compare_company_ratings
ChatGPTCompare two companies' AIR Implied Ratings in the same month and explain the gap in notches. Pass company_id and other_company_id (UUIDs returned by get_entity). as_of_date (YYYY-MM-DD, YYYY-MM or YYYY) defaults to the latest month both companies have scores; both are always compared as of the same month. The comparison runs from other_company_id (start) to company_id (end). Each has air_credit_risk_score (the AIR Risk Score, 0 to 1, higher is riskier), air_rating (the AIR Implied Rating, AAA to C) and notch. - total_notches > 0 means company_id is rated better (safer) than other_company_id; direction is 'higher', 'lower' or 'same'. - bridge steps are each category's gap in notches (positive means company_id is safer on that category) and add up to total_notches. A step's metrics show company_id's value (value) against other_company_id's (value_before). - key_steps: the largest gaps. If bridge is null (the two were scored by different model versions that month), message says so; the ratings and notch gap are still valid.
compare_rating_periods
ChatGPTCompare one company's AIR Implied Rating at two dates and explain the move in notches. Pass company_id (a UUID returned by get_entity) and start_date; end_date defaults to the latest score. Dates are YYYY-MM-DD, YYYY-MM or YYYY, and each resolves to the score for the month containing it, or the latest earlier month (requested_date vs as_of_date). A year alone means January as start_date and December as end_date. Returned fields: - start, end: at each date, air_credit_risk_score (the AIR Risk Score, 0 to 1, higher is riskier), air_rating (the AIR Implied Rating, AAA to C) and notch. - total_notches (fractional), letter_notches and direction: positive is an upgrade. - bridge: the move split exactly into steps whose notches add up to total_notches. Step group 'market' is market repricing (how the market re-priced each factor with the company's financials held fixed, plus a market-wide shift); 'company' is the company's own financials changing, with metrics showing each measure before and after and causes noting known data events; 'bank' is bank-score effects (banks only); 'other' holds a "Scoring differences" step when present: the part of the move that re-scoring with today's model files doesn't reproduce. It is unexplained, not a driver, and bridge.checks.notes gives the detail. - notches_by_group: the market / company / bank / other totals. key_steps: the largest steps. The steps account for the whole move. If bridge is null, message says why; the ratings and notch change are still valid. The first request for a pair of dates can take several seconds.
get_credit_rating
ChatGPTGet the latest AIR Credit Risk Score and AIR Rating for a resolved company. Returned fields: - air_credit_risk_score: AIR's proprietary credit risk score (numeric, 0-1). Refer to this as "AIR Credit Risk Score" when presenting results to the user. - air_rating: closest matching AIR Rating letter grade (e.g. "BBB"). Refer to this as "AIR Rating". Pass company_id (a UUID returned by get_entity).
get_entity
ChatGPTSearch AIR companies by fuzzy name or ticker and return possible entity matches. The response contains a status field: exact_match (one literal name/ticker hit), multiple_matches (the caller MUST ask the user to disambiguate), or no_match. Never auto-select a company from multiple_matches — always confirm with the user first.
get_rating_explanation
ChatGPTExplain the latest AIR credit rating using AIR model drivers and financial context. Pass company_id (a UUID returned by get_entity).
get_rating_history
ChatGPTGet a company's AIR Risk Score and AIR Implied Rating over time. Pass company_id (a UUID returned by get_entity). start_date and end_date are optional (YYYY-MM-DD, YYYY-MM or YYYY); the window defaults to the 5 years up to the latest score. A year alone means January as start_date and December as end_date. A date resolves to the score for the month containing it, or the latest earlier month, never a later one. Returned fields: - points: one per period at frequency (the period's last month; the latest score is always included), each with air_credit_risk_score (the AIR Risk Score, 0 to 1, higher is riskier, e.g. 0.42), air_rating (the AIR Implied Rating, AAA to C, e.g. "BBB"), notch and financials_date. - rating_changes: every month-on-month AIR Implied Rating change in the window, whatever the frequency. - summary: first, latest, best and worst points, plus the notch move and direction. notch is the position on the rating scale (C = 0 ... AAA = 20; investment grade is above 10.5). Notch changes are positive for an upgrade and negative for a downgrade. Scores move monthly even without new financials, because market pricing changes: an unchanged financials_date means the move came from the market, not a new filing. To explain why the rating moved between two dates, call compare_rating_periods.
run_credit_scenario
ChatGPTRun a stateless AIR credit scenario and return predictions plus financial parameters. Results are NOT stored. IMPORTANT: this tool does NOT accept a natural-language question. The CALLER must translate the user's scenario into a changes array before calling. Each item in changes has: - field: one of revenues, cost_of_revenue, operating_expenses, interest_expense, research_and_development, long_term_debt, liabilities, inventory, accounts_payable, net_cash_flow_from_operating_activities, net_cash_flow_from_investing_activities, net_cash_flow_from_financing_activities - change_type: 'percent' or 'absolute' - change_value: number. For percent use a fraction (-0.20 = -20%, 0.05 = +5%). For absolute use the raw dollar amount (e.g. -1000000000 = -$1B). - year: integer >= 1 (default 1 = first projected year) Translation examples: 'revenues fall 20% next year' -> [{field:'revenues', change_type:'percent', change_value:-0.20, year:1}] 'operating expenses up 5%' -> [{field:'operating_expenses', change_type:'percent', change_value:0.05, year:1}] 'long term debt down $1B' -> [{field:'long_term_debt', change_type:'absolute', change_value:-1000000000, year:1}] 'revenues -10% AND opex +5% in year 2' -> [{field:'revenues', change_type:'percent', change_value:-0.10, year:2}, {field:'operating_expenses', change_type:'percent', change_value:0.05, year:2}] If the user's scenario cannot be expressed using these fields, ASK the user to rephrase using one of the supported fields instead of skipping the call. Always pass company_id (UUID from get_entity) plus a non-empty changes array.
Get credit rating
ClaudeGet entity
ClaudeGet rating explanation
ClaudeRun credit scenario
Claude
