Document cross-project Python support policy

.spellcheck-en-custom.txt

@@ -125,6 +125,7 @@ LLMs
 llms
 LLVM
 lora
+LTS
 Makefiles
 Markdownlint
 Martinoli

docs/python-version-support.md

@@ -0,0 +1,65 @@
+# Python Version Support
+
+This document describes the project Python version support policy.
+
+## Goals
+
+1. Ensure that the project is compatible with latest Python releases.
+2. Ensure that the project is compatible with the latest Python LTS releases.
+3. Ensure reasonable choice of Python versions for users to choose from.
+4. Ensure that CI and engineering resources are not wasted on supporting
+   outdated Python versions.
+
+## Approach
+
+### Minimal Python version
+
+The minimal Python version to support in all projects is determined based on
+the availability of Python versions in popular operating systems (Fedora,
+Debian, MacOS) and products built on top of the packages (RHEL AI).
+
+At the moment of writing, the minimal Python version that all projects should
+support is 3.11.
+
+### Support multiple Python versions
+
+To accommodate users that are not willing to upgrade to the latest Python
+version, projects should strive to support several (upstream Python
+versions)[https://devguide.python.org/versions/] in some capacity. At the very
+least, all projects should run lightweight tests (unit, functional) against all
+versions from the minimal supported version up to the latest version.
+
+At the moment of writing, the versions to support are: 3.11, 3.12, 3.13.
+
+### Dominant Python version
+
+To concentrate resources on the most used Python version, the project should
+pick one Python version that will be used to run all CI jobs. Other Python
+versions may be supported in a more limited capacity (e.g. only running unit
+tests).
+
+In general, there's a single dominant Python version at any given time. During
+adoption of a new dominant version, the project may have to temporarily support
+two dominant versions at the same time. This is a temporary situation and plans
+should be made to get back to the single dominant version as soon as possible,
+to avoid wasting resources.
+
+At the moment of writing, the dominant Python version is 3.11.
+
+### Drop support for old Python versions
+
+To avoid wasting resources on supporting outdated Python versions, projects
+should proactively drop support for old versions below the minimal Python
+version indicated above.
+
+To drop support for a Python version, projects should remove all tests that are
+specific to this version. They should also update `pyproject.yaml` metadata to
+specify the minimal Python version equal to the one indicated above.
+Documentation should be updated to reflect the new minimal version, if needed.
+
+At the moment of writing, the versions to drop support for are: 3.10 or
+earlier.
+
+## Links
+
+- [Upstream status of Python versions](https://devguide.python.org/versions/)