Merge branch 'main' of https://github.com/isc-tdyar/iris-vector-rag

Author: tom-dyar

CHANGELOG.md

@@ -1,83 +1 @@
 # Changelog
-
-## [0.5.14] - 2025-11-24
-
-### Added
-- **DSPy Optimization Support** for entity extraction (Feature 063)
-  - New `optimized_program_path` parameter in `OntologyAwareEntityExtractor.__init__()`
-  - Load pre-trained DSPy programs for 31.8% F1 improvement (0.294 → 0.387)
-  - Library-first design with clean API parameter (not environment variable)
-  - Graceful fallback with clear logging when optimization files unavailable
-  - Zero breaking changes (backward compatible optional parameter)
-
-### Testing
-- 8/8 contract tests passing for DSPy optimization feature
-- 17/17 related entity extraction tests passing
-- 7 integration tests created using iris-devtester v1.5.0
-- All transformer import errors resolved (torch/torchvision compatibility)
-
-### Documentation
-- Usage examples in `examples/optimized_dspy_entity_extraction.py`
-- Complete test results in `docs/features/063-dspy-optimization/`
-- Updated STATUS.md, PROGRESS.md, TODO.md with Feature 063 completion
-
-### Dependencies
-- Updated iris-devtester to v1.5.0 from PyPI (significantly faster)
-- Resolved torch/torchvision compatibility (2.4.0/0.19.0)
-
-### Chore
-- Removed tools/nodejs/node_modules from git tracking (build artifacts)
-
-## [0.5.10] - 2025-11-23
-
-### Documentation
-- Updated CHANGELOG to explicitly document UV environment compatibility fix from v0.5.6
-- Clarified that the `intersystems-irispython` UV incompatibility fix (Issue #5) is already included
-
-### Notes
-- No code changes - this release updates documentation only
-- All fixes from v0.5.9 are included
-- UV compatibility fix was already present in v0.5.6 (commit 478d3f1b) and carried forward
-
-## [0.5.9] - 2025-11-23
-
-### Added
-- **Custom Metadata Filtering** for multi-tenant RAG deployments (Feature 051 User Story 1)
-  - Configure custom filter keys via `storage.iris.custom_filter_keys` in YAML config
-  - Extends default 17 metadata fields with enterprise-specific fields (tenant_id, security_level, department, etc.)
-  - Multi-tenant isolation with tenant-based document filtering
-  - Security classification filtering (confidential, public, etc.)
-  - Departmental access control
-  - MetadataFilterManager class for centralized validation
-
-### Fixed
-- **UV Environment Compatibility** - Fixed intersystems-irispython import failures in UV isolated environments (v0.5.6)
-  - Replaced hardcoded `import intersystems_iris.dbapi._DBAPI as iris` with UV-compatible fallback logic
-  - `get_iris_dbapi_connection()` now uses `_get_iris_dbapi_module()` with robust path resolution
-  - Resolves ImportError in UV's isolated environment even when package is installed
-- **IRIS JSON Filtering** - Implemented LIKE-based pattern matching workaround for IRIS Community Edition
-  - Handles both `"key":"value"` and `"key": "value"` JSON formats (with/without space)
-  - Works around missing JSON_VALUE() and JSON_EXTRACT() SQL functions in Community Edition
-- **UV/pytest iris Module Import** - Fixed intersystems-irispython module caching during pytest collection
-  - Modified `tests/conftest.py` to prioritize .venv site-packages before pytest collection
-  - Resolves import issues for ALL tests using IRIS database connections
-- **similarity_search Parameter** - Added support for both `filter` and `metadata_filter` parameter names for backward compatibility
-
-### Security
-- SQL injection prevention via field name validation (regex + dangerous pattern checks)
-- Values safely escaped with SQL quote doubling
-- Tenant isolation verified via E2E integration tests
-- Whitelist enforcement for default + custom metadata fields
-
-### Testing
-- 13/13 contract tests passing (100% success rate)
-  - 8 unit/contract tests for custom field configuration
-  - 4 E2E integration tests with real IRIS database
-  - 1 MetadataFilterManager unit test
-- Multi-tenant isolation verified with cross-tenant queries
-- LIKE-based JSON filtering tested against real IRIS database
-
-### Performance
-- Filter validation: O(n) where n = number of filter keys (typically < 10)
-- SQL pattern matching: LIKE on serialized JSON (acceptable for metadata queries)
-- Test execution: 3.60 seconds for full test suite