p61c fix

patch_history/RETRIEX_PATCH_61C_POSITIVE_SHOP_QUERY_TOKEN_FILTER_ON_P60_README.md

@@ -0,0 +1,86 @@
+# RetrieX Patch p61C - Positive Shop Query Token Filter on p60
+
+p61C reapplies the positive Shopware query token filter on the confirmed p60 baseline.
+
+## Why p61C exists
+
+p61B was built on a stale base and reintroduced legacy `agent.no_llm_fallback.product_roles.vocabulary_views.*` paths that had already been removed by p59G. It also did not reliably preserve the p60 referential device anchor in the generated query.
+
+p61C uses the confirmed p60 baseline and keeps the p59G/p60 cleanup intact.
+
+## Goal
+
+The final plain Shopware query should contain only product-relevant tokens:
+
+- product/device/accessory names from the active genre vocabulary
+- explicitly allowed product family/application terms
+- protected short technical terms such as pH/RX/TH/TC/TP/TM when configured
+- model/type/code tokens such as `808`, `300`, `TH2100`, `2x100ml` when they match configured regex patterns
+
+Sentence, relation and RAG-only reference words such as `gemessen`, `beim` or `indikatortyp` must not dominate the shop query.
+
+## Expected example
+
+Input query after p60 referential/RAG anchoring:
+
+```text
+testomat 808 gemessen 300 beim indikator
+```
+
+Final shop query after p61C:
+
+```text
+testomat 808 300 indikator
+```
+
+## Configuration
+
+Primary configuration lives in:
+
+```yaml
+config/retriex/genre.yaml
+parameters:
+  retriex.genre.config:
+    configuration_values:
+      shop_query_runtime:
+        positive_token_filter:
+```
+
+Important fields:
+
+- `enabled`: activates the filter for the active genre.
+- `min_query_tokens_after_filter`: set to `1` so a single valid product token can still replace a noisy query.
+- `allowed_terms`: extra genre-specific product family/application terms.
+- `blocked_terms`: terms that are useful for RAG/reference resolution but poor shop search tokens.
+- `code_patterns`: regex patterns for model/type/article/size tokens.
+- `include_current_input_preservation_terms`: includes configured protected short terms from the shop query preservation surface.
+- `include_semantic_shop_search_tokens`: includes the genre's shop semantic product vocabulary.
+- `include_product_role_terms`: includes the genre's device/accessory role vocabulary.
+
+`agent.yaml` contains only an inactive compatibility fallback for this feature. Runtime values should be maintained in `genre.yaml`.
+
+## Scope
+
+No hard-coded product names or stopword lists were added to PHP. The PHP code only applies the configured positive token filter.
+
+No changes to:
+
+- retrieval ranking
+- prompt rules
+- shop result scoring
+- SearchRepair
+- intent routing
+- product identity matching
+
+## Validation
+
+Run:
+
+```bash
+bin/console mto:agent:config:validate
+bin/console mto:agent:regression:test
+bin/console mto:agent:config:audit-source --details
+bin/console mto:agent:config:audit-patterns --details
+```
+
+The p59G no-LLM legacy paths must remain absent from `agent.yaml`, `genre.yaml` source paths and `governance.yaml` frozen hashes.