# Conversation Runs and Evaluation Control

> A deep dive into Gaia 2.8’s shift toward repeatable conversation runs and more controlled evaluation workflows.

## Metadata

- Published: 2025-10-24
- Author: Gaia team
- Post type: deep_dive
- Tags: evaluations, conversation runs, quality, governance

# Gaia 2.8 — Conversation Runs and Evaluation Control

As AI systems mature, quality can’t rely on ad-hoc review.

With **Gaia 2.8**, conversations become **repeatable runs**, and evaluation workflows gain finer control.

---

## The Problem: Manual Reviews Don’t Scale

Teams quickly hit limits when quality checks rely on:

- one-off manual reviews,
- informal scoring,
- and inconsistent evaluation settings.

Gaia 2.8 addresses this by introducing **conversation runs** and more structured evaluation controls.

---

## Conversation Runs — Repeatability by Design

**What shipped**

Gaia 2.8 introduces **conversation runs**, enabling teams to:

- launch evaluations on conversation sets,
- monitor run status,
- and inspect outcomes consistently.

**Why this matters**

Runs turn evaluation into a repeatable process.
They help teams:

- track changes over time,
- compare configurations reliably,
- and reduce evaluation drift.

---

## Fine-Grained Overrides — Human Judgment Where It Counts

**What shipped**

Gaia 2.8 adds **turn-level human overrides** in evaluation workflows, allowing reviewers to adjust scores where automation falls short.

**Why this matters**

AI evaluation is useful, but not perfect.
Human overrides ensure:

- accuracy in edge cases,
- accountability for decisions,
- and alignment with real-world expectations.

---

## Structured Judging — Context That Matters

**What shipped**

Gaia 2.8 introduces context-aware judging and clearer evaluation metadata, improving how results are interpreted.

**Why this matters**

Evaluations are only as good as their context.
Better judging ensures scores reflect real conversation dynamics, not isolated turns.

---

## Looking Ahead

The next release expands evaluation governance with **audit trail coverage** and deeper delivery lifecycle tooling.

Gaia 2.8 makes evaluation repeatable. **The next release makes it accountable.**