# Agent orchestration limits, routing rules and Shopware query prompt wording.
# User-visible chat messages live in config/retriex/chat-messages.yaml.
parameters:
  retriex.agent.config:
    commerce_history_budget_chars: 1000
    product_search_knowledge_chunk_limit: 6
    advisory_product_search_knowledge_chunk_limit: 9
    optimized_shop_query_prefix_pattern: '/^(?:keywords?|suchquery|search\s*query|query)\s*:\s*/iu'
    optimized_shop_query_trim_characters: " \t\n\r\0\x0B\"'`"

    input_normalization:
      enabled: true
      max_input_chars: 500
      max_output_chars: 700
      max_added_tokens: 2
      max_length_ratio_percent: 150
      output_prefix_pattern: '/^(?:normalisiert|korrigiert|corrected|normalized)\s*:\s*/iu'
      placeholder_outputs:
        - 'normalized user input'
        - 'corrected user input'
        - 'user input'
        - 'normalisierte nutzereingabe'
        - 'korrigierte nutzereingabe'
      skip_patterns:
        - '/https?:\/\//iu'
        - '/\bwww\./iu'
        - '/```/u'
      prompt:
        intro: 'Normalize the following user input for RetrieX routing before intent detection.'
        output_format_block: |-
          Output format:
          Return exactly one line: the corrected user input.
        current_user_input_label: 'USER INPUT'
        rules:
          - '- Output only the corrected user input text itself, never the words "normalized user input".'
          - '- Correct only obvious typing mistakes, transposed letters, missing umlauts, spacing, and punctuation that clearly preserve the same meaning.'
          - '- Do not add product names, model numbers, article numbers, measurement values, parameters, brands, or application areas that are not already present in the input.'
          - '- Preserve product names, model numbers, article numbers, chemical symbols, units, pH, Redox, ORP, and measurement values exactly unless only letter casing is corrected.'
          - '- Preserve the input language; do not translate German into English or English into German.'
          - '- Preserve vague references such as "der indikator", "das gerät", "suche im shop", or "dazu" without resolving them from context.'
          - '- If the input is already clear or you are uncertain, return the original input unchanged.'
          - '- No introduction, no explanation, no quotation marks.'

      fuzzy_routing:
        enabled: true
        min_token_length: 4
        medium_token_length: 7
        long_token_length: 11
        max_distance_short: 1
        max_distance_medium: 2
        max_distance_long: 3
        min_similarity_percent: 72
        # Canonical routing terms only, not typo variants.
        # Resolved from config/retriex/vocabulary.yaml view
        # agent.input_normalization_fuzzy_routing_terms.
        # A local terms list may still be added here as an explicit project override.
        vocabulary_views:
          terms: agent.input_normalization_fuzzy_routing_terms

    follow_up_context:
      strong_reference_patterns:
        - '/\bder\s+wert\b/u'
        - '/\bdieser\s+wert\b/u'
        - '/\bdiesen\s+wert\b/u'
        - '/\bdem\s+wert\b/u'
        - '/\bmit\s+welche(?:m|n|r)?\b/u'
        - '/\bwomit\b/u'
        - '/\bdamit\b/u'
        - '/\bdafuer\b/u'
        - '/\bdafür\b/u'
        - '/\bdazu\b/u'
        - '/\bdaraus\b/u'
        - '/\bwelche(?:r|s|m|n)?\s+indikator\b/u'
        - '/\bwelche(?:r|s|m|n)?\s+indikatortyp\b/u'
        - '/\bindikator\s+(?:dafuer|dafür|dazu|hierfuer|hierfür)\b/u'
        - '/\bwelche(?:r|s|m|n)?\s+bereich\b/u'
        - '/\bwelche(?:r|s|m|n)?\s+messbereich\b/u'
        - '/\bwelche(?:r|s|m|n)?\s+grenzwert\b/u'
      explicit_commercial_signal_terms:
        - shop
        - preis
        - preise
        - preisen
        - kostet
        - kosten
        - kaufen
        - bestellen
        - warenkorb
        - lieferzeit
        - verfuegbar
        - verfügbar
        - lager
        - url
        - link
        - artikelnummer
        - sku
        - produktnummer

      commercial_table_follow_up:
        enabled: true
        prompt_patterns:
          - '/\b(?:tabelle|tabellarisch|übersicht|uebersicht|liste|auflistung)\b.{0,80}\b(?:preis|preise|preisen|kosten|kostet|shop)\b/u'
          - '/\b(?:preis|preise|preisen|kosten|kostet|shop)\b.{0,80}\b(?:tabelle|tabellarisch|übersicht|uebersicht|liste|auflistung)\b/u'
          - '/\b(?:mit|inkl|inklusive|plus)\s+(?:preis|preise|preisen|kosten|shopdaten)\b/u'
        history_anchor_patterns:
          - '/\bTestomat(?:®)?\s+\d{3,4}\b/iu'
          - '/\b(?:Indikatortyp|Indikator|Indikatoren|Reagenz|Reagenzien|Zubehör|Zubehoer)\b/iu'
        vocabulary_views:
          table_terms: agent.follow_up_context.commercial_table_follow_up.table_terms
        commercial_terms:
          - preis
          - preise
          - preisen
          - kosten
          - kostet
          - shop
          - shoppreis
          - shoppreise
          - shopdaten
        indicator_marker_patterns:
          - '/\b(?:Indikatortyp|Indikator(?:en)?|indicator(?:\s+type)?|Reagenz(?:ien)?)\b/iu'
        query_template_with_model: '{model} indikator'
        query_template_without_model: 'indikator'
      history_question_pattern: '/^Question:\s*(.+)$/mi'
      history_turn_split_pattern: '/(?=^Question:\s)/m'
      history_question_strip_pattern: '/^Question:\s*.*(?:\R|$)/u'
      context_labels:
        previous_user_question_template: 'Vorherige Nutzerfrage: {question}'
        previous_reference_anchors_template: 'Vorherige technische Referenzanker (nur zur Referenzauflösung, keine Faktenquelle): {anchors}'
        current_follow_up_question_template: 'Aktuelle Folgefrage: {question}'
      reference_anchor:
        product_model_pattern: '/\bTestomat(?:®)?\s+(?:\d{3,4}(?:\s+(?=[A-Z0-9]*[A-Z])[A-Z0-9]{2,8})?|EVO(?:\s+[A-Z]{2,6})?|ECO(?:[-\s]?(?:PLUS|C))?|DUO(?:\s+\d{3,4})?|LAB(?:\s+[A-Z]{2,6})?)\b/iu'
        measurement_value_pattern: '/\b\d+(?:[,.]\d+)?\s*°\s*dH\b/iu'

    final_answer_guard:
      enabled: true
      # Prevents runaway final answers caused by repeated LLM output. This is a
      # runtime safety guard and does not change retrieval, ranking or product matching.
      max_output_chars: 12000
      repeated_line:
        enabled: true
        min_output_chars: 1200
        min_line_chars: 16
        max_line_repetitions: 3
        trailing_window_lines: 220
        ignore_patterns:
          - '/^\s*(?:produkt(?:-|\s)?nummer|artikel(?:-|\s)?nummer|preis|verfügbar|verfuegbar|url|quellen?)\b/iu'
          - '/^\s*(?:status|query|intent|datenbasis|shop(?:-|\s)?suche)\b/iu'

    shop_runtime:
      query_cleanup:
        current_input_preservation:
          enabled: true
          # Terms that must be preserved from the current user input in the final
          # Shopware search query. This prevents short domain terms from being
          # dropped by query optimization or generic cleanup. Adapt vocabulary view
          # defaults or add a local `terms` override instead of changing PHP code.
          vocabulary_views:
            terms: agent.shop_runtime.query_cleanup.current_input_preservation_terms

        stopword_cleanup:
          enabled: true
          min_query_tokens_after_cleanup: 2
          # Plain Shopware text search should contain product-relevant terms only.
          # These terms are UI, instruction, presentation or sorting words and are
          # removed after LLM query optimization. Keep this list simple and local.
          terms: []

        positive_token_filter:
          enabled: false
          min_query_tokens_after_filter: 1
          include_current_input_preservation_terms: true
          include_semantic_shop_search_tokens: true
          include_product_role_terms: true
          allowed_terms: []
          blocked_terms: []
          code_patterns: []
          adjacent_variant_patterns: []
          adjacent_variant_terms: []


      attribute_cleanup:
        enabled: true
        # For direct product/accessory lookups with comparative attribute
        # constraints, keep the concrete product type and application terms but
        # do not send range words/numeric thresholds to the plain text shop
        # search. Example: "Anschlusskabel pH/Redox länger 20m" becomes
        # "anschlusskabel redox" so the shop can return 25m/50m/100m cables.
        min_query_tokens_after_cleanup: 2
        # Direct product/accessory cleanup terms are resolved from
        # config/retriex/vocabulary.yaml. Local lists may still be added here
        # as project-specific overrides, but the default source is vocabulary.
        vocabulary_views:
          product_type_terms: search_repair.direct_product_type_terms
          stop_terms: search_repair.direct_product_attribute_stop_terms
        comparative_constraint_patterns: []


      context_resolution:
        context_usage:
          referential_terms: []
        history_anchor_enrichment:
          enabled: true
          max_query_terms: 2
          template: ''
          vocabulary_views:
            trigger_terms: agent.shop_runtime.context_resolution.history_anchor_enrichment.trigger_terms
          query_terms: []
          query_noise_terms: []
          anchor_patterns: []
        meta_query_guard:
          enabled: true
          cleanup_profile: shop_context_fallback
          context_fallback_enabled: true
          context_fallback_question_limit: 12
          context_fallback_history_budget_chars: 20000
          context_fallback_use_full_history: true
          context_fallback_max_terms: 6
          # Legacy/domain override list. Generic stopwords, user-instruction
          # phrases and presentation terms are provided by profile
          # `shop_context_fallback`. Keep shop/price/domain terms here.
          context_fallback_filter_terms: []
          meta_only_terms: []
        rag_anchor_enrichment:
          enabled: true
          # Enriches overly narrow numeric shop queries with a product anchor from
          # retrieved RAG records when the same record explicitly contains the
          # requested numeric value/unit. This prevents queries such as "0,02"
          # when RAG already identified a concrete product such as a device model.
          min_score: 120
          max_query_terms: 6
          early_chunk_bonus_max: 10
          template: '{anchor} {query} {subject}'
          scores:
            exact_value_with_unit: 120
            exact_value_only: 40
            anchor_bonus: 50
          numeric_focus_patterns: []
          product_title_patterns: []
          anchor_bonus_patterns: []
          subject_terms: []


      result_identity:
        enabled: true
        # Direct product-list answers should only list products whose primary
        # identity (name/URL) matches the requested product type. This prevents
        # devices from being listed as a requested consumable merely because the
        # description mentions such consumables as accessories.
        prefer_primary_identity_matches: true
        compound_prefix_match:
          enabled: true
          # Some Shopware product names combine the requested product type with
          # a noun suffix, for example "pH-Pufferlösung". Keep this list small
          # and explicit so broad direct-result filtering remains safe.
          terms: []
        primary_identity_repair:
          enabled: true
          min_query_tokens_after_cleanup: 2
          # Only used for a retry query when the direct-result guard would
          # otherwise suppress all shop results. Keep product words and context
          # such as brand/pH/Redox, but remove target-device wording that can
          # push Shopware ranking toward devices instead of the requested
          # accessory/consumable.
          stop_terms: []


      answer_constraints:
        length_sort:
          enabled: true
          trigger_patterns: []
          value_patterns: []

        length_filter:
          enabled: true
          min_patterns: []
          max_patterns: []


      direct_answer:
        enabled: true
        max_results: 10
    rag_evidence_guard:
      cleanup_profile: rag_evidence
      # Legacy/domain override list. Generic German stopwords are provided by
      # language cleanup profile `rag_evidence`. Keep RAG/product-role terms here.
      stop_terms:
        - ohne
        - messen
        - messung
        - tester
        - testgerät
        - testgeraet
        - gerät
        - geraet
        - messgerät
        - messgeraet
        - produkt
        - produkte
        - artikel
        - shop
      aggregate_query_patterns:
        - '/\bwie\s+viele\b/u'
        - '/\bwieviele\b/u'
        - '/\banzahl\b/u'
        - '/\bcount\b/u'
        - '/\bgesamtzahl\b/u'
      aggregate_evidence_terms:
        - anzahl
        - gesamtzahl
        - stückzahl
        - stueckzahl
        - count
        - portfolio
        - sortiment
        - bestand
        - bestände
        - bestaende
        - lieferprogramm
      aggregate_answer_evidence_patterns:
        - '/\b(?:anzahl|gesamtzahl|stückzahl|stueckzahl|count)\b.{0,80}\b\d+\b/u'
        - '/\b\d+\s+(?:[a-z0-9+\-]+\s+){0,3}(?:produkte|artikel|geräte|geraete|messgeräte|messgeraete)\b/u'
        - '/\b(?:insgesamt|gesamt|total)\b.{0,80}\b\d+\b/u'
        - '/\b(?:sortiment|portfolio|lieferprogramm)\b.{0,120}\b(?:umfasst|enthält|enthaelt|besteht\s+aus|beinhaltet)\b.{0,80}\b\d+\b/u'
      vocabulary_maps:
        synonyms: agent.rag_evidence_guard.synonyms

    no_llm_fallback:
      max_shop_results: 5
    production_ui:
      shop_results:
        max_cards: 5
    shop_prompt:
      intro: 'Generate a short search query for Shopware 6 from the following user input text.'
      output_format_block: |-
        Output format:
        Keyword1 Keyword2 Keyword3
      recent_conversation_context_label: 'RECENT CONVERSATION CONTEXT'
      current_user_input_label: 'CURRENT USER INPUT'
      rules:
        - '- Output only the final search query.'
        - '- Always convert relevant search terms to their singular form.'
        - '- No introduction, no explanation, no quotation marks.'
        - '- Use only shop-relevant search terms from the user input for a shop search.'
        - '- Maximum 6 search terms, preferably fewer.'
        - '- Remove filler words, polite phrases, and irrelevant words.'
        - '- Preserve product names, brands, model numbers, and compound terms exactly if they are relevant.'
        - '- Preserve the language of the CURRENT USER INPUT for generic product/search terms; do not translate German search terms into English.'
        - '- For German user input, output German shop terms, for example "freies Chlor Messung" instead of "free chlorine measurement".'
        - '- Preserve domain terms from the current user input or resolved context in their original language.'
        - '- Numbers that belong to a product name or model must be preserved when they are present in the CURRENT USER INPUT or a clearly resolved reference.'
        - '- Separate terms using spaces only.'
        - '- If a relevant product name is present, it must be placed at the beginning of the final search query.'
        - '- Try to always identify all products mentioned in the user input text, even in long prompts.'
        - '- Look for terms such as Testomat, Horiba, Tritromat, Pockettester, Redox, ORP, or words like indicator/Indikator.'
        - '- If the current user input is vague or referential, use the recent conversation context only as support.'
        - '- Do not output words that only describe conversation flow, such as "same", "again", "also", or "like above".'
      conversation_context_rules:
        - '- The current user input has highest priority.'
        - '- Use the recent conversation context only to resolve omitted references.'
        - '- Use it only for product carry-over, brand carry-over, model carry-over, or variant follow-ups.'
        - '- Do not revive older products unless the current user input clearly refers to them.'
        - '- If the current input starts a new topic, ignore older product context.'
        - '- Prefer the most recent product reference over older ones.'
      language_preservation:
        enabled: true
        language_markers:
          de:
            - ' ä '
            - ' ö '
            - ' ü '
            - ' ß '
            - ' der '
            - ' die '
            - ' das '
            - ' ein '
            - ' eine '
            - ' einer '
            - ' einen '
            - ' welchem '
            - ' welchen '
            - ' welche '
            - ' welcher '
            - ' kann '
            - ' nutzen '
            - ' zur '
            - ' für '
            - ' fuer '
            - ' messung '
            - ' indikator '
            - ' reagenz '
            - ' chlor '
        translation_replacements:
          de:
            free chlorine: 'freies chlor'
            free chlor: 'freies chlor'
            total chlorine: 'gesamtchlor'
            chlorine measurement: 'chlor messung'
            water hardness: 'wasserhärte'
            measurement: 'messung'
            measuring: 'messung'
            chlorine: 'chlor'
            indicator: 'indikator'
            indicators: 'indikatoren'
            reagent: 'reagenz'
            reagents: 'reagenzien'
            accessory: 'zubehör'
            accessories: 'zubehör'