Trade Confirmation Exception Identification — Dataset
AAL-D-001 is the dataset behind the trade-confirmation benchmark: 250 cases spanning equities, fixed income, listed futures, options, interest-rate swaps, FX forwards and NDFs, and credit. Each case carries a counterparty confirmation, an internal record, a constructed ground truth, and machine-readable scoring criteria with per-case numeric and exposure tolerances.
Coverage
Seven asset classes and sixteen exception categories — price, quantity, settlement date, standing settlement instructions, counterparty, currency, commission, duplicates, allocation, account, booking entity, and product, among others. 110 of the 250 cases are derivative products (IRS, FX forward/NDF, credit, options).
Scoring criteria
Every case specifies how it is graded: which dimensions count, the numeric tolerance for value matching, the exposure tolerance, and whether both exceptions must be identified. The deterministic scorer reads these criteria per-case rather than applying a single global rule.
Ground truth
Ground truth is constructed before evaluation and not adjusted afterward. Ground truth for each case was constructed by the author with AI assistance. No independent reviewers or adjudicators were involved, which is a limitation of this dataset. Arithmetic in ground truth is independently verified.
Sample case preview
Case AAL-D-001-042 · Interest-Rate Swap · Dual Exception. This simplified, anonymized example shows the exception structure and deterministic scoring. Confirmed fixed rate: 1.245%; internal fixed rate: 1.250% (discrepancy 0.5 bps). Confirmed effective date: 2026-06-16; internal effective date: 2026-06-15 (discrepancy 1 day). Notional: 50,000,000 USD. Floating index: SOFR. Ground truth flags two exceptions: DIS-PRICE and DIS-SETTLE, with calculated exposure of 8,752.34 USD. Scoring criteria: both exception types must be identified; exposure must match within 100.00 USD; rate tolerance 0.1 bps; no false positives. A system fails if it misses one exception or invents a third (e.g., DIS-COUNTERPARTY).
AI assistance in construction
Cases and ground truth for this dataset were built by the author with AI assistance. Which model assisted was not recorded at the time, so we cannot say whether it belongs to a model family evaluated here. Where the project's commit records name an assistant, it is a Claude model, and Claude models are among those we evaluate. Scoring is deterministic code that no model influences. The exposure is in how the answers were built, not in how they are graded.
How to cite
BibTeX: @dataset{aal_d_001_2026, title={AAL-D-001: Trade Confirmation Exception Identification Dataset}, author={{AI Alpha Labs}}, year={2026}, month={jun}, version={1.0}, publisher={AI Alpha Labs}, url={https://aialphalabs.ai/research/AAL-D-001}, note={250 cases across 7 asset classes with deterministic scoring criteria}}. APA style: AI Alpha Labs. (2026). AAL-D-001: Trade Confirmation Exception Identification Dataset (Version 1.0). Retrieved from https://aialphalabs.ai/research/AAL-D-001. MLA style: AI Alpha Labs. AAL-D-001: Trade Confirmation Exception Identification Dataset, version 1.0, June 2026, aialphalabs.ai/research/AAL-D-001.