⌘PAPERWORK/MACHINE LAB 09 / SHOWCASE
MIS 752 / AGENTS & WORKFLOWS TRACE EXPLORER

Every action.
Every domino.
Every decision.

An agent doesn't just answer. It thinks, requests a tool, receives an observation, and decides what comes next. Follow the chain—and find exactly where it breaks.

Explore the traces ↗Download raw trace ↓
● SAVED NOTEBOOK OUTPUTNO LIVE MODEL / NO API KEY
RECORDED RUNS03Q1 · Q2 · Q3
MODEL STEPS18Across all runs
FAILED OBSERVATIONS13Explicit tool / parse failures
CONVERGED RUNS0/3No final answer in saved runs
01 / THE EVIDENCE

Inside the loop.

Real traces captured in the uploaded notebook. Select a question to inspect each model output and tool observation.

THOUGHT → ACTION → OBSERVATION ↻
ACTIVE TRACE Q1_METFORMIN

Metformin safety

NOT CONVERGED
12 stages
6 model calls · 0 final answers
!
THE CRITICAL FINDING

“Not in the reference” isn't the same as “tool failed.”

In Q3, the lookup returned missing-policy text while the tool was marked successful. The agent kept trying until its six-step budget expired. Logging made the gap visible—but did not fix it.

02 / HUMAN IN THE LOOP

Who gets paged?

Switch between the notebook's supervision styles and see the review burden on the saved traces.

SUPERVISION POLICY SELECT ONE
CURRENT POLICY OUTCOME FLAGGED
INTERRUPTIONS13
LOG ONLY0
SILENT23

⚠ A missing-reference response can have ok: true, so the flagged policy alone may not catch it.
03 / TOOL PLAYGROUND

Ask the reference.

Try the exact-match lookup behavior from the notebook. This is a local simulation of the tool—not a live AI agent.

REFERENCE_LOOKUP / BROWSER DEMO
OBSERVATION READY
Choose a topic and run lookup.