Merged in bugfix/dynamic-primary-issues (pull request #669)

Bugfix/dynamic primary issues

* add LOB_RELATIONSHIP prompt

* Add test script

* Integrate prompt

* Update one-to-one passing rules

* Merged main into bugfix/dynamic-primary-issues

* Remove promt

* replace quotes


Approved-by: Alex Galarce

fieldExtraction/src/investment/aarete_derived.py

@@ -40,6 +40,7 @@ def fill_na_mapping(answer_dicts):
                 "AARETE_DERIVED_PROGRAM",
                 "constants/mappings/crosswalk_program_lob.json",
             )
+            answer_dict["LOB_RELATIONSHIP"] = "Exclusive"
 
         # Fill AARETE_DERIVED_LOB from PRODUCT
         if string_utils.is_empty(answer_dict.get("AARETE_DERIVED_LOB")):
@@ -49,6 +50,7 @@ def fill_na_mapping(answer_dicts):
                 "PRODUCT",
                 "constants/mappings/crosswalk_product_lob.json",
             )
+            answer_dict["LOB_RELATIONSHIP"] = "Exclusive"
     return answer_dicts
 
 

fieldExtraction/src/investment/dynamic_funcs.py

@@ -26,20 +26,23 @@ def prompt_dynamic(text, field_prompts, filename):
     return llm_answer_final
 
 
-def add_full_context_field(one_to_one_fields, field_to_add):
+def add_full_context_field(one_to_one_fields, field_to_add, answer_dicts):
     field_to_add.field_type = "full_context"
     field_to_add.relationship = "one_to_one"
 
-    if field_to_add.allow_na:
-        field_to_add.prompt = (
-            field_to_add.prompt
-            + " Only answer if the answer applies to the ENTIRE contract. Otherwise, answer N/A."
-        )
-    else:
-        field_to_add.prompt = (
-            field_to_add.prompt
-            + " Do NOT leave this field N/A. Choose the most appropriate answer based on the context."
-        )
+    # Special Handling for Program and Product:
+    # If there are multiple non-empty AARETE_DERIVED_LOB values in the contract, don't ask for Program or Product full context
+    if field_to_add.field_name in ["PROGRAM", "PRODUCT"]:
+        lob_values = [answer_dict.get("AARETE_DERIVED_LOB", "") for answer_dict in answer_dicts]
+        non_empty_lobs = [lob for lob in lob_values if not string_utils.is_empty(lob)]
+        if len(non_empty_lobs) > 1:
+            return one_to_one_fields
+
+    field_to_add.prompt = (
+        field_to_add.prompt
+        + prompt_templates.FULL_CONTEXT_ADDITIONAL_INSTRUCTIONS(field_to_add.allow_na)
+    )
+
     one_to_one_fields.add_field(field_to_add)
     return one_to_one_fields
 
@@ -94,7 +97,7 @@ def get_dynamic_one_to_one_fields(answer_dicts):
                         field_name=field.base_field,
                     )
                 one_to_one_fields = add_full_context_field(
-                    one_to_one_fields, field_to_add
+                    one_to_one_fields, field_to_add, answer_dicts
                 )
                 continue
 
@@ -113,7 +116,7 @@ def get_dynamic_one_to_one_fields(answer_dicts):
                 file_path=config.FIELD_JSON_PATH,
                 field_name=field.base_field,
             )
-            one_to_one_fields = add_full_context_field(one_to_one_fields, field_to_add)
+            one_to_one_fields = add_full_context_field(one_to_one_fields, field_to_add, answer_dicts)
     return one_to_one_fields
 
 

fieldExtraction/src/investment/file_processing.py

@@ -177,6 +177,9 @@ def run_one_to_n_prompts(filename, exhibit_dict, all_exhibit_headers, constants)
     ################## Crosswalk Fields ##################
     one_to_n_results = aarete_derived.get_crosswalk_fields(one_to_n_results)
 
+    ################## Determine LOB Relationship ##################
+    one_to_n_results = one_to_n_funcs.get_lob_relationship(one_to_n_results, exhibit_dict, filename)
+
     ################## Fill NA Mapping ##################
     one_to_n_results = aarete_derived.fill_na_mapping(one_to_n_results)
 

fieldExtraction/src/investment/one_to_n_funcs.py

@@ -1367,3 +1367,28 @@ def outlier_terms_breakout(outlier_term, filename: str) -> dict:
     llm_response = llm_utils.invoke_claude(prompt, "sonnet_latest", filename)
     outlier_answers = string_utils.universal_json_load(llm_response) or {}
     return outlier_answers
+
+
+def get_lob_relationship(answer_dicts, exhibit_dict, filename):
+    
+    for answer_dict in answer_dicts:
+        # Process relationship if LOB is present and either Program or Product is present
+        if not string_utils.is_empty(answer_dict.get("AARETE_DERIVED_LOB")) and (not string_utils.is_empty(answer_dict.get("AARETE_DERIVED_PROGRAM")) or not string_utils.is_empty(answer_dict.get("AARETE_DERIVED_PRODUCT"))):
+            # Get values
+            exhibit_text = exhibit_dict.get(answer_dict.get("EXHIBIT_PAGE"))
+            programs_and_products = answer_dict.get("AARETE_DERIVED_PRODUCT", "") + "|" + answer_dict.get("AARETE_DERIVED_PROGRAM", "")
+            dynamic_primary_values = f"LOB: {answer_dict.get("AARETE_DERIVED_LOB")}\nPROGRAM: {programs_and_products}"
+            
+            # Prompt LLM
+            prompt = prompt_templates.LOB_RELATIONSHIP(exhibit_text, dynamic_primary_values)
+            llm_answer_raw = llm_utils.invoke_claude(prompt, model_id="sonnet_latest", filename=filename)
+            llm_answer_final = string_utils.extract_text_from_delimiters(llm_answer_raw, Delimiter.PIPE)
+            
+            # Assign answer
+            answer_dict["LOB_RELATIONSHIP"] = llm_answer_final
+
+    return answer_dicts
+
+
+
+

fieldExtraction/src/prompts/prompt_templates.py

@@ -1091,3 +1091,47 @@ Return your answers in valid JSON format using the exact field names from the qu
 
 **Important:** Base your answers ONLY on the outlier terms provided. Do not make assumptions or add information not present in the contract language.
 """
+
+
+def LOB_RELATIONSHIP(exhibit_text, dynamic_primary_values):
+    return f"""You will be presented with an exhibit from a Payer-Provider contract.
+
+[BACKGROUND]
+- In healthcare, there are generally multiple Programs for any given LOB.
+    - For example, under the Medicaid LOB, there can be the Programs CHIP, STAR, Medi-Cal, etc.
+- The Exhibit shown contains a number of reimbursement terms, which apply to different LOB and Programs.
+
+It was previously identified that the Exhibit contains the following LOB and Programs:
+{dynamic_primary_values}
+
+[INSTRUCTIONS]
+- Determine if the Exhibit presents an "Inclusive" relationship or "Exclusive" relationship between LOB and Program.
+- Inclusive Relationship: When the contract refers to an LOB, including (but not limited to) the programs in the list.
+    - Some examples of language indicating an inclusive relationship are: "Medicaid and CHIP", "Medicare, including Medicare Advantage", and "Duals, including all special needs programs"
+- Exclusive Relationship: When the contract refers to an LOB and Programs, but the information applies to JUST those specific programs.
+    - Some examples of language indicating an exlusive relationship are "Medicaid: CHIP", "Medicare advantage only", and "Duals - only the following SNP programs"
+    - If the programs are represented with different sections of the Exhibit under the same higher-level LOB section, this is indicative of an Exclusive relationship.
+- Closely read the text, then write either |Inclusive| or |Exclusive| to indicate the nature of the relationship.
+
+[CONTEXT]
+Here is the Exhibit to be analyzed:
+{exhibit_text.replace('"', "'")}
+
+[OUTPUT]
+Briefly explain your reasoning, then put your final answer in |pipes|.
+"""
+
+
+def FULL_CONTEXT_ADDITIONAL_INSTRUCTIONS(allow_na):
+    if allow_na:
+        return " Only provide an answer if that answer clearly applies to the ENTIRE contract. Otherwise, answer N/A."
+    else:
+        return " Do NOT leave this field N/A. Choose the most appropriate answer based on the context."
+
+
+
+
+
+
+
+