RATIO
An Architectural Study of Context-Aware Artificial Intelligence in Enterprise Systems
Abstract
RATIO is a design-science technical report proposing a reference architecture for context-aware AI in enterprise systems. It separates model-assisted interpretation and planning from identity, permissions, evidence retrieval, deterministic validation, approval, controlled execution and verification. A governed task contract defines what an agent may access and do, while a synthetic financial-reconciliation example illustrates the difference between explaining a discrepancy and authorizing a change.
The report addresses prompt injection, stale data, transaction failures, human oversight and operating cost, and proposes a comparative evaluation protocol. It is grounded in practitioner experience; it is not peer reviewed and reports no completed comparative experiments or measured performance results.
Themes
The report in three themes.
The report follows a request from interpretation to a verified outcome, and asks who holds authority at each step.
Reasoning and authority
The model interprets requests and proposes plans. Application policy decides what may be accessed or changed, with risk-based approval requirements and deterministic validation.
Evidence and transaction integrity
Scoped retrieval, source provenance, approvals tied to specific record versions, controlled execution and postcondition checks make the path from request to outcome inspectable.
Evaluation and accountability
A proposed comparative protocol evaluates final states, policy compliance, failure recovery, human effort and cost per verified success. The evaluation has not been completed.
Status and scope
What the report is, and what it is not.
RATIO proposes an architecture. It does not report how any product performs, Ancilor included.
Record
- Author
- Argjend Bajraktari
- ORCID
- 0009-0009-6135-7509
- Report type
- Design-science technical report
- Version
- 1.0, final manuscript dated 3 November 2026
- Published
- Zenodo, 3 November 2026
- DOI
- 10.5281/zenodo.23911976
- All versions
- 10.5281/zenodo.23911975
- Licence
- Creative Commons Attribution 4.0 International
- Peer review
- Not peer reviewed
- Evidence
- No completed comparative experiments or measured performance results are reported.
Suggested citation
Bajraktari, A. (2026). RATIO: An Architectural Study of Context-Aware Artificial Intelligence in Enterprise Systems (Version 1.0) [Technical report]. Zenodo. https://doi.org/10.5281/zenodo.23911976
Contribution
An architectural synthesis, a governed task contract, a synthetic reconciliation example and an evaluation protocol. The report does not claim a new foundation-model training method, a formal security proof, certified regulatory compliance or measured superiority over other systems.
Materials
No company dataset, production source code, benchmark fixture package or empirical-results dataset accompanies this version. RATIO names the public research framework; it is not a claim that every component exists as a separately marketed product.
Interest
The author works in enterprise AI and has a commercial interest through Harrimont. The report presents a practitioner perspective, not an independent vendor evaluation.
Full text
Read the full report.
The versioned Zenodo record holds the report, its DOI and the publication metadata. The report includes the architecture, the worked example, the threat model, the evaluation design, references and author declarations.
Open access · English · CC BY 4.0