Implements a Weibull stretched-exponential decay function for scoring memory importance based on recency, frequency, and intrinsic importance. Exported as a factory function `createDecayEngine(options)` that returns a `scoreMemory` function. Configurable half-life, tier-specific beta, and frequency/intrinsic components. Completely self-contained with no internal imports, making it reusable as a generic decay scoring utility for any application needing time-decay ranking.

Provides `expandQuery(query, source)` which appends technical synonyms to a user query for improved BM25 retrieval. Uses a static map of Chinese terms to English equivalents (e.g., '崩溃' → '崩溃 crash error') and avoids false positives via substring checks. Gated by `source` parameter to only apply to manual recall, not auto-recall. Highly reusable as a standalone utility for expanding search queries with domain-specific synonyms.

Implements a `TraceCollector` class that instruments each stage of a retrieval pipeline with start/end counts, duration, and drop IDs. Exposes `getLastDiagnostics()` returning a structured JSON object with stage-by-stage counts, drop summary, and failure stage. Designed as a pass-through collector that can be injected into any multi-stage pipeline (not just retrieval). No internal imports, making it a reusable pattern for observability in any data processing flow.

Provides `parseSmartMetadata()`, `buildSmartMetadata()`, and `stringifySmartMetadata()` for converting between raw metadata strings and structured objects, supporting both camelCase and snake_case field names, legacy compatibility, clamping, and lifecycle-aware boolean checks (`isMemoryExpired`, `isMemoryActiveAt`). Self-contained with no internal imports beyond standard types. Can be extracted as a generic metadata handling utility for any system that stores structured metadata in a JSON string column.