# RAG 2.0: from retrieval to governed context

> RAG 2.0 shifts from simple retrieval to governed context with ownership, minimal context, and operational quality.

- Author: Viktor Berthelius (BRTHLS)
- Published: 2026-03-11
- Updated: 2026-06-29
- Category: ai operating models
- Language: en
- Canonical: https://www.brthls.com/magazine/rag-2-0-governed-context-en
- Source: BRTHLS Magazine — https://www.brthls.com

---

## Problem

Many companies say they have RAG because they connect a search engine to a model. But in practice, the context remains dirty, disordered, and without ownership.

The result is predictable: inconsistent answers, dubious decisions, and a false sense of control.

## Thesis

RAG 2.0 is not just retrieval. It is governed context: sources, permissions, versioning, and quality criteria.

> **Callout —** Without context governance, RAG only amplifies noise.

## Framework

Three layers that turn RAG into a system:

- **Fuentes gobernadas:** ownership, permissions and versioning of each source.
- **Contexto minimo:** what the agent must know to decide confidently.
- **Calidad operativa:** evaluation criteria per source and per use case.

Mini‑case: a team connected 12 sources to their RAG. The answers improved in volume but not in reliability. By reducing to 4 sources with clear ownership and quality criteria, precision rose and reversions dropped.

**Anti‑example:** thinking that more documents = more intelligence.

**Posture:** RAG without governance is not an advantage. It is operational risk.

**Breathing:** In practice, the problem is not the model. It is the dirtiness of the context.

## Protocol (3 steps)

1. **Define ownership de fuentes:** each source has a responsible party and clear permissions.
2. **Fija contexto minimo:** for each use case, what context is mandatory.
3. **Evalua calidad por fuente:** if a source fails two cycles, it is corrected or removed.

| Signal | Metric | Threshold |
| --- | --- | --- |
| Source quality | % valid answers | > 90% |
| Reversion | % decisions reverted | must fall cycle to cycle |
| Coverage | % cases with minimal context | > 95% |

<details>
<summary>Quick checklist for governed RAG</summary>

- Does each source have an owner and permissions?
- Is minimal context defined per case?
- Are quality criteria defined per source?

</details>

Related:
- [Context Architecture: from loose prompts to knowledge operating system](/magazine/context-architecture-from-prompts-to-knowledge-os-en)
- [Context Architecture: why prompt engineering doesn't scale business](/magazine/context-architecture-prompt-engineering-no-escala-negocio-en)
- [10 mistakes that sink AI initiatives in mid-sized companies](/magazine/ai-initiative-mistakes-mid-sized-en)

## Next step

If your RAG today answers but doesn't decide well, schedule a diagnosis at [contact](/en/contact).

## Related signals

The operational difference appears when the team connects context, criteria and cadence in the same decision system. The operational difference appears when the team connects context, criteria and cadence in the same decision system. The operational difference appears when the team connects context, criteria and cadence in the same decision system. The operational difference appears when the team connects context, criteria and cadence in the same decision system.

---

*Translated from the Spanish original with AI assistance and reviewed for accuracy. [Read the original in Spanish](/magazine/rag-2-0-de-recuperacion-a-contexto-gobernado).*

---

_Cite as: Berthelius, V. (2026). "RAG 2.0: from retrieval to governed context". BRTHLS Magazine. https://www.brthls.com/magazine/rag-2-0-governed-context-en_