Structural

Structural metrics are deterministic checks on model outputs (and golden answers where noted): syntax validation (JSON, Python, SQL, XML), schema conformance, readability and overlap scores, type and format checks, counts, and equality against a reference. They do not use an LLM for scoring. Each page lists its shortname, fields, and an example payload (and optional metric_args when the metric supports them).

Metrics

The same pages appear under Structural in the docs sidebar:

• Alpha percentage (alpha_perc)
• Alphanumeric percentage (alphanum_perc)
• BLEU score (bleu)
• Character count match (char_ct_match)
• Exact match (exact_match)
• Is boolean (is_boolean)
• Is date (is_date)
• Is numeric (is_numeric)
• Is string (is_string)
• Is valid JSON (is_valid_json)
• Is valid Python (is_valid_python)
• Is valid SQL (is_valid_sql)
• Is valid XML (is_valid_xml)
• JSON equal (json_equal)
• JSON schema match (json_schema_match)
• Numeric match (numeric_match)
• Paragraph count match (par_ct_match)
• ROUGE score (rouge)
• Sentence count match (sent_ct_match)
• Text readability (text_readability)
• Word count match (word_ct_match)
• XML equal (xml_equal)
• XML schema match (xml_schema_match)