Not sure that all the term frequencies adjustments I coded are being applied? And not sure how to verify it?

[pierpaolocreanza]

So, I defined a name comparison composed of various levels, several of which have term frequency adjustments. However, when I inspect the comparison viewer dashboard I only find evidence of tf adjustments being applied to the first exact level match on full name. How can I be sure that these adjustments are indeed being used to compute the final match score? Or is their absence from the dashboard enough evidence that something isn't working?

My comparison:

name_comparison = cl.CustomComparison(
    comparison_levels=[

        # Null 
        NullLevel("inv_name_clean_final"),

        # Exact match for inv_name_clean_final
        ExactMatchLevel("inv_name_clean_final", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value),

        # Nulls
        NullLevel("last_name"),
        NullLevel("first_name"),
        NullLevel("middle_initial"),

        # First, last name & middle inital must be equal
        And(
            ExactMatchLevel("first_name", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value),
            ExactMatchLevel("last_name", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value),
            ExactMatchLevel("middle_initial")
        ),

        # First & last name must be equal
        And(
            ExactMatchLevel("first_name", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value),
            ExactMatchLevel("last_name", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value)
        ),

        # First name equal to last name of the other record, and last name equal to first name (swapped order)
        ColumnsReversedLevel("first_name", "last_name", symmetrical=True),

        # Custom level for Jaccard or Jaro similarity >= 0.9 on inv_name_clean_final
        CustomLevel(
            sql_condition="""
                JACCARD(inv_name_clean_final_l, inv_name_clean_final_r) >= 0.9
                OR JARO_SIMILARITY(inv_name_clean_final_l, inv_name_clean_final_r) >= 0.9
            """,
            label_for_charts="Jaccard or Jaro >= 0.9 on inv_name_clean_final"
        ),

        # Same last name and similar first name
        And(ExactMatchLevel("last_name", term_frequency_adjustments=True).configure(tf_minimum_u_value=tf_minimum_u_value),
            JaroWinklerLevel("first_name",0.85)
        ),

        # Custom level for Jaccard or Jaro similarity >= 0.9 on last_name and JW>=0.8 on first
        CustomLevel(
            sql_condition="""
                first_name_l IS NOT NULL 
                AND first_name_r IS NOT NULL 
                AND last_name_l IS NOT NULL 
                AND last_name_r IS NOT NULL 
                AND length(trim(first_name_l)) > 0 
                AND length(trim(first_name_r)) > 0 
                AND length(trim(last_name_l)) > 0 
                AND length(trim(last_name_r)) > 0 
                AND (
                    JACCARD(last_name_l, last_name_r) >= 0.9
                    OR JARO_SIMILARITY(last_name_l, last_name_r) >= 0.9
                )
                AND JARO_WINKLER_SIMILARITY(first_name_l, first_name_r) >= 0.85
                """,
            label_for_charts="Jaccard or Jaro >= 0.9 on last_name, JW >=0.85 on first_name"
        ),
        
        # Fallback for all other cases
        ElseLevel()
    ],
    output_column_name="name_comparison",
    comparison_description="Inventor name comparison with various levels for clean and stripped names, first/last name handling, and token reordering"
)

and here are two examples of two levels, both of which should have tf adjustments, but where only the one for the full name is showing up.

[pierpaolocreanza]

Bump this up to see if anyone had encountered this issue where TF adjustments are not applied as expected?