feat(tools-pve/asr-pipeline): SQLite model config, task delete/rerefine, multi-recording review, lexicon CRUD

servers/tools-pve/asr-pipeline/asrpipe/config.py

@@ -1,22 +1,31 @@
 #!/usr/bin/env python3
 """Configuration for the standalone ASR post-processing pipeline and web service.
 
-This program has **no dependency on Claude Code, opencode, or any agent CLI**.
+Model endpoints, provider selections, and API keys are stored in SQLite
-It talks to LiteLLM / model providers over plain HTTPS and does its own file
+(`app_setting` table in `ASR_STATE_DIR/intake.db`) so they can be managed
-and git handling, so it can run inside Docker (via Arcane), systemd, or a shell.
+dynamically from the Web UI without restarting containers. Environment variables
-
+serve as initial bootstrap defaults when a setting is not yet in SQLite.
-Everything is overridable by environment variable.
 """
 from __future__ import annotations
 
 import os
+import sqlite3
+import threading
+import time
 from dataclasses import dataclass, field
 from pathlib import Path
+from typing import Any, Callable
 
 PKG = Path(__file__).resolve().parent
 ROOT = PKG.parent
 OBSIDIAN = Path(os.environ.get("ASR_WORKSPACE", str(ROOT)))
 
+_CACHE_LOCK = threading.Lock()
+_SETTINGS_CACHE: dict[str, str] = {}
+_SETTINGS_TS: float = 0.0
+_SETTINGS_DB_PATH: Path | None = None
+_CACHE_TTL = 1.0
+
 
 def _env(name: str, default: str) -> str:
     v = os.environ.get(name)
@@ -37,6 +46,165 @@ def _envi(name: str, default: int) -> int:
         return default
 
 
+def _state_db_path() -> Path:
+    state_dir = Path(_env("ASR_STATE_DIR", str(ROOT / "state")))
+    return state_dir / "intake.db"
+
+
+def _ensure_settings_table(con: sqlite3.Connection) -> None:
+    con.execute("""
+      CREATE TABLE IF NOT EXISTS app_setting(
+        key TEXT PRIMARY KEY,
+        value TEXT NOT NULL DEFAULT '',
+        updated_at REAL NOT NULL
+      )
+    """)
+
+
+def load_db_settings(force: bool = False) -> dict[str, str]:
+    """Read `app_setting` rows from `intake.db` with a 1s in-memory cache."""
+    global _SETTINGS_CACHE, _SETTINGS_TS, _SETTINGS_DB_PATH
+    db_path = _state_db_path()
+    now = time.time()
+    with _CACHE_LOCK:
+        if (
+            not force
+            and _SETTINGS_DB_PATH == db_path
+            and (now - _SETTINGS_TS) < _CACHE_TTL
+        ):
+            return dict(_SETTINGS_CACHE)
+        if not db_path.exists():
+            _SETTINGS_CACHE = {}
+            _SETTINGS_TS = now
+            _SETTINGS_DB_PATH = db_path
+            return {}
+        try:
+            con = sqlite3.connect(db_path, timeout=5)
+            con.row_factory = sqlite3.Row
+            _ensure_settings_table(con)
+            rows = con.execute("SELECT key, value FROM app_setting").fetchall()
+            con.close()
+            _SETTINGS_CACHE = {str(r["key"]): str(r["value"]) for r in rows}
+            _SETTINGS_TS = now
+            _SETTINGS_DB_PATH = db_path
+            return dict(_SETTINGS_CACHE)
+        except Exception:
+            return dict(_SETTINGS_CACHE)
+
+
+def get_setting(key: str, default: str = "") -> str:
+    data = load_db_settings()
+    val = data.get(key)
+    return val if val not in (None, "") else default
+
+
+def set_settings(updates: dict[str, Any]) -> None:
+    """Persist multiple settings into `intake.db` and invalidate cache."""
+    global _SETTINGS_TS
+    db_path = _state_db_path()
+    db_path.parent.mkdir(parents=True, exist_ok=True)
+    now = time.time()
+    con = sqlite3.connect(db_path, timeout=15)
+    try:
+        con.execute("PRAGMA journal_mode=WAL")
+        _ensure_settings_table(con)
+        for k, v in updates.items():
+            if v is None:
+                continue
+            con.execute(
+                "INSERT INTO app_setting(key, value, updated_at) VALUES(?,?,?) "
+                "ON CONFLICT(key) DO UPDATE SET value=excluded.value, updated_at=excluded.updated_at",
+                (str(k), str(v), now),
+            )
+        con.commit()
+    finally:
+        con.close()
+    with _CACHE_LOCK:
+        _SETTINGS_TS = 0.0
+    if "CFG" in globals():
+        for k in updates:
+            CFG.__dict__.pop(f"_override_{k}", None)
+    load_db_settings(force=True)
+
+
+def bootstrap_default_settings() -> None:
+    """Seed `app_setting` from environment variables for keys not yet stored."""
+    db_path = _state_db_path()
+    db_path.parent.mkdir(parents=True, exist_ok=True)
+    now = time.time()
+    defaults = {
+        "asr_provider": _env("ASR_PROVIDER", _env("MEETING_ASR_BACKEND", "qwen3cam")).lower(),
+        "diarize_url": _env("ASR_DIARIZE_URL", _env("MEETING_DIARIZE_URL", "http://192.168.50.100:10099/asr")),
+        "voiceprint_enroll_url": _env("ASR_VOICEPRINT_ENROLL_URL", "http://192.168.50.100:10099/voiceprints/enroll"),
+        "litellm_base_url": _env("LITELLM_BASE_URL", _env("ASR_LLM_BASE_URL", "http://192.168.50.49:4000/v1")).rstrip("/"),
+        "litellm_api_key": _env("LITELLM_API_KEY", _env("ASR_LLM_API_KEY", _env("OPENAI_API_KEY", ""))),
+        "clean_model": _env("ASR_CLEAN_MODEL", "aistudio/gemini-flash-lite-latest"),
+        "keyword_model": _env("ASR_KEYWORD_MODEL", "aistudio/gemini-flash-lite-latest"),
+        "draft_model": _env("ASR_DRAFT_MODEL", "codex/gpt-sol-latest"),
+        "gemini_asr_model": _env("ASR_GEMINI_MODEL", "aistudio/gemini-3.5-transcribe"),
+        "openai_asr_model": _env("ASR_OPENAI_MODEL", "openai/gpt-4o-transcribe-diarize"),
+        "soniox_api_base": _env("SONIOX_API_BASE", "https://api.soniox.com/v1").rstrip("/"),
+        "soniox_api_key": _env("SONIOX_API_KEY", ""),
+        "soniox_model": _env("SONIOX_MODEL", "stt-async-v5"),
+        "grok_api_base": _env("XAI_API_BASE", "https://api.x.ai/v1").rstrip("/"),
+        "grok_api_key": _env("XAI_API_KEY", _env("GROK_API_KEY", "")),
+        "grok_asr_model": _env("ASR_GROK_MODEL", "grok-voice-transcribe-2.0"),
+        "hindsight_api": _env("HINDSIGHT_API", "http://192.168.50.47:8888").rstrip("/"),
+        "hindsight_bank": _env("HINDSIGHT_BANK", "personal-context"),
+        "use_hindsight": "1" if _env("ASR_HINDSIGHT", "1") != "0" else "0",
+        "use_termbank": "1" if _env("ASR_TERMBANK", "1") != "0" else "0",
+        "git_sync": "1" if _env("ASR_GIT_SYNC", "0") != "0" else "0",
+    }
+    con = sqlite3.connect(db_path, timeout=15)
+    try:
+        con.execute("PRAGMA journal_mode=WAL")
+        _ensure_settings_table(con)
+        for k, v in defaults.items():
+            con.execute(
+                "INSERT OR IGNORE INTO app_setting(key, value, updated_at) VALUES(?,?,?)",
+                (k, str(v), now),
+            )
+            # If key was seeded empty earlier but env now provides a non-empty secret, populate it
+            if v and k.endswith("_api_key"):
+                con.execute(
+                    "UPDATE app_setting SET value=?, updated_at=? WHERE key=? AND value=''",
+                    (str(v), now, k),
+                )
+        con.commit()
+    finally:
+        con.close()
+    load_db_settings(force=True)
+
+
+class _SettingAttr:
+    """Descriptor that resolves from instance override -> SQLite `app_setting` -> env fallback."""
+
+    def __init__(
+        self,
+        key: str,
+        fallback: Callable[[], str],
+        transform: Callable[[str], Any] | None = None,
+    ) -> None:
+        self.key = key
+        self.fallback = fallback
+        self.transform = transform
+        self.override_attr = f"_override_{key}"
+
+    def __get__(self, instance: Any, owner: Any = None) -> Any:
+        if instance is None:
+            return self
+        if self.override_attr in instance.__dict__:
+            return instance.__dict__[self.override_attr]
+        raw = get_setting(self.key, self.fallback())
+        return self.transform(raw) if self.transform else raw
+
+    def __set__(self, instance: Any, value: Any) -> None:
+        instance.__dict__[self.override_attr] = value
+
+    def __delete__(self, instance: Any) -> None:
+        instance.__dict__.pop(self.override_attr, None)
+
+
 @dataclass
 class Config:
     # ---- paths ----------------------------------------------------------
@@ -46,61 +214,130 @@ class Config:
         _env("ASR_WORKSPACE", str(ROOT))))
     root: Path = ROOT
 
-    # ---- LiteLLM / LLM routing (ZenMux deprecated) ----------------------
+    # ---- SQLite-backed dynamic model & provider settings ----------------
-    litellm_base_url: str = field(default_factory=lambda: _env(
+    litellm_base_url = _SettingAttr(
-        "LITELLM_BASE_URL",
+        "litellm_base_url",
-        _env("ASR_LLM_BASE_URL", "http://192.168.50.49:4000/v1")).rstrip("/"))
+        lambda: _env("LITELLM_BASE_URL", _env("ASR_LLM_BASE_URL", "http://192.168.50.49:4000/v1")),
+        lambda s: s.rstrip("/"),
+    )
+    clean_model = _SettingAttr(
+        "clean_model",
+        lambda: _env("ASR_CLEAN_MODEL", "aistudio/gemini-flash-lite-latest"),
+    )
+    draft_model = _SettingAttr(
+        "draft_model",
+        lambda: _env("ASR_DRAFT_MODEL", "codex/gpt-sol-latest"),
+    )
+    keyword_model = _SettingAttr(
+        "keyword_model",
+        lambda: _env("ASR_KEYWORD_MODEL", "aistudio/gemini-flash-lite-latest"),
+    )
 
-    clean_model: str = field(default_factory=lambda: _env(
+    asr_provider = _SettingAttr(
-        "ASR_CLEAN_MODEL", "aistudio/gemini-flash-lite-latest"))
+        "asr_provider",
-    draft_model: str = field(default_factory=lambda: _env(
+        lambda: _env("ASR_PROVIDER", _env("MEETING_ASR_BACKEND", "qwen3cam")),
-        "ASR_DRAFT_MODEL", "codex/gpt-sol-latest"))
+        lambda s: s.strip().lower() or "qwen3cam",
-    keyword_model: str = field(default_factory=lambda: _env(
+    )
-        "ASR_KEYWORD_MODEL", "aistudio/gemini-flash-lite-latest"))
+    diarize_url = _SettingAttr(
+        "diarize_url",
+        lambda: _env("ASR_DIARIZE_URL", _env("MEETING_DIARIZE_URL", "http://192.168.50.100:10099/asr")),
+    )
+    voiceprint_enroll_url = _SettingAttr(
+        "voiceprint_enroll_url",
+        lambda: _env("ASR_VOICEPRINT_ENROLL_URL", "http://192.168.50.100:10099/voiceprints/enroll"),
+    )
 
-    endpoint: str = field(default_factory=lambda: _env(
+    soniox_api_key = _SettingAttr(
-        "ASR_ENDPOINT",
+        "soniox_api_key",
-        _env("LITELLM_BASE_URL",
+        lambda: _env("SONIOX_API_KEY", ""),
-             _env("ASR_LLM_BASE_URL", "http://192.168.50.49:4000/v1")).rstrip("/")
+    )
-        + "/chat/completions"))
+    soniox_api_base = _SettingAttr(
-    transcriptions_endpoint: str = field(default_factory=lambda: _env(
+        "soniox_api_base",
-        "ASR_TRANSCRIPTIONS_ENDPOINT",
+        lambda: _env("SONIOX_API_BASE", "https://api.soniox.com/v1"),
-        _env("LITELLM_BASE_URL",
+        lambda s: s.rstrip("/"),
-             _env("ASR_LLM_BASE_URL", "http://192.168.50.49:4000/v1")).rstrip("/")
+    )
-        + "/audio/transcriptions"))
+    soniox_model = _SettingAttr(
+        "soniox_model",
+        lambda: _env("SONIOX_MODEL", "stt-async-v5"),
+    )
+
+    gemini_asr_model = _SettingAttr(
+        "gemini_asr_model",
+        lambda: _env("ASR_GEMINI_MODEL", "aistudio/gemini-3.5-transcribe"),
+    )
+    openai_asr_model = _SettingAttr(
+        "openai_asr_model",
+        lambda: _env("ASR_OPENAI_MODEL", "openai/gpt-4o-transcribe-diarize"),
+    )
+
+    grok_api_key = _SettingAttr(
+        "grok_api_key",
+        lambda: _env("XAI_API_KEY", _env("GROK_API_KEY", "")),
+    )
+    grok_api_base = _SettingAttr(
+        "grok_api_base",
+        lambda: _env("XAI_API_BASE", "https://api.x.ai/v1"),
+        lambda s: s.rstrip("/"),
+    )
+    grok_asr_model = _SettingAttr(
+        "grok_asr_model",
+        lambda: _env("ASR_GROK_MODEL", "grok-voice-transcribe-2.0"),
+    )
+
+    hindsight_api = _SettingAttr(
+        "hindsight_api",
+        lambda: _env("HINDSIGHT_API", "http://192.168.50.47:8888"),
+        lambda s: s.rstrip("/"),
+    )
+    hindsight_bank = _SettingAttr(
+        "hindsight_bank",
+        lambda: _env("HINDSIGHT_BANK", "personal-context"),
+    )
+    use_hindsight = _SettingAttr(
+        "use_hindsight",
+        lambda: _env("ASR_HINDSIGHT", "1"),
+        lambda s: str(s).lower() not in ("0", "false", "no", "off"),
+    )
+    use_termbank = _SettingAttr(
+        "use_termbank",
+        lambda: _env("ASR_TERMBANK", "1"),
+        lambda s: str(s).lower() not in ("0", "false", "no", "off"),
+    )
+    git_sync = _SettingAttr(
+        "git_sync",
+        lambda: _env("ASR_GIT_SYNC", "1"),
+        lambda s: str(s).lower() not in ("0", "false", "no", "off"),
+    )
+
+    @property
+    def endpoint(self) -> str:
+        if "_override_endpoint" in self.__dict__:
+            return self.__dict__["_override_endpoint"]
+        explicit = os.environ.get("ASR_ENDPOINT")
+        if explicit:
+            return explicit
+        return f"{self.litellm_base_url}/chat/completions"
+
+    @endpoint.setter
+    def endpoint(self, value: str) -> None:
+        self.__dict__["_override_endpoint"] = value
+
+    @property
+    def transcriptions_endpoint(self) -> str:
+        if "_override_transcriptions_endpoint" in self.__dict__:
+            return self.__dict__["_override_transcriptions_endpoint"]
+        explicit = os.environ.get("ASR_TRANSCRIPTIONS_ENDPOINT")
+        if explicit:
+            return explicit
+        return f"{self.litellm_base_url}/audio/transcriptions"
+
+    @transcriptions_endpoint.setter
+    def transcriptions_endpoint(self, value: str) -> None:
+        self.__dict__["_override_transcriptions_endpoint"] = value
 
     concurrency: int = field(default_factory=lambda: _envi("CONCURRENCY", 6))
     http_timeout: float = field(default_factory=lambda: _envf("HTTP_TIMEOUT", 420.0))
     max_retries: int = field(default_factory=lambda: _envi("MAX_RETRIES", 5))
-
-    # ---- ASR providers --------------------------------------------------
-    asr_provider: str = field(default_factory=lambda: _env(
-        "ASR_PROVIDER", _env("MEETING_ASR_BACKEND", "qwen3cam")).lower())
-    diarize_url: str = field(default_factory=lambda: _env(
-        "ASR_DIARIZE_URL",
-        _env("MEETING_DIARIZE_URL", "http://192.168.50.100:10099/asr")))
-    voiceprint_enroll_url: str = field(default_factory=lambda: _env(
-        "ASR_VOICEPRINT_ENROLL_URL",
-        "http://192.168.50.100:10099/voiceprints/enroll"))
-
-    soniox_api_key: str = field(default_factory=lambda: _env("SONIOX_API_KEY", ""))
-    soniox_api_base: str = field(default_factory=lambda: _env(
-        "SONIOX_API_BASE", "https://api.soniox.com/v1").rstrip("/"))
-    soniox_model: str = field(default_factory=lambda: _env(
-        "SONIOX_MODEL", "stt-async-v5"))
-
-    gemini_asr_model: str = field(default_factory=lambda: _env(
-        "ASR_GEMINI_MODEL", "aistudio/gemini-3.5-transcribe"))
-    openai_asr_model: str = field(default_factory=lambda: _env(
-        "ASR_OPENAI_MODEL", "openai/gpt-4o-transcribe-diarize"))
-
-    grok_api_key: str = field(default_factory=lambda: _env(
-        "XAI_API_KEY", _env("GROK_API_KEY", "")))
-    grok_api_base: str = field(default_factory=lambda: _env(
-        "XAI_API_BASE", "https://api.x.ai/v1").rstrip("/"))
-    grok_asr_model: str = field(default_factory=lambda: _env(
-        "ASR_GROK_MODEL", "grok-voice-transcribe-2.0"))
-
     segment_seconds: int = field(default_factory=lambda: _envi(
         "SEGMENT_SECONDS", int(_env("MEETING_ASR_SEGMENT_SECONDS", "540"))))
 
@@ -119,10 +356,6 @@ class Config:
         default_factory=lambda: _envf("GATE_MAX_RATIO", 1.35))
 
     # ---- knowledge sources ----------------------------------------------
-    use_termbank: bool = field(
-        default_factory=lambda: _env("ASR_TERMBANK", "1") != "0")
-    use_hindsight: bool = field(
-        default_factory=lambda: _env("ASR_HINDSIGHT", "1") != "0")
     use_keyword_recall: bool = field(
         default_factory=lambda: _env("ASR_KEYWORD_RECALL", "1") != "0")
     use_lexicon: bool = field(
@@ -131,10 +364,6 @@ class Config:
         default_factory=lambda: _env("ASR_NORMALIZE", "1") != "0")
     use_evidence: bool = field(
         default_factory=lambda: _env("ASR_EVIDENCE", "1") != "0")
-    hindsight_api: str = field(default_factory=lambda: _env(
-        "HINDSIGHT_API", "http://192.168.50.47:8888"))
-    hindsight_bank: str = field(default_factory=lambda: _env(
-        "HINDSIGHT_BANK", "personal-context"))
     hindsight_timeout: float = field(
         default_factory=lambda: _envf("HINDSIGHT_TIMEOUT", 12.0))
     flag_min_score: float = field(
@@ -154,13 +383,17 @@ class Config:
         default_factory=lambda: _envi("REVIEW_MAX_SECONDS", 7200))
 
     # ---- behaviour ------------------------------------------------------
-    git_sync: bool = field(
-        default_factory=lambda: _env("ASR_GIT_SYNC", "1") != "0")
     notify: bool = field(
         default_factory=lambda: _env("ASR_NOTIFY", "1") != "0")
     cache_llm: bool = field(
         default_factory=lambda: _env("ASR_CACHE", "0") != "0")
 
+    def __post_init__(self) -> None:
+        # Clear dataclass-populated default descriptors so __get__ is invoked dynamically
+        for k in list(self.__dict__.keys()):
+            if isinstance(getattr(type(self), k, None), _SettingAttr):
+                del self.__dict__[k]
+
     @property
     def inbox(self) -> Path:
         return self.vault / ".asr-inbox"
@@ -217,6 +450,9 @@ class Config:
         return Path(_env("ASR_ENV_FILE", str(ROOT / ".env")))
 
     def api_key(self) -> str:
+        db_key = get_setting("litellm_api_key", "")
+        if db_key:
+            return db_key
         for k in ("LITELLM_API_KEY", "ASR_LLM_API_KEY", "OPENAI_API_KEY",
                   "ANTHROPIC_AUTH_TOKEN"):
             v = os.environ.get(k)
@@ -233,8 +469,7 @@ class Config:
                         if val:
                             return val
         raise RuntimeError(
-            "no API key: set LITELLM_API_KEY / ASR_LLM_API_KEY / OPENAI_API_KEY "
+            "no API key: configure LiteLLM API Key in /settings or set LITELLM_API_KEY")
-            f"or provide {p}")
 
 
 CFG = Config()

servers/tools-pve/asr-pipeline/asrpipe/lexicon.py

@@ -298,7 +298,7 @@ def sync_from_vault(force: bool = False) -> dict:
     now = time.time()
     n_ent = n_alias = 0
     con.execute("DELETE FROM alias WHERE origin IN ('card','glossary')")
-    con.execute("DELETE FROM entity")
+    con.execute("DELETE FROM entity WHERE card <> ''")
     con.execute("DELETE FROM relation WHERE source LIKE 'card:%'")
     for sub, kind in (("People", "人名"), ("Companies", "公司"), ("Projects", "项目")):
         d = CFG.vault / sub
@@ -524,3 +524,261 @@ def stats() -> dict:
     d["synced_at"] = r["v"] if r else ""
     con.close()
     return d
+
+
+# ------------------------------------------------------------------ Web UI CRUD
+def list_aliases(q: str = "", origin_filter: str = "", limit: int = 200) -> list[dict]:
+    con = connect()
+    sql = "SELECT * FROM alias WHERE 1=1"
+    args: list = []
+    if q.strip():
+        pat = f"%{q.strip()}%"
+        sql += " AND (alias LIKE ? OR canonical LIKE ? OR note LIKE ?)"
+        args.extend([pat, pat, pat])
+    if origin_filter == "learned":
+        sql += " AND origin <> 'card'"
+    elif origin_filter == "card":
+        sql += " AND origin = 'card'"
+    sql += " ORDER BY confirmed DESC, created_at DESC, alias ASC LIMIT ?"
+    args.append(limit)
+    rows = [dict(r) for r in con.execute(sql, args).fetchall()]
+    con.close()
+    return rows
+
+
+def update_alias(
+    old_alias: str,
+    old_canonical: str,
+    new_alias: str,
+    new_canonical: str,
+    *,
+    kind: str = "",
+    confirmed: int = 1,
+    note: str = "",
+) -> bool:
+    con = connect()
+    row = con.execute(
+        "SELECT * FROM alias WHERE alias=? AND canonical=?",
+        (old_alias, old_canonical),
+    ).fetchone()
+    origin = row["origin"] if row else "manual"
+    hits = row["hits"] if row else 0
+    created_at = row["created_at"] if row else time.time()
+    if not kind:
+        er = con.execute("SELECT kind FROM entity WHERE canonical=?", (new_canonical,)).fetchone()
+        kind = er["kind"] if er else (row["kind"] if row else guess_kind(new_canonical))
+    con.execute(
+        "DELETE FROM alias WHERE alias=? AND canonical=?",
+        (old_alias, old_canonical),
+    )
+    con.execute(
+        "INSERT OR REPLACE INTO alias(alias,canonical,kind,origin,confirmed,hits,created_at,note) "
+        "VALUES(?,?,?,?,?,?,?,?)",
+        (new_alias, new_canonical, kind, origin, int(bool(confirmed)), hits, created_at, note),
+    )
+    con.commit()
+    con.close()
+    return True
+
+
+def delete_alias(alias: str, canonical: str) -> bool:
+    con = connect()
+    cur = con.execute("DELETE FROM alias WHERE alias=? AND canonical=?", (alias, canonical))
+    con.commit()
+    deleted = cur.rowcount > 0
+    con.close()
+    return deleted
+
+
+def list_corrections(q: str = "", limit: int = 200) -> list[dict]:
+    con = connect()
+    sql = "SELECT * FROM correction WHERE 1=1"
+    args: list = []
+    if q.strip():
+        pat = f"%{q.strip()}%"
+        sql += " AND (surface LIKE ? OR canonical LIKE ? OR note LIKE ?)"
+        args.extend([pat, pat, pat])
+    sql += " ORDER BY confirmed DESC, n DESC, last_seen DESC LIMIT ?"
+    args.append(limit)
+    rows = [dict(r) for r in con.execute(sql, args).fetchall()]
+    con.close()
+    return rows
+
+
+def update_correction(
+    old_surface: str,
+    old_canonical: str,
+    new_surface: str,
+    new_canonical: str,
+    *,
+    confirmed: int = 1,
+    note: str = "",
+) -> bool:
+    con = connect()
+    row = con.execute(
+        "SELECT * FROM correction WHERE surface=? AND canonical=?",
+        (old_surface, old_canonical),
+    ).fetchone()
+    n = row["n"] if row else 1
+    now = time.time()
+    con.execute("DELETE FROM correction WHERE surface=? AND canonical=?", (old_surface, old_canonical))
+    con.execute(
+        "DELETE FROM phonetic_confusion WHERE heard=? AND intended=?",
+        (old_surface, old_canonical),
+    )
+    con.execute(
+        "INSERT OR REPLACE INTO correction(surface,canonical,n,confirmed,last_seen,note) "
+        "VALUES(?,?,?,?,?,?)",
+        (new_surface, new_canonical, n, int(bool(confirmed)), now, note),
+    )
+    con.execute(
+        "INSERT OR REPLACE INTO phonetic_confusion(heard,intended,n,confirmed,last_seen,note) "
+        "VALUES(?,?,?,?,?,?)",
+        (new_surface, new_canonical, n, int(bool(confirmed)), now, note),
+    )
+    con.commit()
+    con.close()
+    return True
+
+
+def delete_correction(surface: str, canonical: str) -> bool:
+    con = connect()
+    cur = con.execute("DELETE FROM correction WHERE surface=? AND canonical=?", (surface, canonical))
+    con.execute("DELETE FROM phonetic_confusion WHERE heard=? AND intended=?", (surface, canonical))
+    con.commit()
+    deleted = cur.rowcount > 0
+    con.close()
+    return deleted
+
+
+def list_entities(q: str = "", kind: str = "", limit: int = 200) -> list[dict]:
+    con = connect()
+    sql = "SELECT * FROM entity WHERE 1=1"
+    args: list = []
+    if q.strip():
+        pat = f"%{q.strip()}%"
+        sql += " AND (canonical LIKE ? OR company LIKE ? OR role LIKE ? OR related LIKE ?)"
+        args.extend([pat, pat, pat, pat])
+    if kind.strip():
+        sql += " AND kind=?"
+        args.append(kind.strip())
+    sql += " ORDER BY (CASE WHEN card='' THEN 0 ELSE 1 END), canonical ASC LIMIT ?"
+    args.append(limit)
+    rows = [dict(r) for r in con.execute(sql, args).fetchall()]
+    con.close()
+    return rows
+
+
+def upsert_entity(
+    canonical: str,
+    kind: str = "人名",
+    *,
+    company: str = "",
+    role: str = "",
+    related: str = "",
+    card: str = "",
+) -> None:
+    con = connect()
+    con.execute(
+        "INSERT OR REPLACE INTO entity(canonical,kind,company,role,related,card,synced_at) "
+        "VALUES(?,?,?,?,?,?,?)",
+        (canonical, kind or guess_kind(canonical), company, role, related, card, time.time()),
+    )
+    con.commit()
+    con.close()
+
+
+def update_entity(
+    old_canonical: str,
+    new_canonical: str,
+    kind: str,
+    *,
+    company: str = "",
+    role: str = "",
+    related: str = "",
+) -> bool:
+    con = connect()
+    row = con.execute("SELECT * FROM entity WHERE canonical=?", (old_canonical,)).fetchone()
+    card = row["card"] if row else ""
+    con.execute("DELETE FROM entity WHERE canonical=?", (old_canonical,))
+    con.execute(
+        "INSERT OR REPLACE INTO entity(canonical,kind,company,role,related,card,synced_at) "
+        "VALUES(?,?,?,?,?,?,?)",
+        (new_canonical, kind or guess_kind(new_canonical), company, role, related, card, time.time()),
+    )
+    if old_canonical != new_canonical:
+        con.execute("UPDATE alias SET canonical=? WHERE canonical=?", (new_canonical, old_canonical))
+        con.execute("UPDATE correction SET canonical=? WHERE canonical=?", (new_canonical, old_canonical))
+    con.commit()
+    con.close()
+    return True
+
+
+def delete_entity(canonical: str) -> bool:
+    con = connect()
+    cur = con.execute("DELETE FROM entity WHERE canonical=?", (canonical,))
+    con.commit()
+    deleted = cur.rowcount > 0
+    con.close()
+    return deleted
+
+
+def list_relations(q: str = "", limit: int = 200) -> list[dict]:
+    con = connect()
+    sql = "SELECT * FROM relation WHERE 1=1"
+    args: list = []
+    if q.strip():
+        pat = f"%{q.strip()}%"
+        sql += " AND (subject LIKE ? OR predicate LIKE ? OR object LIKE ? OR source LIKE ?)"
+        args.extend([pat, pat, pat, pat])
+    sql += " ORDER BY active DESC, confidence DESC, last_seen DESC LIMIT ?"
+    args.append(limit)
+    rows = [dict(r) for r in con.execute(sql, args).fetchall()]
+    con.close()
+    return rows
+
+
+def update_relation(
+    old_subject: str,
+    old_predicate: str,
+    old_object: str,
+    old_source: str,
+    subject: str,
+    predicate: str,
+    object_: str,
+    *,
+    confidence: float = 0.8,
+    active: int = 1,
+) -> bool:
+    con = connect()
+    row = con.execute(
+        "SELECT * FROM relation WHERE subject=? AND predicate=? AND object=? AND source=?",
+        (old_subject, old_predicate, old_object, old_source),
+    ).fetchone()
+    first_seen = row["first_seen"] if row else time.time()
+    source = old_source or "manual"
+    con.execute(
+        "DELETE FROM relation WHERE subject=? AND predicate=? AND object=? AND source=?",
+        (old_subject, old_predicate, old_object, old_source),
+    )
+    con.execute(
+        "INSERT OR REPLACE INTO relation(subject,predicate,object,confidence,source,first_seen,last_seen,active) "
+        "VALUES(?,?,?,?,?,?,?,?)",
+        (subject, predicate, object_, float(confidence), source, first_seen, time.time(), int(bool(active))),
+    )
+    con.commit()
+    con.close()
+    return True
+
+
+def delete_relation(subject: str, predicate: str, object_: str, source: str) -> bool:
+    con = connect()
+    cur = con.execute(
+        "DELETE FROM relation WHERE subject=? AND predicate=? AND object=? AND source=?",
+        (subject, predicate, object_, source),
+    )
+    con.commit()
+    deleted = cur.rowcount > 0
+    con.close()
+    return deleted
+

servers/tools-pve/asr-pipeline/asrpipe/review.py

@@ -307,14 +307,14 @@ def _request_revision(draft_path: str) -> bool:
         return False
 
 
-def answer(question_id: int, value: str) -> dict:
+def answer(question_id: int, value: str, *, allow_update: bool = False) -> dict:
     con = lexicon.connect()
     row = con.execute("SELECT * FROM review_question WHERE id=?",
                       (question_id,)).fetchone()
     if not row:
         con.close()
         raise KeyError(f"review question {question_id} not found")
-    if row["status"] != "pending":
+    if row["status"] != "pending" and not allow_update:
         result = {
             "id": question_id, "answer": row["answer"], "learned": False,
             "already_answered": True,
@@ -376,3 +376,47 @@ def answer(question_id: int, value: str) -> dict:
         "draft_feedback_saved": saved, "voiceprint_enrolled": voiceprint_enrolled,
         "revision_started": revision_started,
     }
+
+
+def recent_answered(limit: int = 30) -> list[dict]:
+    con = lexicon.connect()
+    rows = con.execute(
+        "SELECT * FROM review_question WHERE status='answered' "
+        "ORDER BY answered_at DESC, id DESC LIMIT ?",
+        (limit,),
+    ).fetchall()
+    con.close()
+    return [dict(r) for r in rows]
+
+
+def reopen_question(question_id: int) -> bool:
+    con = lexicon.connect()
+    cur = con.execute(
+        "UPDATE review_question SET status='pending', answer='', answered_at=NULL WHERE id=?",
+        (question_id,),
+    )
+    con.commit()
+    ok = cur.rowcount > 0
+    con.close()
+    return ok
+
+
+def delete_for_job(transcript_path: str = "", draft_path: str = "", job_id: str = "") -> int:
+    con = lexicon.connect()
+    rows = con.execute("SELECT id, recording, payload FROM review_question").fetchall()
+    to_delete: list[int] = []
+    for r in rows:
+        payload = json.loads(r["payload"] or "{}")
+        if (
+            (transcript_path and r["recording"] == transcript_path)
+            or (draft_path and payload.get("draft_path") == draft_path)
+            or (job_id and payload.get("job_id") == job_id)
+        ):
+            to_delete.append(int(r["id"]))
+    if to_delete:
+        marks = ",".join("?" for _ in to_delete)
+        con.execute(f"DELETE FROM review_question WHERE id IN ({marks})", to_delete)
+        con.commit()
+    con.close()
+    return len(to_delete)
+

servers/tools-pve/asr-pipeline/asrservice/db.py

@@ -4,15 +4,19 @@ import sqlite3
 import time
 from typing import Any
 
-from .config import DB, DEFAULT_PROVIDER, ensure_dirs
+from asrpipe.config import CFG, bootstrap_default_settings, get_setting, load_db_settings, set_settings
+from .config import DB, ensure_dirs
 
 STATUSES = {
     "uploading", "queued", "transcribing", "refining", "review",
     "publishing", "complete", "failed",
 }
 
+_BOOTSTRAPPED_DB: str = ""
+
 
 def connect() -> sqlite3.Connection:
+    global _BOOTSTRAPPED_DB
     ensure_dirs()
     con = sqlite3.connect(DB, timeout=30)
     con.row_factory = sqlite3.Row
@@ -37,8 +41,14 @@ def connect() -> sqlite3.Connection:
         status TEXT NOT NULL, detail TEXT NOT NULL DEFAULT '', created_at REAL NOT NULL
       );
       CREATE INDEX IF NOT EXISTS ix_job_event_job ON job_event(job_id,id);
+      CREATE TABLE IF NOT EXISTS app_setting(
+        key TEXT PRIMARY KEY,
+        value TEXT NOT NULL DEFAULT '',
+        updated_at REAL NOT NULL
+      );
     """)
     cols = {r[1] for r in con.execute("PRAGMA table_info(intake_job)")}
+    default_prov = CFG.asr_provider or "qwen3cam"
     additions = {
         "draft": "TEXT NOT NULL DEFAULT ''",
         "result": "TEXT NOT NULL DEFAULT ''",
@@ -46,13 +56,19 @@ def connect() -> sqlite3.Connection:
         "sha256": "TEXT NOT NULL DEFAULT ''",
         "attempts": "INTEGER NOT NULL DEFAULT 0",
         "error_code": "TEXT NOT NULL DEFAULT ''",
-        "asr_provider": f"TEXT NOT NULL DEFAULT '{DEFAULT_PROVIDER}'",
+        "asr_provider": f"TEXT NOT NULL DEFAULT '{default_prov}'",
         "asr_model": "TEXT NOT NULL DEFAULT ''",
     }
     for name, ddl in additions.items():
         if name not in cols:
             con.execute(f"ALTER TABLE intake_job ADD COLUMN {name} {ddl}")
     con.commit()
+    if _BOOTSTRAPPED_DB != str(DB):
+        _BOOTSTRAPPED_DB = str(DB)
+        try:
+            bootstrap_default_settings()
+        except Exception:
+            pass
     return con
 
 
@@ -69,7 +85,7 @@ def create_job(
     asr_model: str = "",
 ) -> dict[str, Any]:
     now = time.time()
-    prov = (asr_provider or DEFAULT_PROVIDER or "qwen3cam").strip().lower()
+    prov = (asr_provider or CFG.asr_provider or "qwen3cam").strip().lower()
     with connect() as con:
         con.execute(
             "INSERT INTO intake_job(id,original_name,stored_name,title,context,size,"
@@ -88,7 +104,7 @@ def update_job(job_id: str, status: str, detail: str = "", **fields: Any) -> boo
         raise ValueError(f"invalid status: {status}")
     allowed = {
         "transcript", "draft", "result", "sha256", "attempts",
-        "error_code", "asr_provider", "asr_model",
+        "error_code", "asr_provider", "asr_model", "title", "context",
     }
     changes = {k: v for k, v in fields.items() if k in allowed and v is not None}
     sets = ["status=?", "updated_at=?", "detail=?"]
@@ -106,6 +122,44 @@ def update_job(job_id: str, status: str, detail: str = "", **fields: Any) -> boo
     return bool(cur.rowcount)
 
 
+def update_job_meta(
+    job_id: str,
+    *,
+    title: str | None = None,
+    context: str | None = None,
+    asr_provider: str | None = None,
+    asr_model: str | None = None,
+) -> bool:
+    sets = ["updated_at=?"]
+    values: list[Any] = [time.time()]
+    if title is not None:
+        sets.append("title=?")
+        values.append(title[:120])
+    if context is not None:
+        sets.append("context=?")
+        values.append(context[:2000])
+    if asr_provider is not None:
+        sets.append("asr_provider=?")
+        values.append(asr_provider[:40])
+    if asr_model is not None:
+        sets.append("asr_model=?")
+        values.append(asr_model[:80])
+    values.append(job_id)
+    with connect() as con:
+        cur = con.execute(f"UPDATE intake_job SET {','.join(sets)} WHERE id=?", values)
+    return bool(cur.rowcount)
+
+
+def delete_job(job_id: str) -> dict[str, Any] | None:
+    job = get_job(job_id)
+    if not job:
+        return None
+    with connect() as con:
+        con.execute("DELETE FROM job_event WHERE job_id=?", (job_id,))
+        con.execute("DELETE FROM intake_job WHERE id=?", (job_id,))
+    return job
+
+
 def get_job(job_id: str) -> dict[str, Any] | None:
     with connect() as con:
         row = con.execute("SELECT * FROM intake_job WHERE id=?", (job_id,)).fetchone()
@@ -120,13 +174,27 @@ def find_by_stored(stored_name: str) -> dict[str, Any] | None:
         return dict(row) if row else None
 
 
-def list_jobs(limit: int = 30, offset: int = 0, status: str = "") -> list[dict[str, Any]]:
+def list_jobs(
+    limit: int = 30,
+    offset: int = 0,
+    status: str = "",
+    q: str = "",
+    provider: str = "",
+) -> list[dict[str, Any]]:
     limit = min(max(limit, 1), 200)
     offset = max(offset, 0)
-    sql, args = "SELECT * FROM intake_job", []
+    sql = "SELECT * FROM intake_job WHERE 1=1"
+    args: list[Any] = []
     if status:
-        sql += " WHERE status=?"
+        sql += " AND status=?"
         args.append(status)
+    if provider:
+        sql += " AND asr_provider=?"
+        args.append(provider)
+    if q.strip():
+        pat = f"%{q.strip()}%"
+        sql += " AND (title LIKE ? OR original_name LIKE ? OR context LIKE ? OR detail LIKE ?)"
+        args.extend([pat, pat, pat, pat])
     sql += " ORDER BY created_at DESC LIMIT ? OFFSET ?"
     args.extend((limit, offset))
     with connect() as con:
@@ -145,3 +213,32 @@ def pending_recovery() -> list[dict[str, Any]]:
         return [dict(r) for r in con.execute(
             "SELECT * FROM intake_job WHERE status IN ('queued','transcribing','refining','publishing') "
             "ORDER BY created_at").fetchall()]
+
+
+def get_all_settings() -> dict[str, str]:
+    connect().close()
+    return load_db_settings(force=True)
+
+
+def save_settings(updates: dict[str, Any]) -> dict[str, str]:
+    connect().close()
+    set_settings(updates)
+    return load_db_settings(force=True)
+
+
+__all__ = [
+    "STATUSES",
+    "connect",
+    "create_job",
+    "update_job",
+    "update_job_meta",
+    "delete_job",
+    "get_job",
+    "find_by_stored",
+    "list_jobs",
+    "events",
+    "pending_recovery",
+    "get_all_settings",
+    "save_settings",
+    "get_setting",
+]

servers/tools-pve/asr-pipeline/asrservice/worker.py

@@ -4,6 +4,7 @@ import fcntl
 import hashlib
 import json
 import os
+import re
 import shutil
 import sqlite3
 import time
@@ -11,9 +12,10 @@ import uuid
 from pathlib import Path
 
 from asrpipe import graph, lexicon, providers, refine, review, revise, vault
+from asrpipe.config import CFG
 from . import db
 from .config import (
-    DATA, DEFAULT_PROVIDER, DRAFTS, FAILED, INBOX, LOGS, MAX_ATTEMPTS,
+    DATA, DRAFTS, FAILED, INBOX, LOGS, MAX_ATTEMPTS,
     POLL_SECONDS, PROCESSED, PROCESSING, PUBLIC_URL, REVISION_QUEUE,
     STABLE_SECONDS, STATE, SUPPORTED, TRANSCRIPTS, ensure_dirs,
 )
@@ -50,7 +52,7 @@ def transcribe(
 ) -> tuple[Path, providers.TranscribeResult]:
     res = providers.transcribe_audio(
         audio,
-        provider=provider or DEFAULT_PROVIDER,
+        provider=provider or CFG.asr_provider,
         model=model,
         title=title,
         user_context=user_context,
@@ -71,7 +73,7 @@ def transcribe(
 def process_audio(source: Path, job: dict) -> None:
     job_id = job["id"]
     attempts = int(job.get("attempts") or 0) + 1
-    prov = providers.normalize_provider_name(job.get("asr_provider") or DEFAULT_PROVIDER)
+    prov = providers.normalize_provider_name(job.get("asr_provider") or CFG.asr_provider)
     model = str(job.get("asr_model") or "")
     digest = sha256(source)
     prov_label = providers.PROVIDERS.get(prov, {}).get("label", prov)
@@ -120,6 +122,47 @@ def process_audio(source: Path, job: dict) -> None:
          f"{PUBLIC_URL}/jobs/{job_id}/review/")
 
 
+def rerefine_job(job: dict) -> None:
+    """Re-run LLM refinement on an existing raw transcript without re-running ASR."""
+    job_id = job["id"]
+    transcript = Path(str(job.get("transcript") or ""))
+    if not transcript.is_file():
+        raise FileNotFoundError(f"transcript file missing for job {job_id}: {transcript}")
+    raw = transcript.read_text(encoding="utf-8", errors="ignore")
+    intake = {
+        "job_id": job_id,
+        "title": job.get("title", ""),
+        "context": job.get("context", ""),
+        "source": job.get("source", ""),
+        "asr_provider": job.get("asr_provider") or CFG.asr_provider,
+        "asr_model": job.get("asr_model") or "",
+    }
+    intake_line = f"<!-- intake: {json.dumps(intake, ensure_ascii=False)} -->"
+    if re.match(r"^<!--\s*intake:\s*\{.*?\}\s*-->", raw):
+        raw = re.sub(r"^<!--\s*intake:\s*\{.*?\}\s*-->", intake_line, raw, count=1)
+    else:
+        raw = intake_line + "\n" + raw
+    transcript.write_text(raw, encoding="utf-8")
+
+    db.update_job(job_id, "refining", "正在基于现有转写重新生成精修草稿")
+    review.delete_for_job(
+        transcript_path=str(transcript),
+        draft_path=str(job.get("draft") or ""),
+        job_id=job_id,
+    )
+    result = refine.refine(
+        raw, date=time.strftime("%Y-%m-%d"),
+        transcript_path=str(transcript), queue_review=True,
+    )
+    draft = DRAFTS / f"{job_id}.md"
+    vault.write_draft_file(draft, result.draft)
+    questions = review.enqueue_draft(str(transcript), draft)
+    db.update_job(
+        job_id, "review", f"整理稿已重新生成，{questions} 个问题待确认",
+        transcript=str(transcript), draft=str(draft),
+    )
+
+
 def claim(path: Path) -> tuple[Path, dict]:
     job = db.find_by_stored(path.name)
     if not job:
@@ -127,7 +170,7 @@ def claim(path: Path) -> tuple[Path, dict]:
         job = db.create_job(
             job_id, path.name, path.name, title=path.stem,
             size=path.stat().st_size, source="ssh",
-            asr_provider=DEFAULT_PROVIDER,
+            asr_provider=CFG.asr_provider,
         )
     target = PROCESSING / path.name
     if target.exists():
@@ -164,8 +207,15 @@ def process_revisions() -> None:
     for marker in sorted(REVISION_QUEUE.glob("*.json")):
         try:
             payload = json.loads(marker.read_text(encoding="utf-8"))
-            draft = Path(payload["draft"])
+            action = str(payload.get("action") or "revise")
             job_id = str(payload.get("job_id") or "")
+            if action == "rerefine":
+                job = db.get_job(job_id)
+                if job:
+                    rerefine_job(job)
+                marker.unlink(missing_ok=True)
+                continue
+            draft = Path(payload["draft"])
             db.update_job(job_id, "publishing", "正在应用人工确认")
             revise.revise(draft)
             current = db.get_job(job_id) or {}
@@ -181,6 +231,14 @@ def process_revisions() -> None:
             marker.unlink(missing_ok=True)
         except Exception as exc:
             log(f"revision marker={marker.name} failed type={type(exc).__name__} err={exc}")
+            try:
+                payload = json.loads(marker.read_text(encoding="utf-8"))
+                jid = str(payload.get("job_id") or "")
+                if jid:
+                    db.update_job(jid, "failed", f"精修/修订失败：{exc}", error_code=type(exc).__name__)
+            except Exception:
+                pass
+            marker.unlink(missing_ok=True)
 
 
 def recover() -> None:

servers/tools-pve/asr-pipeline/tests/test_service.py

@@ -5,6 +5,7 @@ import os
 import tempfile
 import threading
 import unittest
+import urllib.parse
 import urllib.request
 from http.server import ThreadingHTTPServer
 from pathlib import Path
@@ -20,6 +21,7 @@ os.environ["ASR_NOTIFY"] = "0"
 os.environ["ASR_GIT_SYNC"] = "0"
 
 from asrpipe import lexicon, providers, review, segment  # noqa: E402
+from asrpipe.config import CFG  # noqa: E402
 from asrservice import config as svc_cfg, db as svc_db, web as svc_web  # noqa: E402
 
 
@@ -33,6 +35,28 @@ class ProviderAndServiceTests(unittest.TestCase):
     def tearDown(self):
         self.tmp.cleanup()
 
+    def test_sqlite_settings_override_config_dynamically(self):
+        svc_db.save_settings({
+            "asr_provider": "soniox",
+            "litellm_base_url": "https://api.opfo.xyz/v1",
+            "litellm_api_key": "sk-sqlite-dynamic-key",
+            "clean_model": "aistudio/gemini-flash-latest",
+            "draft_model": "openai/gpt-sol-latest",
+            "soniox_api_key": "soniox-db-secret",
+            "use_hindsight": "0",
+        })
+        self.assertEqual(CFG.asr_provider, "soniox")
+        self.assertEqual(CFG.litellm_base_url, "https://api.opfo.xyz/v1")
+        self.assertEqual(CFG.endpoint, "https://api.opfo.xyz/v1/chat/completions")
+        self.assertEqual(CFG.transcriptions_endpoint, "https://api.opfo.xyz/v1/audio/transcriptions")
+        self.assertEqual(CFG.api_key(), "sk-sqlite-dynamic-key")
+        self.assertEqual(CFG.clean_model, "aistudio/gemini-flash-latest")
+        self.assertEqual(CFG.draft_model, "openai/gpt-sol-latest")
+        self.assertEqual(CFG.soniox_api_key, "soniox-db-secret")
+        self.assertFalse(CFG.use_hindsight)
+        # Restore default for subsequent tests
+        svc_db.save_settings({"asr_provider": "qwen3cam", "use_hindsight": "1"})
+
     def test_soniox_token_diarization_and_segment_parse(self):
         payload = {
             "tokens": [
@@ -135,20 +159,76 @@ class ProviderAndServiceTests(unittest.TestCase):
         updated = draft.read_text(encoding="utf-8")
         self.assertIn("speaker_1: 张三", updated)
 
-    def test_db_and_web_endpoints(self):
+    def test_lexicon_crud_operations(self):
-        job_id = "0123456789abcdef01234567"
+        # Entity CRUD
-        job = svc_db.create_job(
+        lexicon.upsert_entity("测试生物", "公司", company="测试集团", role="子公司", related="李四")
-            job_id=job_id,
+        ents = lexicon.list_entities(q="测试生物")
-            original_name="meeting.m4a",
+        self.assertEqual(len(ents), 1)
-            stored_name=f"20261102-{job_id}.m4a",
+        self.assertEqual(ents[0]["company"], "测试集团")
-            title="架构周会",
+
-            context="验证 Web API",
+        lexicon.update_entity("测试生物", "测试生物科技", "公司", company="新集团", role="总部", related="李四,王五")
+        ents2 = lexicon.list_entities(q="测试生物科技")
+        self.assertEqual(len(ents2), 1)
+        self.assertEqual(ents2[0]["company"], "新集团")
+        self.assertTrue(lexicon.delete_entity("测试生物科技"))
+        self.assertEqual(len(lexicon.list_entities(q="测试生物科技")), 0)
+
+        # Alias & Correction CRUD
+        lexicon.learn_alias("老测试", "测试人", kind="人名", origin="manual", note="初始")
+        lexicon.update_alias("老测试", "测试人", "小测试", "测试人", kind="人名", confirmed=1, note="已改")
+        aliases = lexicon.list_aliases(q="小测试")
+        self.assertEqual(len(aliases), 1)
+        self.assertEqual(aliases[0]["note"], "已改")
+        self.assertTrue(lexicon.delete_alias("小测试", "测试人"))
+
+        lexicon.learn_correction("误识词", "正词", note="c1")
+        lexicon.update_correction("误识词", "正词", "新误识词", "正词", confirmed=1, note="c2")
+        cors = lexicon.list_corrections(q="新误识词")
+        self.assertEqual(len(cors), 1)
+        self.assertEqual(cors[0]["note"], "c2")
+        self.assertTrue(lexicon.delete_correction("新误识词", "正词"))
+
+        # Relation CRUD
+        lexicon.upsert_relation("张三", "任职于", "洛启生物", confidence=0.8, source="manual")
+        lexicon.update_relation("张三", "任职于", "洛启生物", "manual", "张三", "实控人", "洛启生物", confidence=0.95, active=1)
+        rels = lexicon.list_relations(q="张三")
+        self.assertTrue(any(r["predicate"] == "实控人" for r in rels))
+        self.assertTrue(lexicon.delete_relation("张三", "实控人", "洛启生物", "manual"))
+
+    def test_db_web_endpoints_multi_review_and_job_delete(self):
+        job1_id = "0123456789abcdef01234567"
+        job2_id = "fedcba987654321001234567"
+        t1 = svc_cfg.TRANSCRIPTS / f"{job1_id}.txt"
+        t2 = svc_cfg.TRANSCRIPTS / f"{job2_id}.txt"
+        t1.write_text("【speaker_0】录音一内容\n", encoding="utf-8")
+        t2.write_text("【speaker_0】录音二内容\n", encoding="utf-8")
+
+        svc_db.create_job(
+            job_id=job1_id,
+            original_name="meeting1.m4a",
+            stored_name=f"20261102-{job1_id}.m4a",
+            title="第一场尽调会",
+            context="验证多录音确认",
             size=2048,
             asr_provider="soniox",
             asr_model="stt-async-v5",
         )
-        self.assertEqual(job["asr_provider"], "soniox")
+        svc_db.update_job(job1_id, "review", "待确认", transcript=str(t1))
-        self.assertEqual(job["asr_model"], "stt-async-v5")
+
+        svc_db.create_job(
+            job_id=job2_id,
+            original_name="meeting2.m4a",
+            stored_name=f"20261102-{job2_id}.m4a",
+            title="第二场投决会",
+            context="验证批量操作",
+            size=4096,
+            asr_provider="qwen3cam",
+        )
+        svc_db.update_job(job2_id, "review", "待确认", transcript=str(t2))
+
+        q1 = review.enqueue(str(t1), "第一场问题A？", {"kind": "draft_question", "candidates": ["选项A1", "选项A2"]})
+        q2 = review.enqueue(str(t1), "第一场问题B？", {"kind": "draft_question", "candidates": ["选项B1"]})
+        q3 = review.enqueue(str(t2), "第二场问题C？", {"kind": "draft_question", "candidates": ["选项C1"]})
 
         svc_web.AGENT_TOKEN = "test-agent-token"
         server = ThreadingHTTPServer(("127.0.0.1", 0), svc_web.Handler)
@@ -164,6 +244,17 @@ class ProviderAndServiceTests(unittest.TestCase):
                 with urllib.request.urlopen(req, timeout=5) as resp:
                     return resp.status, resp.read().decode("utf-8")
 
+            def post_form(path: str, data: dict[str, str]) -> tuple[int, str]:
+                encoded = urllib.parse.urlencode(data).encode("utf-8")
+                req = urllib.request.Request(
+                    f"http://127.0.0.1:{port}{path}",
+                    data=encoded,
+                    headers={"Content-Type": "application/x-www-form-urlencoded"},
+                    method="POST",
+                )
+                with urllib.request.urlopen(req, timeout=5) as resp:
+                    return resp.status, resp.read().decode("utf-8")
+
             status, body = fetch("/health")
             self.assertEqual(status, 200)
             self.assertEqual(json.loads(body)["ok"], True)
@@ -173,22 +264,57 @@ class ProviderAndServiceTests(unittest.TestCase):
                 headers={"Authorization": "Bearer test-agent-token"},
             )
             self.assertEqual(status, 200)
-            prov_data = json.loads(body)
+            prov_ids = [p["id"] for p in json.loads(body)["providers"]]
-            prov_ids = [p["id"] for p in prov_data["providers"]]
             self.assertEqual(prov_ids, ["qwen3cam", "soniox", "gemini", "openai", "grok"])
 
-            status, body = fetch(f"/jobs/{job_id}/")
+            # Multi-recording review page shows both recordings grouped
+            status, body = fetch("/review/")
             self.assertEqual(status, 200)
-            self.assertIn("架构周会", body)
+            self.assertIn("第一场尽调会", body)
-            self.assertIn("soniox", body)
+            self.assertIn("第二场投决会", body)
+            self.assertIn("第一场问题A？", body)
+            self.assertIn("第二场问题C？", body)
 
-            status, body = fetch("/lexicon")
+            # Batch submit both questions for recording 1
+            status, body = post_form(f"/jobs/{job1_id}/review/", {
+                "question_ids": f"{q1},{q2}",
+                f"preset_{q1}": "选项A1",
+                f"custom_{q2}": "自定义回答B",
+                "batch_submit": "1",
+            })
             self.assertEqual(status, 200)
-            self.assertIn("词库", body)
+            self.assertEqual(len(svc_web.pending(svc_db.get_job(job1_id))), 0)
+            self.assertEqual(len(svc_web.pending(svc_db.get_job(job2_id))), 1)
 
-            status, body = fetch("/settings")
+            # Edit job metadata
+            status, body = post_form(f"/jobs/{job1_id}/meta", {
+                "title": "第一场尽调会（已改名）",
+                "context": "更新后的背景",
+                "provider": "openai",
+                "model": "openai/gpt-4o-transcribe-diarize",
+                "next_action": "save",
+            })
             self.assertEqual(status, 200)
-            self.assertIn("LiteLLM", body)
+            self.assertIn("第一场尽调会（已改名）", body)
+
+            # Save settings via Web UI POST
+            status, body = post_form("/settings", {
+                "litellm_base_url": "http://192.168.50.49:4000/v1",
+                "clean_model": "aistudio/gemini-flash-lite-latest",
+                "keyword_model": "aistudio/gemini-flash-lite-latest",
+                "draft_model": "codex/gpt-sol-latest",
+                "asr_provider": "soniox",
+                "use_hindsight": "1",
+            })
+            self.assertEqual(status, 200)
+            self.assertEqual(CFG.asr_provider, "soniox")
+
+            # Delete job 2 and verify its pending review question is also cleaned up
+            status, _ = post_form(f"/jobs/{job2_id}/delete", {"delete_files": "1"})
+            self.assertEqual(status, 200)
+            self.assertIsNone(svc_db.get_job(job2_id))
+            self.assertFalse(t2.exists())
+            self.assertNotIn(q3, [r["id"] for r in review.pending(200)])
         finally:
             server.shutdown()
             server.server_close()