superlocalmemory 4.2.4__tar.gz → 4.2.5__tar.gz

{superlocalmemory-4.2.4/src/superlocalmemory.egg-info → superlocalmemory-4.2.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: superlocalmemory
-Version: 4.2.4
+Version: 4.2.5
 Summary: The governed memory brain for AI agents: one local memory for Claude Code, Codex, Cursor, Copilot, ChatGPT and any MCP client
 Author-email: Varun Pratap Bhardwaj <admin@superlocalmemory.com>
 License-Expression: AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/agents/slm-governance-advisor.md

@@ -81,4 +81,4 @@ slm-scope · slm-governance · slm-profile · slm-remember · slm-recall · slm-
 # What NOT to do
 Never session_init twice; never forget without dry-run preview; never run `slm gdpr erase --yes` unasked; never store secrets; never bypass role checks; never claim an erasure succeeded without verifying via recall.
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/agents/slm-loop-runner.md

@@ -79,4 +79,4 @@ assessment. The gate is the authority.
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/agents/slm-memory-advisor.md

@@ -51,4 +51,4 @@ slm-recall · slm-remember · slm-session · slm-scope · slm-profile · slm-gov
 # What NOT to do
 Never session_init twice; never forget dry_run=False without reporting preview; never switch the active profile unasked; never dump a whole file into remember; never invent a memory; never claim "saved" without success:true / clean CLI exit; never bypass scope or governance restrictions.
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/agents/slm-optimize-advisor.md

@@ -41,4 +41,4 @@ slm-compress · slm-cache · slm-status · slm-profile
 # What NOT to do
 Never compress code-for-edit/JSON-to-parse/<500 chars; never store secrets/ccr_ids; never let optimize failure block/alter the task; never claim a specific savings %; never hand a ccr_id to another agent.
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/rules/AGENTS.md

@@ -187,4 +187,4 @@ Using this memory from a web assistant or another computer: see the slm-web-acce
 - **slm-governance-advisor** — scope/role compliance, retention and lifecycle, GDPR
 - **slm-loop-runner** — bounded, gate-verified loops (the `/slm-loop` command delegates to it)
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/rules/CLAUDE.md.fragment

@@ -1,4 +1,4 @@
-<!-- BEGIN SuperLocalMemory v4.2.4 -->
+<!-- BEGIN SuperLocalMemory v4.2.5 -->
 
 ## SuperLocalMemory (SLM) — Agent Rules
 
@@ -51,6 +51,6 @@ slm-recall · slm-remember · slm-session · slm-status · slm-cache · slm-comp
 ### Subagents
 slm-memory-advisor (memory decisions, session hygiene, scope/profile guidance) · slm-optimize-advisor (context compression + KV cache) · slm-governance-advisor (scope/roles/compliance/GDPR) · slm-loop-runner (bounded, gate-verified loops)
 
-<!-- END SuperLocalMemory v4.2.4 -->
+<!-- END SuperLocalMemory v4.2.5 -->
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-bot-memory/SKILL.md

@@ -178,4 +178,4 @@ store every local bot on that computer can reach.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-cache/SKILL.md

@@ -144,4 +144,4 @@ These subcommands control daemon-level cache settings and bulk removal. They do
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-compress/SKILL.md

@@ -147,4 +147,4 @@ Content over 1 MB (1 000 000 bytes UTF-8) is processed but `reversible` is force
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-getting-started-bot/SKILL.md

@@ -194,4 +194,4 @@ quality on a shared box.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-governance/SKILL.md

@@ -274,4 +274,4 @@ Before running any destructive or state-changing operation (`forget`,
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-graph/SKILL.md

@@ -324,4 +324,4 @@ All are host-only: a remote caller cannot use them.
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-loop/SKILL.md

@@ -171,4 +171,4 @@ Guides: `docs/skill-evolution.md`, `docs/bounded-loops-bridge.md`.
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-media/SKILL.md

@@ -337,4 +337,4 @@ minutes (previous vectors are kept) and let them decide.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-mesh/SKILL.md

@@ -386,4 +386,4 @@ mesh availability.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-profile/SKILL.md

@@ -161,4 +161,4 @@ anything but `core` or `mesh`.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-recall/SKILL.md

@@ -515,4 +515,4 @@ See `slm-profile`.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-remember/SKILL.md

@@ -372,4 +372,4 @@ sessions on this machine, so use it only when the user asks to move. See
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-scope/SKILL.md

@@ -180,4 +180,4 @@ to review the impact. See `slm-remember` for the full deletion discipline.
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-session/SKILL.md

@@ -288,4 +288,4 @@ answered. To also surface shared or global facts, call `recall` with
 
 ---
 
-*SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later*

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/plugin-src/skills/slm-status/SKILL.md

@@ -228,4 +228,4 @@ the memory-engine upgrade is pending.
 
 ---
 
-SuperLocalMemory v4.2.4 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.2.5 · Qualixar · AGPL-3.0-or-later

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "4.2.4"
+version = "4.2.5"
 description = "The governed memory brain for AI agents: one local memory for Claude Code, Codex, Cursor, Copilot, ChatGPT and any MCP client"
 readme = "README.md"
 license = "AGPL-3.0-or-later"
@@ -289,6 +289,9 @@ dev = [
     "httpx2==2.12.0",
     "pytest>=9.0.2",
     "pytest-asyncio>=0.21",
+    # The release gate runs the ~22k-test suite in parallel (-n); serial it needs
+    # more than the job's whole time budget. Dev group only, never published.
+    "pytest-xdist>=3.6",
     "twine>=6.2.0",
     "wheel==0.46.3",
 ]

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/__init__.py

@@ -32,7 +32,7 @@ if "OMP_NUM_THREADS" not in os.environ:
     os.environ["OMP_NUM_THREADS"] = "2"
 # ---------------------------------------------------------------------------
 
-__version__ = "4.2.4"
+__version__ = "4.2.5"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.6.1",

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/cli/commands.py

@@ -4042,6 +4042,7 @@ def _cmd_doctor_checks(args: Namespace, state: dict) -> None:
             "hidden_by_forgetting": _h.hidden_by_forgetting,
             "inconsistently_hidden": _h.inconsistently_hidden,
             "missing_projection": _h.missing_projection,
+            "indexing": _h.indexing,
             "index_verified": _h.index_verified,
             "reachability": round(_h.reachability, 4),
             "healthy": _h.healthy,

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/core/config.py

@@ -274,6 +274,17 @@ class ScopeConfig:
 # Encoding Config
 # ---------------------------------------------------------------------------
 
+def _load_coverage_settings(encoding: "EncodingConfig", section: object) -> None:
+    """The two coverage limits from config.json; anything but a whole number
+    of 0 or more keeps the default."""
+    if not isinstance(section, dict):
+        return
+    for key in ("coverage_facts_per_memory", "coverage_backfill_per_memory"):
+        value = section.get(key)
+        if isinstance(value, int) and not isinstance(value, bool) and value >= 0:
+            setattr(encoding, key, value)
+
+
 @dataclass
 class EncodingConfig:
     """Configuration for the encoding (memory creation) pipeline."""
@@ -281,6 +292,11 @@ class EncodingConfig:
     # Fact extraction
     chunk_size: int = 10           # Conversation turns per extraction chunk
     max_facts_per_chunk: int = 10  # V3.3.11: increased from 5 to preserve more details
+    # 4.2.5: verbatim facts per memory for sentences extraction missed (0 = off)
+    coverage_facts_per_memory: int = 40
+    # 4.2.5: the same, for memories saved before it (the daemon backfill);
+    # config.json "encoding" turns it on, e.g. 10
+    coverage_backfill_per_memory: int = 0  # off until measured on a large store
     min_fact_confidence: float = 0.3
 
     # Entity resolution
@@ -1449,6 +1465,7 @@ class SLMConfig:
 
         # V3.3 config fields (additive — defaults work if missing from JSON)
         config.hopfield = load_hopfield_config(data.get("hopfield"))
+        _load_coverage_settings(config.encoding, data.get("encoding"))
         fg = data.get("forgetting", {})
         if fg:
             config.forgetting = ForgettingConfig(**{

superlocalmemory-4.2.5/src/superlocalmemory/core/coverage_backfill.py

@@ -0,0 +1,212 @@
+# Copyright (c) 2026 Varun Pratap Bhardwaj / Qualixar
+# Licensed under AGPL-3.0-or-later
+"""Memories saved before 4.2.5 get a fact for every sentence.
+
+Since 4.2.5 a save keeps a verbatim fact for each sentence its extracted facts
+miss (``encoding/coverage_facts.py``). This pass gives older memories the same,
+in the daemon, in bounded batches that yield to recall.
+
+It only reaches memories whose whole-memory fact is live. That fact already
+makes every sentence retrievable, so the pass never exposes text recall could
+not already find; it makes it findable one sentence at a time. A memory with an
+erasure tombstone, or whose whole fact is archived, quarantined, merged,
+expired or a stub, is left as it is. Coverage is judged against every fact the
+memory ever had, live or not, so a corrected or forgotten sentence is not
+brought back.
+
+Each added fact is appended to ``coverage-backfill.jsonl`` (fact and memory
+ids) so the pass can be undone exactly. Newest memories first: they are the
+ones being asked about. ``coverage-backfill.json`` holds the last memory
+reached; a stop loses nothing and a restart resumes there.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import threading
+from collections.abc import Callable
+from dataclasses import replace
+from datetime import UTC, datetime
+from pathlib import Path
+
+from superlocalmemory.encoding.coverage_facts import DEFAULT_LIMIT, coverage_facts
+from superlocalmemory.storage.models import AtomicFact
+
+logger = logging.getLogger(__name__)
+
+BATCH = 25
+LEDGER = "coverage-backfill.jsonl"
+LOCK = "coverage-backfill.lock"
+STATE = "coverage-backfill.json"
+#: Longer memories are stored truncated with a source pointer; left as they are.
+MAX_CHARS = 24000
+#: Above any rowid: the newest-first pass starts here.
+_TOP = 2**62
+
+_CANDIDATES = """
+SELECT m.rowid AS rid, m.memory_id, m.profile_id, m.content, m.session_id,
+       m.session_date, m.scope, m.shared_with
+FROM memories m
+WHERE m.rowid < ? AND length(m.content) BETWEEN 40 AND ?
+  AND EXISTS (SELECT 1 FROM atomic_facts s
+              WHERE s.memory_id = m.memory_id AND s.content = m.content
+                AND s.quarantined = 0 AND s.non_answer = 0
+                AND s.lifecycle != 'archived' AND s.merged_into IS NULL
+                AND s.valid_until IS NULL)
+  {erased}
+ORDER BY m.rowid DESC
+LIMIT ?
+"""
+_ERASED = "AND NOT EXISTS (SELECT 1 FROM projection_tombstones t WHERE t.memory_id = m.memory_id)"
+
+
+def _candidates_sql(execute: Callable) -> str:
+    # M035 creates the tombstone table; a store without it has no erasures.
+    has = execute("SELECT 1 FROM sqlite_master WHERE type='table' "
+                  "AND name='projection_tombstones'", ())
+    return _CANDIDATES.format(erased=_ERASED if has else "")
+_HELD = "SELECT content FROM atomic_facts WHERE memory_id = ? AND content != ?"
+
+
+def plan(execute: Callable, row, limit: int = DEFAULT_LIMIT) -> list[AtomicFact]:
+    """The facts *row*'s memory is missing, ready for the writer."""
+    content = row["content"]
+    held = [AtomicFact(content=r["content"]) for r in execute(_HELD, (row["memory_id"], content))]
+    shared = json.loads(row["shared_with"] or "null")
+    return [
+        replace(fact, memory_id=row["memory_id"], profile_id=row["profile_id"],
+                scope=row["scope"] or "personal", shared_with=shared)
+        for fact in coverage_facts(content, held, session_id=row["session_id"] or "",
+                                   session_date=row["session_date"], limit=limit)
+    ]
+
+
+def pending(execute: Callable, limit: int = DEFAULT_LIMIT,
+            before_rowid: int = _TOP) -> tuple[int, int]:
+    """(memories, facts) the pass would add, without writing anything."""
+    memories = facts = 0
+    sql = _candidates_sql(execute)
+    while True:
+        rows = execute(sql, (before_rowid, MAX_CHARS, 500))
+        if not rows:
+            return memories, facts
+        for row in rows:
+            n = len(plan(execute, row, limit))
+            memories += 1 if n else 0
+            facts += n
+        before_rowid = rows[-1]["rid"]
+
+
+def _try_lock(fd: int) -> bool:
+    try:
+        if os.name == "nt":
+            import msvcrt
+
+            os.lseek(fd, 0, os.SEEK_SET)
+            msvcrt.locking(fd, msvcrt.LK_NBLCK, 1)
+        else:
+            import fcntl
+
+            fcntl.flock(fd, fcntl.LOCK_EX | fcntl.LOCK_NB)
+        return True
+    except OSError:
+        return False
+
+
+def _unlock(fd: int) -> None:
+    if os.name == "nt":
+        import msvcrt
+
+        os.lseek(fd, 0, os.SEEK_SET)
+        msvcrt.locking(fd, msvcrt.LK_UNLCK, 1)
+    else:
+        import fcntl
+
+        fcntl.flock(fd, fcntl.LOCK_UN)
+
+
+class CoverageBackfill:
+    """Bounded, resumable, undoable. ``write`` stores one prebuilt fact with
+    full enrichment (the engine's ``run_store_fact_direct``)."""
+
+    def __init__(self, db, write: Callable[[AtomicFact], object], *, state_dir: Path,
+                 batch: int = BATCH, max_batches: int | None = None,
+                 limit: int = DEFAULT_LIMIT) -> None:
+        self._db = db
+        self._write = write
+        self._dir = Path(state_dir)
+        self._batch = batch
+        self._max_batches = max_batches
+        self._limit = limit
+        self._stop = threading.Event()
+        self.written = 0
+        self.finished = False
+
+    def stop(self, timeout: float = 0.0) -> None:
+        """Stop after the memory in progress."""
+        self._stop.set()
+
+    def _state(self) -> dict:
+        try:
+            return json.loads((self._dir / STATE).read_text(encoding="utf-8"))
+        except (OSError, ValueError):
+            return {"before_rowid": _TOP, "finished": False}
+
+    def _save(self, before_rowid: int, finished: bool) -> None:
+        self._dir.mkdir(parents=True, exist_ok=True)
+        tmp = self._dir / (STATE + ".tmp")
+        tmp.write_text(json.dumps({"before_rowid": before_rowid, "finished": finished,
+                                   "written": self.written}), encoding="utf-8")
+        tmp.replace(self._dir / STATE)
+
+    def _record(self, fact: AtomicFact) -> None:
+        with (self._dir / LEDGER).open("a", encoding="utf-8") as ledger:
+            ledger.write(json.dumps({"fact_id": fact.fact_id, "memory_id": fact.memory_id,
+                                     "profile_id": fact.profile_id,
+                                     "at": datetime.now(UTC).isoformat()}) + "\n")
+
+    def run_until_done(self) -> int:
+        """Run in the calling thread; return the facts added. One runner at a
+        time across processes: a CLI engine and the daemon never both add."""
+        self._dir.mkdir(parents=True, exist_ok=True)
+        fd = os.open(self._dir / LOCK, os.O_RDWR | os.O_CREAT, 0o600)
+        try:
+            if not _try_lock(fd):
+                return 0
+            try:
+                return self._run_locked()
+            finally:
+                _unlock(fd)
+        finally:
+            os.close(fd)
+
+    def _run_locked(self) -> int:
+        from superlocalmemory.core.recall_gate import background_work
+
+        state = self._state()
+        if state.get("finished"):
+            self.finished = True  # newer memories were covered when saved
+            return 0
+        before, batches = int(state.get("before_rowid") or _TOP), 0
+        sql = _candidates_sql(self._db.execute)
+        while not self._stop.is_set() and (self._max_batches is None
+                                           or batches < self._max_batches):
+            rows = self._db.execute(sql, (before, MAX_CHARS, self._batch))
+            if not rows:
+                self.finished = True
+                break
+            with background_work():
+                for row in rows:
+                    if self._stop.is_set():
+                        break
+                    for fact in plan(self._db.execute, row, self._limit):
+                        self._write(fact)
+                        self._record(fact)
+                        self.written += 1
+                    before = row["rid"]
+            self._save(before, False)
+            batches += 1
+        self._save(before, self.finished)
+        return self.written

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/core/embedding_reindex.py

@@ -96,7 +96,7 @@ class ReindexRunner:
         self._api_lock = threading.Lock()
         self._prebuilt: dict[int, Any] = {}
         self._thread: threading.Thread | None = None
-        self._last_purge = 0.0
+        self._last_purge: float | None = None  # None: never purged (monotonic starts at boot)
         self.notice: str | None = None
         self.quiet: dict | None = None
         #: job id -> change-log position of the last full diff that found nothing
@@ -335,7 +335,7 @@ class ReindexRunner:
                 stats["clear"] = "done" if cursor is None else {"cursor": cursor}
                 update_job(conn, latest["job_id"], stats=json.dumps(stats))
             return True
-        if time.monotonic() - self._last_purge > _PURGE_EVERY_S:
+        if self._last_purge is None or time.monotonic() - self._last_purge > _PURGE_EVERY_S:
             self._last_purge = time.monotonic()
             if sp.table_exists(conn, sp.PURGE):
                 with steps.write_txn(conn, self.db_path):

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/core/engine.py

@@ -486,6 +486,7 @@ class MemoryEngine:
             self._maintenance_scheduler = MaintenanceScheduler(
                 self._db, self._config, self._profile_id,
                 embedder=self._embedder,  # v3.8.2: periodic NULL-embedding self-heal
+                fact_writer=self._store_coverage_fact,  # 4.2.5: coverage backfill
             )
             self._maintenance_scheduler.start()
         except Exception as exc:
@@ -609,6 +610,22 @@ class MemoryEngine:
             profile_id=profile_id,
         )
 
+    def _store_coverage_fact(self, fact: AtomicFact) -> str:
+        """Add one sentence fact to its existing memory (core/coverage_backfill.py).
+
+        ``store_fact_direct`` gives a prebuilt fact a memory of its own; a
+        coverage fact belongs to the memory it was cut from, so deleting that
+        memory deletes it too.
+        """
+        from superlocalmemory.core.store_pipeline import run_store_fact_direct
+
+        return run_store_fact_direct(
+            fact, fact.profile_id, db=self._db, embedder=self._embedder,
+            entity_resolver=self._entity_resolver, ann_index=self._ann_index,
+            graph_builder=self._graph_builder, retrieval_engine=self._retrieval_engine,
+            vector_store=self._vector_store,
+        )
+
     def store_fact_direct(
         self, fact: AtomicFact, *, profile_id: str | None = None,
     ) -> str:

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/core/health_repairs.py

@@ -321,8 +321,12 @@ def _tick_action(engine, action, idle):
 
 
 def run_safe_repairs(engine, *, idle):
-    last = getattr(engine, '_memory_health_checked', 0.0)
-    if time.monotonic() - last < CHECK_INTERVAL_S:
+    # ``None`` means "never checked". A 0.0 sentinel is wrong here: the monotonic
+    # clock counts from boot on Linux, so on a host up for less than
+    # CHECK_INTERVAL_S (a fresh CI runner, a just-restarted server) ``now - 0.0``
+    # is still below the interval and the first check would be skipped.
+    last = getattr(engine, '_memory_health_checked', None)
+    if last is not None and time.monotonic() - last < CHECK_INTERVAL_S:
         return
     engine._memory_health_checked = time.monotonic()
     more = False
@@ -331,7 +335,7 @@ def run_safe_repairs(engine, *, idle):
             return
         more = _tick_action(engine, action, idle) or more
     if more:  # continue bounded automatic batches on the next idle tick
-        engine._memory_health_checked = 0.0
+        engine._memory_health_checked = None
 
 
 def _native_total(native):

{superlocalmemory-4.2.4 → superlocalmemory-4.2.5}/src/superlocalmemory/core/maintenance_scheduler.py

@@ -21,7 +21,8 @@ import logging
 from datetime import timedelta
 import os
 import threading
-from typing import TYPE_CHECKING
+from collections.abc import Callable
+from typing import TYPE_CHECKING, Any
 
 if TYPE_CHECKING:
     from superlocalmemory.core.config import SLMConfig
@@ -104,8 +105,12 @@ class MaintenanceScheduler:
         config: SLMConfig,
         profile_id: str = "default",
         embedder: object | None = None,
+        fact_writer: Callable[[Any], object] | None = None,
     ) -> None:
         self._db = db
+        # 4.2.5: stores one prebuilt fact with full enrichment; the coverage
+        # backfill needs it (core/coverage_backfill.py).
+        self._fact_writer = fact_writer
         self._config = config
         self._profile_id = profile_id
         # v3.8.2 self-heal: when provided, periodic maintenance backfills
@@ -180,6 +185,12 @@ class MaintenanceScheduler:
         self._initial_hub_timer = threading.Timer(390.0, self._hub_baseline_pass)
         self._initial_hub_timer.daemon = True
         self._initial_hub_timer.start()
+        # 4.2.5: memories saved before coverage-at-save get a fact for every
+        # sentence. Same one-shot pattern, after the others; the periodic
+        # cycle then continues it in bounded steps.
+        self._initial_coverage_timer = threading.Timer(450.0, self._coverage_pass)
+        self._initial_coverage_timer.daemon = True
+        self._initial_coverage_timer.start()
         logger.info(
             "Maintenance scheduler started (interval=%dm)",
             self._config.forgetting.scheduler_interval_minutes,
@@ -219,6 +230,29 @@ class MaintenanceScheduler:
         except Exception as exc:
             self._record_step("hub baseline", False, str(exc))
 
+    def _coverage_pass(self, max_batches: int | None = 40) -> None:
+        """Add the missing sentence facts of older memories, bounded and
+        resumable (core/coverage_backfill.py)."""
+        limit = self._config.encoding.coverage_backfill_per_memory
+        if not self._running or self._fact_writer is None or limit <= 0:
+            return  # 0 turns the pass off
+        try:
+            from superlocalmemory.core.coverage_backfill import CoverageBackfill
+            from superlocalmemory.infra.data_root import state_path
+
+            self._coverage_job = CoverageBackfill(
+                self._db, self._fact_writer, state_dir=state_path("backfill"),
+                max_batches=max_batches,
+                limit=limit,
+            )
+            written = self._coverage_job.run_until_done()
+            if written:
+                logger.info("Coverage backfill: %d sentence facts added (%s)", written,
+                            "done" if self._coverage_job.finished else "more next cycle")
+            self._record_step("coverage backfill", True)
+        except Exception as exc:  # noqa: BLE001
+            self._record_step("coverage backfill", False, str(exc))
+
     def _project_backfill(self) -> None:
         """Tag a bounded batch of session-end summaries with their project
         (storage/project_backfill.py). Idempotent; a no-op once converged."""
@@ -318,6 +352,13 @@ class MaintenanceScheduler:
         if _hub_timer is not None:
             _hub_timer.cancel()
             self._initial_hub_timer = None
+        _coverage_timer = getattr(self, "_initial_coverage_timer", None)
+        if _coverage_timer is not None:
+            _coverage_timer.cancel()
+            self._initial_coverage_timer = None
+        _coverage_job = getattr(self, "_coverage_job", None)
+        if _coverage_job is not None:  # stops after the memory in progress
+            _coverage_job.stop()
         _hub_pass = getattr(self, "_hub_pass", None)
         if _hub_pass is not None:  # stops between batches
             _hub_pass.stop(timeout=0)
@@ -571,6 +612,7 @@ class MaintenanceScheduler:
         # #150: once per cycle, all profiles in one sweep (each row stays in
         # its own profile); batched, so the write lock is held per batch.
         self._project_backfill()
+        self._coverage_pass()
 
         # Anything that has failed on several cycles running is not a blip.
         # Every step here logs and continues, which is right -- one broken step