> ## Documentation Index
> Fetch the complete documentation index at: https://docs.amlok.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Research

> Explore AMLOK's public financial-record corpus, completed model training, measured benchmarks and methods for financial risk intelligence.

AMLOK researches financial-data understanding for business and consumer underwriting and continuous monitoring. This research collection brings together original public records, source-level field definitions, completed baseline training and reproducible financial-document measurements.

## Research foundation

Financial activity appears in several systems. An order becomes an invoice, payment, processor settlement, bank deposit and accounting entry. Resolving these relationships is necessary to measure cash flow, obligations and changes without counting the same economic event several times.

The research covers financial language, document extraction, temporal entity relationships, deterministic financial calculations and outcome modeling. Public financing records provide a source-backed foundation for the tasks their fields support. Transaction histories and borrower-permissioned business systems require their own authorized datasets and evaluation populations.

## Data and completed studies

| Research output | Recorded evidence |
| - | - |
| Business lending records | 2,207,426 SBA loan-program observations; 1,897,410 contain a first-disbursement date. |
| Financing disclosures | 1,049 Canadian-source intervention and financing-disclosure rows. |
| Consumer lending records | 208,818 mortgage originations from the 2024 New York HMDA cohort. |
| Financial statements | 3,905,397 numeric facts from 6,516 SEC filing submissions. |
| Supervised model training | Logistic regression and gradient-boosted trees trained on 19,135 real SBA observations, with separate validation and held-out evaluation cohorts. |
| Document retrieval | TF-IDF and BM25 evaluated on 1,147 FinQA questions from real financial reports. |

Counts retain their source units. A financial fact, disclosure row and loan observation are different objects and are not added into a unique-deal total. The [dataset inventory](https://docs.amlok.io/research/datasets.md) identifies coverage and available data points. The [measured results](https://docs.amlok.io/research/public-results.md) distinguish the completed tasks and their metrics.

## Explore the research

> [card] [Datasets and data points](https://docs.amlok.io/research/datasets.md)
>
> Collected records, field definitions, coverage and labeling workflow.

> [card] [Measured results](https://docs.amlok.io/research/public-results.md)
>
> Real-data model evaluation and financial-document retrieval.

> [card] [Source findings](https://docs.amlok.io/research/public-record-findings.md)
>
> Financing terms, source semantics and observed extraction challenges.

> [card] [Model and training methods](https://docs.amlok.io/research/model-architecture.md)
>
> Financial-language, event and graph research methods.

> [card] [Comparison methodology](https://docs.amlok.io/research/frontier-benchmark.md)
>
> Matched evidence, evaluation populations and scoring protocols.

> [card] [Research artifacts](https://docs.amlok.io/research/reproducibility.md)
>
> Source hashes, result manifests and reproducible methods.

## Scientific scope

The completed supervised models use structured loan fields. The document study measures evidence retrieval. These results do not measure a foundation model, transaction-stream understanding or frontier-model performance. The architecture and comparison pages describe the methods for those separate research tasks. The [model card](https://docs.amlok.io/research/model-card.md) records the scope and limitations of each completed study.

## Product boundary

[Insights](https://docs.amlok.io/pricing/insights.md) performs full underwriting. [Signal](https://docs.amlok.io/pricing/signal.md) monitors changes and includes reassessment. Institutions retain lending decisions, borrower communication and intervention. Research introduces no new paid product or public API schema.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.