+Requires-Dist: pypdf>=4.0; extra == "extract"
+Requires-Dist: python-docx>=1.1; extra == "extract"
+Requires-Dist: python-pptx>=0.6.23; extra == "extract"
+Requires-Dist: openpyxl>=3.1; extra == "extract"
+Dynamic: license-file
+
+# 知库体检（KBA）
+
+> 只读体检本地知识库：健康分、重复裁决建议、治理计划；数据不出本机。
+> 独立项目，不修改 `D:\content_hub`；复用其算法与规则（独立实现）。
+
+## 状态（2026-11-05）
+
+**v0.3.0rc4（内部预发布）：P0 报告诚实层 + P1 抽取层 + P1 准确性前两批 + 抽取并行/TTL + 标注评测。**
+**Gate 1 外部语料验证未通过，不能对外称稳定版或“完整审阅”。**
+
+- 真实内部语料（215 文件 / 58.2MB）只读跑通：抽取 110/113（失败 0、no_text 3）、文本覆盖 97.4%、健康 80、重复 2 组、标签有效覆盖 75.8%、自检索 48%、owner 0%；仍是单个内部语料，不是 Gate 1
+- 真实语料 54 文件 AI 辅助标注样本：抽取 97.35%、PII 8/8；近重复/同名/页级标签由预测簇反推，仅验证链路；标签 exact 0、Jaccard 0.13，说明推断标签过标，需人工校准
+- P0：A/B/C 数据置信度、健康分按可用指标归一化并显示权重覆盖度、抽取三态、Office 计入盘点、正文缓存 TTL 与 `--purge-text`、报告“数据覆盖与结论适用范围”；C 级元数据报告不展示综合健康分
+- P1 抽取：`kba extract` 用 pypdf / python-docx / python-pptx / openpyxl 抽取 PDF、Word、PPT、Excel 正文；Excel 结构指纹（sheet/公式模式/named range）识别“同模型不同版本”；旧版 `.ppt/.xls/.doc` 显式标 unsupported
+- P1 准确性：PDF 去页眉页脚、PDF/PPT 页级切分、跨文档页级重合（不自动合并）、终稿/底稿信号 + 正文字数/相似度差异、人工裁决队列、PII 正则初筛（仅计数）、标签/领域三层推断（frontmatter + 路径/文件名 + 内容）、负责人覆盖与本地断链检测、`kba clean --purge-extract`
+- P1 运维：`kba extract --workers` 并行抽取、默认 7 天 sidecar TTL、`tools/bench_scan.py` 合成基准
+- 评测：`kba eval` 读取标注 JSON，输出近重复/同名/页级/版本/PII 的 P/R/F1、标签 exact/Jaccard、抽取成功率；合成语料只证明回归，不代表外部准确率
+- 本地连接器：免费 `connector-local`（kba-local）+ `kba scan-summary`；付费专家依赖它，连接器不可用才回退粘贴摘要
+- 产出：[demo/report.html](demo/report.html)、[demo/share.md](demo/share.md)（均为 P0 前版本，重新生成后含新章节）
+- 38 个测试通过（P0 验收 + 抽取/结构指纹/失败态/UTF-16/页级/版本/PII/标签/断链/purge/并行/TTL/评测）
+- **免费入口已上线**：专家 v0.1.4 审核通过、平台可召唤，prompt-only，不依赖 CLI；**付费未开通**，付费专家/连接器包已就绪但未提交平台
+
+## 快速开始
+
+```bash
+cd D:\kba
+pip install -e ".[extract]"       # 含 PDF/Office 抽取依赖；不装也能跑元数据模式
+# 或安装内部预发布 wheel（本地构建，不对外发布）：
+# python -m pip install "dist/kba_cli-0.3.0rc4-py3-none-any.whl[extract]"
+
+python -m kba doctor                # 环境自检
+python -m kba extract --root "D:\你的资料" --out "D:\kba-extract"
+python -m kba audit start --root "D:\你的资料" --extract-dir "D:\kba-extract" --foreground
+python -m kba audit analyze
+python -m kba audit report --out report.html
+python -m kba audit share  --out share.md
+python -m kba eval --root "D:\你的资料" --labels labels.json --json  # 标注语料评测
+python -m kba clean --purge-text    # 立即清理 job/text 正文缓存（默认 TTL 7 天）
+python -m kba clean --purge-extract --dir "D:\kba-extract"  # 清理抽取 sidecar
+```
+
+只做元数据盘点时可跳过 `kba extract` 与 `--extract-dir`；报告会标 C 级且不展示综合健康分。
+
+平台内使用（WorkBuddy）：
+
+```
+连接器：connector/（CLI + Skill 方案）
+技能：kb-audit / kb-clusters / kb-taxonomy
+```
+
+## 目录
+
+```
+kba/
+├── kba/                 # Python 包（cli/doctor/jobs/fingerprint/extract/tags/links/pii/analyze/report/telemetry）
+├── connector/           # WorkBuddy 发布包（meta / cli.json / 3 个 SKILL）
+├── plugin-free/         # 免费引导插件包（技能入口：.codebuddy-plugin + skills）
+├── plugin-free-expert/  # 免费引导专家包（专家入口：plugin.json + agents + avatar）
+├── plugin-paid-expert/  # 付费专家包 v0.1.0（报告解读；不读本地文件）
+├── plugin-paid/         # ¥1 付费试用插件包（旧方案）
+├── packs/generic/       # 通用行业包（taxonomy/tags/rules/report）
+├── tests/               # smoke tests
+├── tools/               # vendor 同步 / 插件打包 / 提审检查 / 合成基准
+├── demo/                # 真实语料报告（本地生成，不入库）
+└── docs/                # PRD、平台 6 问、访谈提纲、定价实验
+```
+
+## 文档
+
+| 文档 | 说明 |
+|---|---|
+| [CHANGELOG.md](CHANGELOG.md) | 版本变更与预发布说明（0.3.0rc4） |
+| [docs/eval-labels.md](docs/eval-labels.md) | `kba eval` 标注格式、指标口径与 Gate 1 用法 |
+| [docs/free-entry-runbook.md](docs/free-entry-runbook.md) | 免费专家 v0.1.4 召唤验收、冻结规则与反馈表 |
+| [plugin-paid-expert/README.md](plugin-paid-expert/README.md) | 付费专家包 v0.1.0：定位、发布字段与提审检查 |
+| [connector-local/README.md](connector-local/README.md) | 免费本地连接器 kba-local：结构、命令与包源前置 |
+| [docs/handoff.md](docs/handoff.md) | **新任务交接**：现状、决策、P0/P1 状态、验收与启动提示词 |
+| [docs/master-assessment-and-plan.md](docs/master-assessment-and-plan.md) | **单一事实源**：完整缺口、可信度分级、P0/P1/P2 计划与验收门 |
+| [docs/prd-conformance-review.md](docs/prd-conformance-review.md) | PRD 与 WorkBuddy 专家规范符合性评审、偏差与修正清单 |
+| [docs/PRD.md](docs/PRD.md) | PRD v1.8：需求、FR、接口、决策建议、v0.3.0rc4 实现状态 |
+| [docs/platform-questions.md](docs/platform-questions.md) | 平台 6 问调研答案、提审避坑清单与剩余 2 个待验证项 |
+| [docs/interview-guide.md](docs/interview-guide.md) | 10 人次访谈提纲与评分 |
+| [docs/interview-guide-investment-research.md](docs/interview-guide-investment-research.md) | 投研首站的 10 人次访谈提纲（卖方/买方/独立/IT） |
+| [docs/vertical-investment-research.md](docs/vertical-investment-research.md) | 投研垂直方案：ICP、痛点、体检产品、定价与 90 天验证 |
+| [docs/pricing-experiment.md](docs/pricing-experiment.md) | 免费/付费切分与 ¥99/199/299 实验 |
+| [docs/paid-trial-runbook.md](docs/paid-trial-runbook.md) | ¥1 付费试用技能的完整执行步骤、判据与分流 |
+| [docs/pre-submit-review.md](docs/pre-submit-review.md) | 提审前评审：自动化检查、逐项合规、修复项与剩余风险 |
+| [docs/submission-log.md](docs/submission-log.md) | 平台提交记录、审核状态与过审后的操作顺序 |
+| [docs/repo-setup.md](docs/repo-setup.md) | 远程仓库、CI、分支保护与公开前的拆分计划 |
+| [docs/developer-profile.md](docs/developer-profile.md) | 开发者主页：头像、昵称、简介与联系方式填写建议 |
+| [docs/product-gap-assessment.md](docs/product-gap-assessment.md) | 产品缺口深度评估：数据接入、分析准确性、报告可信度与优先级 |
+
+## 已知限制
+
+- **只读**：v0.3.0rc4 不提供 apply/undo，治理执行仍在服务侧。
+- **抽取是可选的**：不跑 `kba extract` 时仍是元数据模式，报告标 C 级、不展示综合健康分；`pip install -e ".[extract]"` 是抽取前置。
+- **旧版 Office 只计数**：`.ppt/.xls/.doc` 不做 LibreOffice 转换，显式标 unsupported；`.pptx/.xlsx` 已可抽取。
+- **Excel 结构指纹只识别有公式/named range 的模型**：纯数值工作簿即使 sheet 同名也不会被当作“同模型不同版本”，避免误报。
+- **页级重合不是整档裁决**：PDF/PPT 单页高重合会报告，但不进入重复簇、不自动折叠；版本差异目前是终稿/底稿信号 + 字数/相似度/增删字符摘要，未做逐段正文语义 diff。
+- **PII 是正则初筛**：只覆盖邮箱/手机号/身份证/银行卡/密钥/合同编号模式，可能误报/漏报；报告只显示计数与文件名，不展示原文。
+- **编码仍有边界**：UTF-8/GB18030/UTF-16(BOM) 可读；Big5/Shift-JIS 在 GB18030 可解码时仍可能是乱码（已报 `text_unreadable` 的只覆盖解码失败场景）。
+- **正文缓存有 TTL 但不加密**：`job/text/` 默认保留 7 天；`kba extract` 默认在抽取前清理超过 7 天的 sidecar（`--ttl-days -1` 关闭）；`kba clean --purge-text` / `--purge-extract` 可立即清理。金融/律所客户仍需显式说明缓存策略。
+- **标签/领域推断是关键词规则**：frontmatter + 路径/文件名 + 内容关键词，单篇最多 6 个标签；可能误标/漏标，未接入行业词表校验。
+- **负责人只读 frontmatter**：`owner/负责人` 字段；ACL、最后访问、文档系统 owner 未接。
+- **断链只查 Markdown 本地相对链接**：跳过 http/https、`file:`/`javascript:` 等 URI scheme 和 `@ref` 占位符；不查外链、仓库外目标、非 Markdown 引用。
+- **性能基线只是合成 5k**：`tools/bench_scan.py` 本机 5k 文本约 scan 9.6s / analyze 1.9s；50k 与真实 PDF/Office 语料未测。
+- **时效判断不可靠**：文件系统 mtime 在复制/迁移后会重置；报告会提示“仅供参考”，后续接入文档系统 `updated`。
+- **PDF/docx 标签不在文件内**：推断可补一部分，但二进制标签仍不如文本 frontmatter 可靠。
+- **行业包未加载**：`packs/generic` 目前是配置草案，v0.4 起生效。
+- **许可证暂用 MIT**：待 Q10 决策后可能切 Apache-2.0。
+
+## 下一步
+
+1. **P1 剩余准确性**：逐段正文语义 diff、ACL/文档系统 owner、标签误报治理与行业词表；用 3 个外部语料验收抽取 ≥90%、近重复召回 ≥80%。
+2. **报告可验收**：PDF 报告、报告哈希/签字页、执行摘要；默认脱敏分享。
+3. **工程门禁**：Windows CI 已在 `.gitcode/workflows/ci.yml` 增加；补 5 万文件性能基线、云占位/权限拒绝/Big5 fixtures。抽取并行与 sidecar TTL 已加。
+4. 按 [docs/paid-trial-runbook.md](docs/paid-trial-runbook.md) 走路径 B：免费包 → 过审 → 收入变现 → ¥1 付费技能验证。
+5. 按 [docs/interview-guide.md](docs/interview-guide.md) 约 10 人访谈；3 个外部语料跑通前不卖“完整审阅”。
+6. 连接器提审；根据 ¥1 验证结果决定付费报告形态。

kba_cli-0.3.0rc4.dist-info/RECORD

@@ -0,0 +1,22 @@
+kba/__init__.py,sha256=03x9LGTV8b9fEQFcQMmZhL-fxVu9wIUe0dQqhd6_lGw,72
+kba/__main__.py,sha256=4JMK66Wj4uLZTKbF-sT3LAxOsr6buig77PmOkJCRRxw,83
+kba/analyze.py,sha256=E0u8iLFaw7KdthGEQ1pLELJJcDW85VzAbwY-xk1bhPE,42172
+kba/cli.py,sha256=WlzI-SR7Huney9Mqp1Uywe9yVCt68PHZEH0SO3xqRxw,21925
+kba/doctor.py,sha256=LTduI5k-0msoq1ghxoSjG__KwAp1g-eXR0SzSWg5lFQ,2591
+kba/eval.py,sha256=07seiG3rDiAwoeep3HMHXHTaT3ekcqmJYE6mRBJAh9Y,13355
+kba/extract.py,sha256=5hBciTJzFNQnDUPylSSqqjZrD7UMETW7loHdXKXXzII,18313
+kba/fingerprint.py,sha256=1zB8mNNLio-9q2ZsNn5gknK30PQRxr2J9PkJ-83Ezxs,12963
+kba/jobs.py,sha256=g3nNRNshYUdpgea4zTHpUuTQK16TmskCYrPwh6MouTM,1780
+kba/links.py,sha256=23uqOZzEp07YjCweyZe7UUA6Hzw4pkJHYuWj_m-iMKQ,1435
+kba/pii.py,sha256=ch7Fk26dq2Rkevh3K8BBuswMGKO0brJiQGMv_D2wTm4,1348
+kba/report.py,sha256=CFmdHyNFCrs3hRP_o58RKi8Js3K9iLJzZtHKMekwYz0,32286
+kba/summary.py,sha256=Ap5HU7IoLrtbr5Jvm-pfjpdZUoEclaRvwPlgoJcp4lg,2245
+kba/tags.py,sha256=SsYf1BmR0XWDSQ5ocSOKAMx8yGM4j910Y8C0IiUxwHQ,5200
+kba/telemetry.py,sha256=L5zllit89CiwCFwBgCa8e6K0wBQOeM7euGqRXr3q1Ig,2121
+kba/util.py,sha256=wFrqY_RWlZSmTJz_KSDOP4Qti2omyzrJwU0AblrqsXM,6485
+kba_cli-0.3.0rc4.dist-info/licenses/LICENSE,sha256=u8T5zOdeAeV1yV7GwyCHuclCloDGM2XbUNebaUPDntA,1060
+kba_cli-0.3.0rc4.dist-info/METADATA,sha256=Z0mdb2_0w3EvsMMkTSYNLuAGPUN9ZYWj7CYP9AHSk7k,11346
+kba_cli-0.3.0rc4.dist-info/WHEEL,sha256=mOdXR9C3wMVov7cSY8PY-XCzo3SmmWCHvGQr5csJoao,91
+kba_cli-0.3.0rc4.dist-info/entry_points.txt,sha256=n_iS4p6qLR1TPczTuiCs9knjV8Fwsp2dgG1i056JlKg,37
+kba_cli-0.3.0rc4.dist-info/top_level.txt,sha256=dQToisLm3VfZPs8NeWqpwJnLQUUNsyVvm0Q1E_sNB3U,4
+kba_cli-0.3.0rc4.dist-info/RECORD,,

kba_cli-0.3.0rc4.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (84.2.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

kba_cli-0.3.0rc4.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+kba = kba.cli:main

kba_cli-0.3.0rc4.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 KBA
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

kba_cli-0.3.0rc4.dist-info/top_level.txt

@@ -0,0 +1 @@
+kba