# Research that ships

> Sireto's research follows the problems NormLogic has to solve: reliable requirement extraction, canonical regulatory models, agents that operate against rules, temporal and versioned regulation, auditable reasoning, and human-AI modelling workflows.

Canonical: https://sireto.com/research

## Six fields, one problem

Artificial intelligence, software engineering, knowledge representation, legal informatics, formal methods, distributed systems. The objective is systems that can be understood, tested and trusted.

## Open questions

- Reliable requirement extraction: making AI-assisted extraction of obligations, permissions, conditions and exceptions measurable and repeatable, with datasets and benchmarks.
- Canonical regulatory models: representations expressive enough for legal nuance yet precise enough to execute (LegalRuleML, Akoma Ntoso, domain-specific languages).
- Agents that operate against rules: architectures where an agent's actions are checked against explicit structured regulatory models and each decision carries its justification.
- Temporal and versioned regulation: which version applied when, what an amendment changed, and how change propagates.
- Auditable reasoning: a chain of evidence from source provision to decision.
- Human-AI modelling workflows: how experts and AI systems build and maintain regulatory models together (NormFlow).

## Method

1. Explore an emerging idea against a real regulatory problem.
2. Build a reference implementation, evaluate it systematically, publish what we can.
3. Ship what works into NormLogic or client systems.