fix(channelmemory): harden passive extraction reliability (#1386)

Co-authored-by: Collective Developer <man@collective.dev>

docs/06-store-data-model.md

@@ -720,7 +720,7 @@ L0 (Working Memory)           L1 (Episodic Memory)        L2 (Semantic Memory)
 
 | Interface | Purpose | Key Methods |
 |-----------|---------|-------------|
-| `EpisodicStore` | Tier 1.5 memory CRUD + hybrid search | `Create`, `Search`, `ExistsBySourceID`, `ListUnpromoted`, `MarkPromoted` |
+| `EpisodicStore` | Tier 1.5 memory CRUD + hybrid search | `Create`, `Search`, `ExistsBySourceID`, `GetBySourceID`, `ListUnpromoted`, `MarkPromoted` |
 | `EvolutionMetricsStore` | Stage 1: record metrics (retrieval, tool, feedback) | `RecordMetric`, `AggregateToolMetrics`, `AggregateRetrievalMetrics` |
 | `EvolutionSuggestionStore` | Stage 2: generate & track improvement suggestions | `CreateSuggestion`, `ListSuggestions`, `UpdateSuggestionStatus` |
 | `VaultStore` | Knowledge Vault: document registry + links | `UpsertDocument`, `Search`, `CreateLink`, `GetOutLinks`, `GetBacklinks` |
@@ -744,7 +744,7 @@ L0 (Working Memory)           L1 (Episodic Memory)        L2 (Semantic Memory)
 `ChannelMemoryExtractionStore` is implemented for PostgreSQL and SQLite. It is
 tenant-scoped, stores no raw message bodies, and uses deterministic hashes to
 deduplicate the same channel/history/type/summary candidate across repeated
-runs.
+runs for the same channel instance.
 
 ### 12 Promoted Agent Columns
 

docs/project-changelog.md

@@ -4,6 +4,23 @@ Significant changes, features, and fixes in reverse chronological order.
 
 ---
 
+## 2026-07-07
+
+### Passive channel memory extraction reliability
+
+**Fixes**
+
+- Fixed partial settings updates so enabling passive memory no longer clears review mode or other omitted options.
+- Extraction runs now checkpoint only messages actually sent within the extraction input budget, preventing skipped channel messages.
+- Pending review counts now count the full queue, retention uses the configured `retention_hours`, duplicate approvals preserve the existing episodic ID, and failed item writes mark the run as failed.
+- Passive memory candidates now dedupe across runs for the same channel instance and item hash in both PostgreSQL and SQLite.
+
+**Tests**
+
+- Added regression coverage for config patching, checkpoint bounds, pending counts, run failure finalization, retention, duplicate approval, and SQLite cross-run dedupe.
+
+---
+
 ## 2026-07-03
 
 ### Usage cost display precision