superlocalmemory 4.1.7 → 4.1.9

package/.claude-plugin/marketplace.json

@@ -21,7 +21,7 @@
       "license": "AGPL-3.0-or-later",
       "name": "superlocalmemory",
       "source": "./plugin",
-      "version": "4.1.7"
+      "version": "4.1.9"
     },
     {
       "author": {
@@ -39,7 +39,7 @@
       "license": "AGPL-3.0-or-later",
       "name": "superlocalmemory-codex",
       "source": "./codex-plugin",
-      "version": "4.1.7"
+      "version": "4.1.9"
     }
   ]
 }

package/CHANGELOG.md

@@ -5,6 +5,52 @@ All notable changes to SuperLocalMemory will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [4.1.9] — Silence is not a verdict
+
+### Fixed
+- **A recall nobody engaged with was scored as an average one.** How useful a recall
+  was is worked out from what happened next — whether a memory was quoted, edited, or
+  asked about again. When nothing at all had happened, that calculation still landed
+  on a middling score, and it was saved as if it had been measured. That is not a
+  blank: downstream it reads as mild approval, nudging results toward whatever was
+  shown and making the system steadily more certain it has no preference either way.
+  Recalls with nothing observed are now closed without a score, which is what the
+  batch pass already did — the two paths simply disagreed. On a store in daily use,
+  every one of 492 saved scores turned out to be this value, so nothing recorded there
+  reflected real use.
+- **Signals recorded as explicitly absent were counted as present.** A recall whose
+  signals were all recorded as "did not happen" was treated as having something to
+  say, and scored, because the record existed even though everything in it was empty.
+  Both paths now ask whether anything actually happened rather than whether a record
+  was written.
+
+## [4.1.8] — The same question, answered the same way
+
+### Fixed
+- **Recall could return the same memories in a different order each time it was
+  asked.** Adaptive ranking became the default in 4.1.7, which let a contextual
+  bandit choose how much to weigh each retrieval channel per query. That choice
+  is sampled from what an arm has learned, and arms that have settled no rewards
+  yet hold only their starting assumption — so the sample varied, and two
+  identical questions could weigh semantic against keyword evidence quite
+  differently for no reason drawn from anything observed. Adaptive ranking is
+  opt-in again: set `SLM_RANKING=v2-ensemble` to enable it. Nothing else about
+  retrieval changed, and no stored memory is affected.
+- **One out-of-date label could make the whole store refuse to answer.** A
+  memory filed as a plan stops reading as a plan once its date passes — ordinary
+  use, and the maintenance pass re-reads those on its own. That check was
+  treated like a missing table: every route answered `503 Service unavailable`
+  until the daemon was restarted, and because the readiness answer was decided
+  once at startup, the repair that had already run could not lift it. The check
+  now reports what it is, a data quality signal, and the store keeps serving
+  while the next maintenance pass settles it.
+- **Tool usage stopped being recorded.** Behavioural patterns, per-skill
+  performance, and the engagement measures that tell a recall whether it helped
+  all read the same record of which tools ran. Nothing on a normal install was
+  writing it, so those readings kept reporting from progressively older activity
+  and eventually stood still. Each completed tool call is recorded again, with
+  credential-shaped text removed and summaries capped before they are stored.
+
 ## [4.1.7] — Recall that improves with use
 
 ### Added

package/README.md

@@ -5,7 +5,7 @@
   </picture>
 </p>
 
-<h1 align="center">SuperLocalMemory V4.1.7</h1>
+<h1 align="center">SuperLocalMemory V4.1.9</h1>
 
 <h2 align="center">Rent the LLM. Own the memory.</h2>
 
@@ -27,12 +27,12 @@ guarantee here is stated as a falsifiable invariant, tested under an adversarial
 negative control, and shipped with the harness that regenerates the evidence:
 <code>python benchmark/run_all.py --trials 200 --output-dir results/</code>. What each experiment
 does <em>not</em> exercise is stated too.</p>
-<p align="center"><code>v4.1.7</code> — one control plane: <strong>SLM-Mesh</strong> peer coordination · multi-scope memory (personal / shared / global) · profiles · Cache · Compress · 7-layer retrieval · code graph · Entity Explorer · skill evolution · Modes A/B/C · GDPR retention &amp; audit chain · bounded loops — across CLI, MCP, dashboard, the <strong>Claude plugin</strong>, the <strong>Codex add-on</strong>, and documented IDE integrations.<br/>
+<p align="center"><code>v4.1.9</code> — one control plane: <strong>SLM-Mesh</strong> peer coordination · multi-scope memory (personal / shared / global) · profiles · Cache · Compress · 7-layer retrieval · code graph · Entity Explorer · skill evolution · Modes A/B/C · GDPR retention &amp; audit chain · bounded loops — across CLI, MCP, dashboard, the <strong>Claude plugin</strong>, the <strong>Codex add-on</strong>, and documented IDE integrations.<br/>
 Proxy: <code>slm wrap claude</code> &nbsp;·&nbsp; MCP: add <code>slm_compress</code> to your config &nbsp;·&nbsp; Skill: zero-config</p>
 <p align="center"><strong>Four public arXiv preprints</strong> · V4: <a href="https://arxiv.org/abs/2608.08253">arXiv:2608.08253</a> · companion archive: <a href="https://zenodo.org/records/21853302">Zenodo 21853302</a> (<a href="https://doi.org/10.5281/zenodo.21853302">DOI 10.5281/zenodo.21853302</a>) · prior preprints: <a href="https://arxiv.org/abs/2603.02240">2603.02240</a> · <a href="https://arxiv.org/abs/2603.14588">2603.14588</a> · <a href="https://arxiv.org/abs/2604.04514">2604.04514</a>.</p>
 
 <p align="center">
-  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v4.1.7-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v4.1.7 — Current Release"/></a>
+  <a href="CHANGELOG.md"><img src="https://img.shields.io/badge/v4.1.9-Current_Release-2ea44f?style=for-the-badge&logo=checkmarx&logoColor=white" alt="v4.1.9 — Current Release"/></a>
   <a href="https://arxiv.org/abs/2608.08253"><img src="https://img.shields.io/badge/arXiv-2608.08253-b31b1b?style=for-the-badge&logo=arxiv&logoColor=white" alt="SuperLocalMemory 4.0 paper on arXiv:2608.08253"/></a>
   <a href="https://zenodo.org/records/21853302"><img src="https://img.shields.io/badge/Zenodo-10.5281%2Fzenodo.21853302-1682D4?style=for-the-badge&logo=zenodo&logoColor=white" alt="V4 paper on Zenodo: 10.5281/zenodo.21853302"/></a>
   <a href="https://arxiv.org/abs/2603.14588"><img src="https://img.shields.io/badge/arXiv-2603.14588-b31b1b?style=for-the-badge&logo=arxiv&logoColor=white" alt="arXiv Paper"/></a>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "superlocalmemory",
-  "version": "4.1.7",
+  "version": "4.1.9",
   "description": "Local-first agent memory with MCP and an agent-native CLI. Documented clients include Claude Code, Cursor, and Windsurf.",
   "keywords": [
     "ai-memory",

package/plugin/.claude-plugin/plugin.json

@@ -15,5 +15,5 @@
   "mcpServers": "./.mcp.json",
   "name": "superlocalmemory",
   "repository": "https://github.com/qualixar/superlocalmemory",
-  "version": "4.1.7"
+  "version": "4.1.9"
 }

package/plugin/CLAUDE.md

@@ -1,4 +1,4 @@
-<!-- BEGIN SuperLocalMemory v4.1.7 -->
+<!-- BEGIN SuperLocalMemory v4.1.9 -->
 
 ## SuperLocalMemory (SLM) — Agent Rules
 
@@ -39,6 +39,6 @@ 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)
 
-<!-- END SuperLocalMemory v4.1.7 -->
+<!-- END SuperLocalMemory v4.1.9 -->
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-governance-advisor.md

@@ -77,4 +77,4 @@ slm-scope · slm-governance · slm-profile · slm-remember · slm-recall
 # What NOT to do
 Never session_init twice; never forget without dry-run preview; never store secrets; never bypass role checks; never claim an erasure succeeded without verifying via recall.
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-loop-runner.md

@@ -68,4 +68,4 @@ assessment. The gate is the authority.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/agents/slm-memory-advisor.md

@@ -46,4 +46,4 @@ 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 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 v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/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 carry ccr_ids across profile switches.
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/requirements.txt

@@ -1 +1 @@
-superlocalmemory==4.1.7
+superlocalmemory==4.1.9

package/plugin/skills/slm-cache/SKILL.md

@@ -145,4 +145,4 @@ These subcommands control daemon-level cache settings. They do not read or write
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/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 v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/skills/slm-governance/SKILL.md

@@ -245,4 +245,4 @@ Before running any destructive operation (`forget`, `compact_memories`):
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-graph/SKILL.md

@@ -312,4 +312,4 @@ profile. See `slm-profile` for the full profile switching workflow.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/skills/slm-loop/SKILL.md

@@ -96,4 +96,4 @@ paused, name the approval needed; when errored, quote the short detail.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin/skills/slm-mesh/SKILL.md

@@ -279,4 +279,4 @@ mesh availability.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-profile/SKILL.md

@@ -146,4 +146,4 @@ Name them differently in your MCP config (e.g. `superlocalmemory-personal` and
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-recall/SKILL.md

@@ -323,4 +323,4 @@ before recalling, then switch back. See `slm-profile` for workspace switching.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-remember/SKILL.md

@@ -270,4 +270,4 @@ different workspace, use `switch_profile` first. See `slm-profile`.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-scope/SKILL.md

@@ -173,4 +173,4 @@ to review the impact. See `slm-remember` for the full deletion discipline.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-session/SKILL.md

@@ -253,4 +253,4 @@ explicitly and call `recall` with `include_global`/`include_shared` after
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin/skills/slm-status/SKILL.md

@@ -163,4 +163,4 @@ multi-profile setup. To switch the active profile, see `slm-profile`.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/agents/slm-memory-advisor.md

@@ -46,4 +46,4 @@ 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 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 v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/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 carry ccr_ids across profile switches.
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/rules/AGENTS.md

@@ -137,4 +137,4 @@ When the SLM MCP server is unavailable, use these CLI equivalents:
 - **slm-optimize-advisor** — context compression and KV cache
 - **slm-governance-advisor** — scope/role compliance, retention policies, GDPR
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/skills/slm-cache/SKILL.md

@@ -145,4 +145,4 @@ These subcommands control daemon-level cache settings. They do not read or write
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/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 v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/skills/slm-governance/SKILL.md

@@ -245,4 +245,4 @@ Before running any destructive operation (`forget`, `compact_memories`):
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-graph/SKILL.md

@@ -312,4 +312,4 @@ profile. See `slm-profile` for the full profile switching workflow.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/skills/slm-loop/SKILL.md

@@ -96,4 +96,4 @@ paused, name the approval needed; when errored, quote the short detail.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/plugin-src/skills/slm-mesh/SKILL.md

@@ -279,4 +279,4 @@ mesh availability.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-profile/SKILL.md

@@ -146,4 +146,4 @@ Name them differently in your MCP config (e.g. `superlocalmemory-personal` and
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-recall/SKILL.md

@@ -323,4 +323,4 @@ before recalling, then switch back. See `slm-profile` for workspace switching.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-remember/SKILL.md

@@ -270,4 +270,4 @@ different workspace, use `switch_profile` first. See `slm-profile`.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-scope/SKILL.md

@@ -173,4 +173,4 @@ to review the impact. See `slm-remember` for the full deletion discipline.
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-session/SKILL.md

@@ -253,4 +253,4 @@ explicitly and call `recall` with `include_global`/`include_shared` after
 
 ---
 
-*SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later*
+*SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later*

package/plugin-src/skills/slm-status/SKILL.md

@@ -163,4 +163,4 @@ multi-profile setup. To switch the active profile, see `slm-profile`.
 
 ---
 
-SuperLocalMemory v4.1.7 · Qualixar · AGPL-3.0-or-later
+SuperLocalMemory v4.1.9 · Qualixar · AGPL-3.0-or-later

package/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "superlocalmemory"
-version = "4.1.7"
+version = "4.1.9"
 description = "Local-first agent memory with auditable hybrid retrieval"
 readme = "README.md"
 license = "AGPL-3.0-or-later"

package/src/superlocalmemory/__init__.py

@@ -32,7 +32,7 @@ if "OMP_NUM_THREADS" not in os.environ:
     os.environ["OMP_NUM_THREADS"] = "2"
 # ---------------------------------------------------------------------------
 
-__version__ = "4.1.7"
+__version__ = "4.1.9"
 
 _REQUIRED_VERSIONS = {
     "sentence_transformers": "5.3.0",

package/src/superlocalmemory/core/recall_pipeline.py

@@ -340,23 +340,26 @@ class _ReadOnlyLearningView:
 def _resolve_ranking_mode(env: "dict[str, str] | os._Environ[str]") -> str:
     """Map the ``SLM_RANKING`` env var to a canonical mode.
 
-    ``SLM_RANKING`` is an explicit operator policy. Absence of that policy now
-    means the full pipeline, not ``off``.
-
-    It defaulted to ``off`` because a stored signal that had gone stale must
-    not start reordering results merely because someone upgraded. That risk
-    depended on a settler that scored a recall it could not observe, which no
-    longer happens: an unobserved recall leaves the ranking untouched instead
-    of being recorded as an average one, and each observation counts for less
-    the more predictable it was. A ranking layer nobody switches on is a
-    ranking layer that never learns, so the default now enables it.
-
-    Set ``SLM_RANKING=off`` to opt out.
+    ``SLM_RANKING`` is an explicit operator policy, and the absence of one
+    means ``off``: an upgrade must not start reordering results on its own.
+
+    Fixing the settlement path was necessary for the ensemble to be worth
+    enabling, but it is not sufficient. Settlement decides what an arm learns
+    *after* a query; the channel weights come from sampling that arm's
+    posterior *before* it. Until arms have posteriors shaped by real
+    observations, that sample is drawn from the prior, so two identical
+    queries can be weighted differently for no reason the store can justify —
+    variance with nothing bought by it. A memory system's answers should move
+    because the evidence moved.
+
+    So the ensemble stays opt-in until arms carry settled rewards. Set
+    ``SLM_RANKING=v2-ensemble`` to enable it, or ``v1``/``v2`` for the
+    narrower passes.
     """
     raw = (env.get("SLM_RANKING", "") or "").strip().lower()
     if raw in _RANKING_MODES:
         return raw
-    return "v2-ensemble"
+    return "off"
 
 
 def apply_ranking(

package/src/superlocalmemory/hooks/post_tool_outcome_hook.py

@@ -83,6 +83,122 @@ def _validate(marker: str) -> str | None:
         return None
 
 
+# --- Behavioural telemetry -------------------------------------------------
+# ``tool_events`` is the record of what the agent did, and several readers
+# depend on it: assertion mining, skill-performance mining, and the engagement
+# features that settle a recall against the actions which followed it. Each of
+# those reads the table; none of them writes it. The only writers were an
+# explicit ``log_tool_event`` call and a bulk importer, so on an ordinary
+# install the table simply stopped receiving invocations and every reader
+# quietly aged out with it.
+#
+# This hook already runs on exactly the right edge -- after each tool completes,
+# with the session that ran it -- so it records the invocation here. It happens
+# before the marker scan and independently of it: whether a recalled memory was
+# named in the output decides what a *reward* is worth, not whether the action
+# occurred at all.
+_MAX_SUMMARY_LEN = 500
+
+# Compiled once at import. This path now runs on every tool call rather than
+# only on the ~1-in-5 that carry a marker, so per-call ``re`` compilation or a
+# module import would be paid on the hot path for no benefit.
+_SECRET_PATTERNS = (
+    (re.compile(r"\b(?:sk-|pk-|api[_-]?key[_-]?)[A-Za-z0-9_-]{10,}\b"),
+     "[REDACTED]"),
+    (re.compile(r"\b[A-Za-z0-9+/]{40,}={0,2}\b"), "[REDACTED]"),
+    (re.compile(r"password\s*[=:]\s*\S+", re.IGNORECASE),
+     "password=[REDACTED]"),
+)
+
+
+def _scrub(text: str) -> str:
+    """Strip credential-shaped substrings from telemetry text."""
+    for pattern, replacement in _SECRET_PATTERNS:
+        text = pattern.sub(replacement, text)
+    return text
+
+
+def _summary(raw: object) -> str:
+    """Truncate-then-scrub ``raw`` for a summary column.
+
+    Truncation comes first so the regex cost is bounded by
+    ``_MAX_SUMMARY_LEN`` and not by the size of a tool response.
+    """
+    if raw is None:
+        return ""
+    if not isinstance(raw, str):
+        try:
+            import json as _json
+            raw = _json.dumps(raw, default=str)
+        except Exception:
+            try:
+                raw = str(raw)
+            except Exception:
+                return ""
+    return _scrub(raw[:_MAX_SUMMARY_LEN])
+
+
+def _record_tool_event(
+    session_id: str,
+    tool_name: str,
+    payload: dict,
+    response_text: str,
+) -> bool:
+    """Append one ``tool_events`` row. Never raises; returns whether it wrote.
+
+    Failure is silent by design: telemetry must never be the reason a tool call
+    reports a problem to the user, and the hook contract is to exit 0 whatever
+    happens.
+    """
+    if not tool_name:
+        return False
+    try:
+        import sqlite3
+        from datetime import datetime, timezone
+
+        try:
+            from superlocalmemory.hooks.session_registry import (
+                resolve_active_profile,
+            )
+            profile_id = resolve_active_profile() or "default"
+        except Exception:
+            profile_id = "default"
+
+        project_path = ""
+        raw_cwd = payload.get("cwd")
+        if isinstance(raw_cwd, str):
+            project_path = raw_cwd[:_MAX_SUMMARY_LEN]
+
+        conn = sqlite3.connect(str(_memory_db_path()), timeout=0.05)
+        try:
+            conn.execute("PRAGMA busy_timeout=50")
+            conn.execute(
+                "INSERT INTO tool_events "
+                "(session_id, profile_id, project_path, tool_name, event_type,"
+                " input_summary, output_summary, duration_ms, metadata,"
+                " created_at) "
+                "VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)",
+                (
+                    session_id,
+                    profile_id,
+                    project_path,
+                    tool_name,
+                    "complete",
+                    _summary(payload.get("tool_input")),
+                    _summary(response_text),
+                    0,
+                    "{}",
+                    datetime.now(timezone.utc).isoformat(),
+                ),
+            )
+            conn.commit()
+        finally:
+            conn.close()
+        return True
+    except Exception:
+        return False
+
+
 def _inner_main() -> str:
     """Return an ``outcome`` string (for perf log); never raises."""
     payload = read_stdin_json()
@@ -119,6 +235,12 @@ def _inner_main() -> str:
 
     # Response scan — capped BEFORE regex (bound O(cap)).
     response_text = summarize_response(payload.get("tool_response"))
+
+    # The invocation is recorded before anything is decided about markers. An
+    # action the agent took is a fact about the session on its own terms; the
+    # marker only decides whether it also settles a pending recall.
+    _record_tool_event(session_id, tool_name, payload, response_text)
+
     if not response_text:
         return "no_response"
 

package/src/superlocalmemory/learning/reward.py

@@ -113,6 +113,30 @@ _MAX_CLOCK_SKEW_MS: Final[int] = 60_000
 # ---------------------------------------------------------------------------
 
 
+#: Dwell below this is not engagement; the label formula uses the same bound.
+_DWELL_THRESHOLD_MS: Final[int] = 2000
+
+
+def _carries_evidence(signals: Mapping[str, object]) -> bool:
+    """Did anything actually happen after this recall?
+
+    ``_compute_label`` is built as ``0.5 + bonuses - penalties``, so a signal
+    set with nothing in it evaluates to exactly one half. That number is not an
+    absence of judgement, it is a confident one: it becomes a mid-strength
+    positive training label and a Beta update that tightens a posterior around
+    its prior. An empty dict and a dict of falsy values mean the same thing here
+    and must both be refused.
+    """
+    if not signals:
+        return False
+    if signals.get("cite") or signals.get("edit") or signals.get("requery"):
+        return True
+    try:
+        return int(signals.get("dwell_ms", 0) or 0) >= _DWELL_THRESHOLD_MS
+    except (TypeError, ValueError):
+        return False
+
+
 def _compute_label(signals: Mapping[str, object]) -> float:
     """Deterministic label in ``[0.0, 1.0]`` per the manifest A.1 formula.
 
@@ -141,8 +165,9 @@ def _compute_label(signals: Mapping[str, object]) -> float:
         dwell_int = 0
 
     dwell_bonus = 0.0
-    if dwell_int >= 2000:
-        dwell_bonus = min(0.15, 0.05 + (dwell_int - 2000) / 80_000.0)
+    if dwell_int >= _DWELL_THRESHOLD_MS:
+        dwell_bonus = min(
+            0.15, 0.05 + (dwell_int - _DWELL_THRESHOLD_MS) / 80_000.0)
 
     label = (
         0.5
@@ -585,9 +610,19 @@ class EngagementRewardModel:
                     signals = json.loads(pending["signals_json"] or "{}")
                 except json.JSONDecodeError:  # pragma: no cover — defensive
                     signals = {}
-                reward = _compute_label(signals)
                 now_ms = self._clock_ms()
                 timestamp_iso = _iso_from_ms(now_ms)
+                if not _carries_evidence(signals):
+                    # Finalize so it stops being rescanned, but record no
+                    # outcome: nothing was observed, and a score here would
+                    # be invented rather than measured. Mirrors reap_stale.
+                    conn.execute(
+                        "UPDATE pending_outcomes "
+                        "SET status = 'settled' WHERE outcome_id = ?",
+                        (outcome_id,),
+                    )
+                    return _FALLBACK_REWARD
+                reward = _compute_label(signals)
 
                 # NOTE: Split INSERT across lines so the Stage-5b CI
                 # gate's single-line regex (LLD-00 §13) does not fire.
@@ -746,7 +781,7 @@ class EngagementRewardModel:
             #
             # The row is still finalized, so it stops being rescanned; it is
             # simply not turned into evidence it never was.
-            if not signals:
+            if not _carries_evidence(signals):
                 settle_ids.append(row["outcome_id"])
                 continue
             reward = _compute_label(signals)

package/src/superlocalmemory/storage/migrations/M048_upcoming_holds_only_what_is_upcoming.py

@@ -205,3 +205,25 @@ def verify(conn: sqlite3.Connection) -> bool:
         )
         return False
     return True
+
+
+def blocks_serving(conn: sqlite3.Connection) -> bool:
+    """Should a daemon refuse to serve while this check does not hold? No.
+
+    Nothing here is about schema. ``verify()`` reads ``content`` and compares
+    today's reading of it against the ``fact_type`` already stored, so a False
+    means some memories are filed as plans that no longer read as plans. Every
+    table and column a query needs is present either way; the store answers
+    normally, and at worst a handful of memories carry a stale label until the
+    next maintenance pass re-reads them.
+
+    The distinction matters because this is a standing guard over data that
+    ordinary use re-violates by design: a plan whose date passes stops reading
+    as upcoming, which is the rule working, not a fault. Treating that like a
+    missing table let one drifted row answer 503 on every route for as long as
+    the process lived — and because the readiness snapshot is taken once at
+    startup, the background pass that repairs the data could not lift the
+    refusal it caused. An outage produced by a quality check is worse than the
+    thing the check is for. Same reasoning as ``M043.blocks_serving``.
+    """
+    return False