p43N

patch_history/RETRIEX_PATCH_43N_LANGUAGE_CLEANUP_GROUP_SETS_README.md

@@ -0,0 +1,46 @@
+# RetrieX Patch 43N - Language Cleanup Group Sets
+
+## Goal
+
+Reduce repeated cleanup-profile group compositions in `language.yaml` without changing effective cleanup behavior.
+
+## Changes
+
+- Added optional `stopword_group_sets` and `phrase_group_sets` to `config/retriex/language.yaml`.
+- Moved repeated `de_core` + `conversation` profile composition into `stopword_group_sets.de_conversation`.
+- Moved repeated `user_instruction` phrase composition into `phrase_group_sets.user_instruction`.
+- Updated `LanguageCleanupConfig` to resolve group sets before local profile groups.
+- Existing `stopword_groups`, `phrase_groups`, and `meta_term_groups` remain supported.
+
+## Non-goals
+
+- No new domain/runtime logic.
+- No scoring changes.
+- No prompt-rule changes.
+- No retrieval changes.
+- No admin UI changes.
+- No new hard-coded fachliche lists in PHP core.
+
+## Expected effective behavior
+
+The effective cleanup profile outputs remain identical to p43M:
+
+- `commerce_query`: same stopwords, phrases, meta terms, protected terms.
+- `rag_evidence`: same stopwords, phrases, meta terms, protected terms.
+- `retrieval_reference_cleanup`: same stopwords, phrases, meta terms, protected terms.
+- `shop_context_fallback`: same stopwords, phrases, meta terms, protected terms.
+
+## Local checks performed
+
+```bash
+php -l src/Config/LanguageCleanupConfig.php
+php -l src/Config/GovernanceConfig.php
+php -l src/Config/AgentRunnerConfig.php
+php -l src/Config/PromptBuilderConfig.php
+php -l src/Config/SearchRepairConfig.php
+python3 YAML parse check for config/retriex/*.yaml
+python3 effective p43M-vs-p43N cleanup profile comparison
+php LanguageCleanupConfig effective resolver comparison
+```
+
+The local `bin/console` checks could not be executed in this artifact environment because the ZIP does not include `vendor/` and Composer dependencies are unavailable here.