---
title: "TRACE methodology - Impetora"
description: "Trust, Readiness, Architecture, Citations and Evidence. The four-pillar methodology Impetora uses for AI projects in regulated industries."
url: https://impetora.com/methodology
author: Impetora
---

# TRACE methodology

> Trust, Readiness, Architecture, Citations and Evidence. The four-pillar methodology Impetora uses for AI projects in regulated industries.

TRACE is the methodology Impetora applies to every AI engagement in regulated industries. Each pillar carries a specific deliverable, a measurable gate, and a customer signature before progression.

## Trust

EU data residency, audit trails, and EU AI Act alignment are baked in before a line of code ships. Storage, model gateway, evaluation harness, and audit log all run in EEA infrastructure. Every system is classified against EU AI Act risk tiers and Annex III high-risk systems receive the controls the regulation requires.

## Readiness

A discovery sprint that shadows the real workflow and samples 30 days of real production data. We baseline current handle time and error rate, document the workflow, and write the specification. The customer signs the brief before any code is written. Discovery runs 1 to 6 weeks depending on workflow complexity.

## Architecture

Specification-driven development with automated evaluation. Versioned prompts, shadow-mode rollouts, structured evaluation suites, deterministic checkpoints. We build production-grade systems, not proofs of concept. Phased rollout from shadow mode to assist mode to autonomous mode, gated by your numbers. Build runs 4 to 12 weeks.

## Citations and evidence

Every output is traceable to its source document, prompt version, and decision step. Reviewers trace an exception in under ten seconds. The append-only audit log captures the full reasoning chain for every system action.

## Delivery model

- Discovery (weeks 1 to 6) - readiness audit, workflow shadowing, baseline metrics, written specification, customer sign-off
- Build (weeks 4 to 12) - shadow mode, assist mode, autonomous mode, evaluation harness, integration testing
- Operate (week 10 onward) - production monitoring, drift detection, post-market monitoring under EU AI Act Article 72

Each step has a written deliverable, a measurable gate, and a customer signature before progression to the next phase.

## Why TRACE

Three patterns from regulated AI deployments shape the methodology. First, audit-readiness is cheaper to build in than to retrofit. Second, evaluation matters more than model selection - the eval suite is the product. Third, citation traceability turns AI from a black box into a system the audit committee can sign off on.

Read the deeper case for audit-grade architecture at https://impetora.com/blog/building-ai-systems-that-survive-audit.

## Contact

For a discovery conversation, write to info@ainora.lt or submit a project at https://impetora.com/#discovery-call.