Credit ratings: the data

Three small files, shown exactly as the model gets them in the conditions that read CSV. The conditions that read RDF get the same rows as facts.ttl: a straight conversion, one resource per row and one property per column, with nothing merged or inferred.

Some details are deliberate:

• Northwind Holdings has two records (E02, E03) with different names and one LEI. Its latest action sits on the other system's record.
• Halden Bank has a RESEARCH record (E17) that puts it in Luxembourg, while its RATINGS record says Germany. By the local rule, RESEARCH holds for country.
• Brindle Capital (E16) is a QA test record, with rating actions that look real.
• Valdoria's Country Ceiling sits in the same table as the ratings.
• Marlow Utilities' actions are out of date order.
• Instruments pay in local codes (XVD, CVK), and one has only an expected rating (EX).

The R10 executed condition gets different tables, written by execute.py: one row per entity with its records listed, ratings with the dates they held, each instrument with whose risk it carries, the ceilings, and every entity's ancestors.

The base data plus 300 invented issuers, written by scale.py from a fixed seed. The added issuers repeat the base data's traps: some have a RESEARCH record with a different country or holding the latest action, some are QA test records with actions of their own, and there are group chains, finance subsidiaries with guaranteed notes and credit opinions, instruments in local currency, expected ratings, withdrawals, Watches, outlook changes, and rows out of date order.