M&S Data Labs | Enterprise Lite Edition (12.5K) - 13-Column Agentic Data

Engineered for Supervised Fine-Tuning (SFT) and benchmarking Financial Agentic AI systems.

Overview This Technical Data Package contains the Lite Edition (12,501 records) of our production-grade synthetic datasets. It is designed specifically to train Large Language Models to process complex financial logic without the compliance risks associated with real-world data. All entities, identifiers, amounts, events, and outcomes are mathematically simulated.

Key Architectural Features

• 13-Element Cognitive Architecture: Each JSONL line represents a fully self-contained transaction block structured with 1 Executive Narrative layer and 12 deeply nested operational pillars: Narrative, Transaction_Identity, Parties_Entity_Profile, Transaction_Intent, Security_Auth, KYC_AML_Risk, Pre_Execution_State, Agent_Action_Trace, Network_API_Events, Financial_Computation, Policy_Control_Decision, Post_Execution_State, and Ground_Truth_Evaluation.
• 100% Synthetic Origin: Zero real-world customer, account, or production banking records were intentionally used.
• Advanced Quality Assurance: Mathematically validated double-entry logic, strict deduplication, and cross-domain stochastic shuffling applied to prevent LLM overfitting.
• Machine-Ready Splits: Data is pre-partitioned into disjointed Train, Validation, and Test sets formatted specifically for LLM instruction-tuning.

What’s Included in This Package?

• Core Data Assets: 12,501 exact records in standard JSONL format.
• Validation Tools: Comprehensive, machine-generated reports (qa_report.json and financial_invariants_report.json) proving 100% schema compliance, plus a Python-based validation script (ms_validator_sdk.py).
• Technical Documentation: Enterprise Factsheet, Data Dictionary, Scenario Catalog, and Evaluation Rubrics.
• Provenance & Security: Cryptographically hashed assets including metadata, checksums, and the official Privacy Statement.

Licensing & Enterprise Support

• Commercial Usage: This Lite Edition unlocks broad internal commercial usage rights for SFT and benchmarking. Model outputs may be used commercially provided the raw dataset is unrecoverable.
• 30-Day Limited Technical Support: To ensure seamless integration, this package is backed by premium technical email support (covering schema interpretation and JSONL parsing assistance) for a full 30 days post-purchase.
• Contact: For support requests or custom dataset generation inquiries, contact our engineering team directly at: contact@msdatalabs.online