Release mnemostack 1.3.0

CHANGELOG.md

@@ -6,6 +6,8 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+## [1.3.0] - 2026-07-27
+
 ### Added
 
 - **Multi-field lexical arms (`recall.text_search_fields`)**: title/heading fields are usually far more precise lexical signals than chunk bodies — under `text_search: lexical`, a mapping of payload field → fusion weight (YAML `text_search_fields: {title: 2.0, text: 1.0}`, env `MNEMOSTACK_TEXT_SEARCH_FIELDS="title:2.0,text:1.0"`) turns each field into its OWN lexically-gated arm: the MatchText gate runs on that field while the result text still comes from `text_key` (a hit surfaced only by the title arm carries its chunk body, not its title, into fusion and synthesis). Arms get distinct names (`qdrant_text` for the `text_key` field, `qdrant_text:<field>` otherwise), so fusion weights, the adaptive (Q-learning) profile, and latency/hit/trace/degraded telemetry stay per-arm — previously two same-class arms shared one name and were indistinguishable everywhere that matters; `Retriever` subclasses now accept a `name=` constructor override for exactly that reason. Weights apply at the fusion level (the gate filters, it cannot score); weight 1.0 is left to the adaptive profile, anything else is a static override. When set, the mapping REPLACES the arm set (list `text_key` explicitly to keep the body arm), and it requires `text_search: lexical` — any other mode fails loud at startup instead of silently ignoring the fields. Wired on every surface (HTTP server, MCP, CLI): `mnemostack text-index` creates the full-text index for EVERY configured gate field, `doctor` live-checks each one and validates the fields/mode pairing, and `--source bm25` still selects the whole lexical family (suffixed names match by their base). Verified against real Qdrant servers: per-field text indexes and title-gated matches replay in the CI smoke suite.

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "mnemostack"
-version = "1.2.0"
+version = "1.3.0"
 description = "Self-hosted hybrid memory & retrieval for AI apps: semantic + BM25 + temporal + graph recall over Qdrant/Memgraph, RRF fusion, 8-stage ranking, multi-tenant filters, answer synthesis."
 readme = "README.md"
 requires-python = ">=3.10"

src/mnemostack/__init__.py

@@ -6,7 +6,7 @@
 slow and can leave async test runners waiting on unrelated imports.
 """
 
-__version__ = "1.2.0"
+__version__ = "1.3.0"
 
 __all__ = [
     "Config",

tests/test_version.py

@@ -12,3 +12,14 @@ def test_package_version_matches_project_metadata():
 
     assert match is not None
     assert mnemostack.__version__ == match.group(1)
+
+
+def test_current_version_has_dated_changelog_entry():
+    changelog = Path("CHANGELOG.md").read_text()
+    version_heading = re.search(
+        rf"^## \[{re.escape(mnemostack.__version__)}\] - \d{{4}}-\d{{2}}-\d{{2}}$",
+        changelog,
+        re.MULTILINE,
+    )
+
+    assert version_heading is not None