fix(recall): cap default limit to 30 + relevance-bound OR-broadening

memory_recall was returning almost the entire store instead of a small
set of relevant matches. Two compounding causes, both fixed here:

1. Default limit was 10000 (commit d03a77a "effectively unlimited").
   recall_memories/MemoryRecall and the memory_recall MCP tool now
   default to 30 (ceiling stays 10000 for callers that opt in).

2. The OR-broadening fallback (fires when the precise AND-match is
   sparse) ordered by the importance hybrid and padded up to `limit`,
   so with limit=10000 it flooded results with high-importance but
   irrelevant memories. It now orders OR-matches by ts_rank(relevance)
   DESC and applies a minimum-rank floor (OR_BROADEN_MIN_RANK=0.01) to
   drop rows that merely contain a query word incidentally.

Tests: add test_recall_default_limit_is_capped (asserts 30 passed to
fetch) and test_recall_or_broadening_is_relevance_bounded (asserts the
OR query is relevance-ordered + rank-floored). Full suite 176 green,
ruff + mypy clean.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

src/claude_memory/api/app.py

@@ -190,6 +190,12 @@ async def store_memory(body: MemoryStore, user: AuthUser = Depends(get_current_u
     return MemoryResponse(id=row["id"], category=row["category"], importance=row["importance"])
 
 
+# OR-broadening fallback: when the precise AND-match is sparse we widen to an
+# OR-match to fill results, but only with rows whose relevance (ts_rank) clears
+# this floor. Below it a row merely contains one query word incidentally — noise.
+OR_BROADEN_MIN_RANK = 0.01
+
+
 @app.post("/api/memories/recall")
 async def recall_memories(body: MemoryRecall, user: AuthUser = Depends(get_current_user)) -> dict[str, Any]:
     pool = await get_pool()
@@ -251,8 +257,9 @@ async def recall_memories(body: MemoryRecall, user: AuthUser = Depends(get_curre
                     FROM memories, to_tsquery('english', $2) query
                     WHERE deleted_at IS NULL
                       AND search_vector @@ query
+                      AND ts_rank(search_vector, query) > {OR_BROADEN_MIN_RANK}
                       {or_cat_filter}
-                    ORDER BY {order_clause}
+                    ORDER BY ts_rank(search_vector, query) DESC
                     LIMIT $3
                     """,
                     *or_params,
@@ -896,7 +903,7 @@ async def memory_store(content: str, category: str = "facts", tags: str = "",
 @mcp_server.tool()
 async def memory_recall(context: str, expanded_query: str = "",
                         category: str | None = None, sort_by: str = "importance",
-                        limit: int = 10000) -> str:
+                        limit: int = 30) -> str:
     """Recall memories by semantic search."""
     pool = await get_pool()
     user_id = _current_user.get()
@@ -957,8 +964,9 @@ async def memory_recall(context: str, expanded_query: str = "",
                     FROM memories, to_tsquery('english', $2) query
                     WHERE deleted_at IS NULL
                       AND search_vector @@ query
+                      AND ts_rank(search_vector, query) > {OR_BROADEN_MIN_RANK}
                       {or_cat_filter}
-                    ORDER BY {order_clause}
+                    ORDER BY ts_rank(search_vector, query) DESC
                     LIMIT $3
                     """,
                     *or_params,

src/claude_memory/api/models.py

@@ -17,7 +17,9 @@ class MemoryRecall(BaseModel):
     expanded_query: str = ""
     category: Optional[str] = None
     sort_by: Literal["importance", "relevance", "recency"] = "importance"
-    limit: int = Field(default=10000, ge=1, le=10000)
+    # Default to a small top-N so recall returns the most relevant matches, not
+    # the whole store. Ceiling stays high for callers that explicitly want more.
+    limit: int = Field(default=30, ge=1, le=10000)
 
 
 class MemoryResponse(BaseModel):