Benchmark
Categories
Every verified question belongs to one category. Each page defines what the category tests, ranks the models on it, lists the hardest questions and the most common failures, and can be followed for changes.
| Category | Tests | Verified questions | Frontier average accuracy |
|---|---|---|---|
| Fact extraction | Reading a single stated fact from the record — an issuer, a date, a denomination — without inference. | 56 | 88% |
| Maturity | Determining an obligation's maturity and testing it against the maximum-maturity characteristic, including amended or extended maturities. | 26 | 100% |
| Currency | Identifying the currency of denomination and testing the specified-currency characteristic, including redenominations and multi-currency series. | 28 | 100% |
| Seniority | Determining ranking (senior, subordinated, secured) from the terms, and testing the not-subordinated characteristic. | 30 | 100% |
| Transferability | Deciding whether an obligation is transferable without consent, including transfer restrictions buried in exchange-offer or restructuring terms. | 30 | 83% |
| Multi-document | Reconciling documents that disagree — a supplement that amends terms, a notice that contradicts a prospectus — and identifying which governs. | 2 | 100% |
| Deliverability | The full deliverable-obligation analysis: every characteristic tested together on one instrument and one credit event. | 30 | 77% |
| Uncertainty | Recognising when the record does not settle the question, and abstaining rather than guessing. | 32 | 69% |
| Restructuring | Applying the restructuring credit-event definition to what actually happened: binding on all holders, deterioration in creditworthiness, the listed changes. | 2 | 67% |
| Entity match | Matching the obligor in the documents to the reference entity, including successor and guarantor questions. | 30 | 98% |
| Evidence | Citing the sections of the record that actually support the answer, and no others. | 30 | 33% |