DEV Community

#evaluation

Posts

đź‘‹ Sign in for the ability to sort posts by relevant, latest, or top.
Your Agent Has Observability. It Doesn't Have Evals.

Your Agent Has Observability. It Doesn't Have Evals.

Comments
10 min read
Cosine Gating Won't Save You From Sycophancy: A Self-Refutation Observed by Three Judges

Cosine Gating Won't Save You From Sycophancy: A Self-Refutation Observed by Three Judges

Comments 2
5 min read
My local 7B thinks "kill a Python process" is a violent crime — and my regex beat it

My local 7B thinks "kill a Python process" is a violent crime — and my regex beat it

Comments
4 min read
RAG Evaluation 2026: The Four Core Metrics and How to Read Them Diagnostically

RAG Evaluation 2026: The Four Core Metrics and How to Read Them Diagnostically

Comments
4 min read
Using Execution Traces to Evaluate AI Agent Behavior

Using Execution Traces to Evaluate AI Agent Behavior

2
Comments
5 min read
Green Scores Without a Pinned Harness Are Screenshots

Green Scores Without a Pinned Harness Are Screenshots

1
Comments
4 min read
AI Agent Evaluation End to End: How the LFORLA Drone Build Benchmark Scores Planning, Sourcing, and Assembly

AI Agent Evaluation End to End: How the LFORLA Drone Build Benchmark Scores Planning, Sourcing, and Assembly

2
Comments 2
4 min read
7 Agent Eval Mistakes That Cost Me Weeks (And the One-Line Fixes That Ended Them)

7 Agent Eval Mistakes That Cost Me Weeks (And the One-Line Fixes That Ended Them)

24
Comments 5
5 min read
El modelo encontró evidencia relevante y aun así falló: por qué “alucinación” se me quedó corta

El modelo encontró evidencia relevante y aun así falló: por qué “alucinación” se me quedó corta

Comments
4 min read
The benchmark that disproved its own result

The benchmark that disproved its own result

Comments 1
6 min read
How to Evaluate AI Agents

How to Evaluate AI Agents

2
Comments
7 min read
JuryTrace: make agent-judge failures inspectable

JuryTrace: make agent-judge failures inspectable

Comments 1
4 min read
My board never scored an outage as a regression. My evidence couldn't prove it.

My board never scored an outage as a regression. My evidence couldn't prove it.

Comments
5 min read
We spent two days bisecting a prompt change. The regression was noise.

We spent two days bisecting a prompt change. The regression was noise.

Comments
1 min read
Nineteen of twenty patched the symptom

Nineteen of twenty patched the symptom

Comments 2
9 min read
đź‘‹ Sign in for the ability to sort posts by relevant, latest, or top.