Agent trajectory evaluation

에이전트 경로 평가

Evaluate the tool choices and sequence that led to an agent’s final answer.

···
html
<div class="stage"><div class="title">TRAJECTORY EVALUATION</div><div class="canvas"><div class="paths"><div><b>EXPECTED</b><span id="e0">search</span><span id="e1">verify</span><span id="e2">answer</span></div><div><b>ACTUAL</b><span id="a0">search</span><span id="a1">skip</span><span id="a2">answer</span></div></div><div class="finding" id="finding">compare tool steps</div></div></div>
css
.stage{width:min(94vw,820px);height:min(88vh,330px);box-sizing:border-box;padding:clamp(9px,2.4vmin,18px);border:1px solid var(--line);border-radius:13px;background:var(--surface);display:flex;flex-direction:column;gap:clamp(7px,2vmin,14px);font:600 clamp(12px,2.5vmin,16px)/1.28 "Pretendard Variable",Pretendard,-apple-system,BlinkMacSystemFont,"Apple SD Gothic Neo",sans-serif;color:var(--fg);overflow:hidden}.title{color:var(--accent);letter-spacing:.035em;flex:none}.canvas{position:relative;flex:1;min-height:0;display:flex;align-items:center;justify-content:center}.muted{color:var(--muted)}.mono{font-family:ui-monospace,SFMono-Regular,monospace}.box{border:1px solid var(--line);border-radius:8px;background:var(--bg);padding:clamp(5px,1.4vmin,12px);text-align:center}.active{border-color:var(--accent)!important;color:var(--accent)!important;background:color-mix(in srgb,var(--accent) 12%,var(--surface))!important}.canvas{flex-direction:column;gap:clamp(7px,2vmin,15px)}.paths{width:100%;max-width:620px;display:flex;flex-direction:column;gap:7px}.paths>div{display:flex;align-items:center;gap:4px}.paths b{width:clamp(72px,20vmin,115px);color:var(--muted)}.paths span{flex:1;border:1px solid var(--line);border-radius:4px;text-align:center;padding:clamp(4px,1.5vmin,8px) 1px}.paths .on{border-color:var(--accent);color:var(--accent)}.paths #a1.on{border-color:var(--accent-3);color:var(--accent-3)}.finding{color:var(--accent)}
js
let n=0;function compare(){const s=n++%3;for(let i=0;i<3;i++){document.getElementById('e'+i).classList.toggle('on',i<=s);document.getElementById('a'+i).classList.toggle('on',i<=s)}document.getElementById('finding').textContent=s<1?'compare tool steps':'verification step missing'}compare();setInterval(compare,900)

Trajectory evaluation checks the steps and tool calls that produced an agent’s answer, even when the final answer is correct. Comparing expected and actual paths can reveal extra calls or missing verification.

The demo advances two paths to the same answer and highlights a poor tool choice. Avoid forcing one exact path when several valid approaches exist.

When to use

Use it when an agent’s tool safety, efficiency, or sequence matters as much as its answer.

Open as page ↗