upgrading version

Author: PabloCabaleiro

CHANGELOG.md

@@ -1,10 +1,18 @@
 # Changelog
 
+<!-- markdownlint-disable MD024 -->
+
 The format follows the principles of [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) and uses [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased](https://github.com/PabloCabaleiro/pondera/tree/main)
 
-### Added (Unreleased)
+## [v0.4.1](https://github.com/PabloCabaleiro/pondera/releases/tag/v0.4.1) - 2025-09-19
+
+<!-- markdownlint-disable-next-line MD024 -->
+### Changed
+
+- Internal: route all evaluations through the unified `multi_evaluate` function (single-case path now uses the same aggregation pipeline).
+- Prompting: moved part of the prior user prompt into the system prompt and refined the system prompt wording for clarity and consistency.
 
 ## [v0.4.0](https://github.com/PabloCabaleiro/pondera/releases/tag/v0.4.0) - 2025-09-12
 
@@ -18,6 +26,7 @@ The format follows the principles of [Keep a Changelog](https://keepachangelog.c
 - Structured error classes introduced: `RunnerError`, `JudgeError`, `TimeoutError` (subclass of `asyncio.TimeoutError`), and `ValidationError` with wrapping of raw exceptions in runner/judge execution and YAML load path.
 - Basic logging: added standard library logging calls (logger name `pondera`) in core API execution path and simple availability test.
 
+<!-- markdownlint-disable-next-line MD024 -->
 ### Changed
 
 - API now always returns `MultiEvaluationResult` (single run wrapped with one `EvaluationResult`) for a stable schema.
@@ -27,6 +36,7 @@ The format follows the principles of [Keep a Changelog](https://keepachangelog.c
 - Fail-fast on missing criterion scores when per-criterion thresholds provided (no silent 0 default); `compute_pass` now raises `ValidationError`.
 - BREAKING: normalized naming: removed `Judgment.pass_fail`/dual serialization; single boolean field `evaluation_passed` everywhere (tests & artifacts updated, no backward alias).
 
+<!-- markdownlint-disable-next-line MD024 -->
 ### Fixed
 
 - Updated tests to align with unified return type.

README.md

@@ -101,9 +101,9 @@ class ConstantJudge(JudgeProtocol):
 
 ```bash
 # Using uv (recommended)
-uv add 'git+ssh://git@github.com/PabloCabaleiro/pondera.git@v0.3.0'
+uv add 'git+ssh://git@github.com/PabloCabaleiro/pondera.git@v0.4.1'
 # or from source in editable mode
-uv pip install 'git+ssh://git@github.com/PabloCabaleiro/pondera.git@v0.3.0'
+uv pip install 'git+ssh://git@github.com/PabloCabaleiro/pondera.git@v0.4.1'
 ```
 
 The judge uses the pydantic-ai ecosystem. Configure provider credentials via env vars (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `AZURE_OPENAI_API_KEY`, etc.) plus optional `PONDERA_` settings.

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "pondera"
-version = "0.2.0"
+version = "0.4.1"
 description = "YAML-first, pluggable runners & LLM-as-a-judge evaluation framework."
 readme = "README.md"
 requires-python = ">=3.10"

src/pondera/__init__.py

@@ -22,4 +22,4 @@
     "ValidationError",
 ]
 
-__version__ = "0.2.0"
+__version__ = "0.4.1"