Session 4: TRACE Evals

10 lessons · Back to full syllabus

What you keep

The full TRACE loop as a system you build - Trace, Read, Analyze, Codify, Enforce. Code assertions first; validate every LLM judge with TPR/TNR.

You ship

An eval suite your capstone runs against, visible from a Streamlit UI.

Live session resources

Lessons

Live

The vibe-check trap

"It works on the examples I tried" is not evaluation - and for non-deterministic systems it is dangerous.

Read lesson
Live

Trace and Read: error analysis

Capture full records, then read traces by hand - qualitative research on real failures.

Read lesson
Live

Analyze: cluster, count, prioritize

Binary pass/fail judgments, ranked by frequency and impact - the spec for your eval tooling.

Read lesson
Live

Codify: build the eval suite

Code-based assertions plus validated LLM-as-judge - an unvalidated judge is just another vibe check.

Read lesson
Live

Enforce: evals in the loop

Wire evals into the development loop so every change gets measured automatically.

Read lesson
Live

Evaluating agents: trajectories, not just final answers

A polite final message can hide a wrong tool call, a silent retry loop, or an action that never actually happened.

Read lesson
Deep dive

Aligning an LLM-as-judge to human labels

The actual calibration workflow: blind labels, held-out validation, TPR/TNR, and re-validating after every model change.

Read lesson
Deep dive

Agent trajectory and world-state grading

Score the path the agent took and what actually changed in the environment, not just what it said.

Read lesson
Async

Session 4 assignment support: Build and run TRACE on your capstone

Path A: traces, taxonomy, code assertions, one fix, Streamlit. Path B: validated judge, goldens, CI sketch.

Read lesson
Assignment

Build and run TRACE on your capstone

Trace, read, analyze, codify, and wire evals to run on demand - show a metric moving after a fix.

Read lesson