index.html

<!doctype html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Agent Reliability Lab · Cloud Lab Works</title><link rel="stylesheet" href="style.css"></head><body>
<header><p class="eyebrow">CLOUD LAB WORKS / EXPERIMENT 01</p><h1>Watch it fail.<br>Check the recovery.</h1><p class="lead">Four controlled tool failures. One policy change at a time. Inspect what happened before believing the verdict.</p><div class="badge">Synthetic MCP reference-client runs · no language model</div></header>
<main><section class="toolbar"><label>Experiment <select id="fixture"></select></label><button id="download">Download evidence</button><label class="upload">Compare a cloud run<input id="upload" type="file" accept=".json,application/json"></label></section>
<p id="status" role="status">Loading recorded evidence…</p><p class="recorded-links">Verified cloud runs: <a href="https://huggingface.co/spaces/cloudlabworks/agent-reliability-lab/resolve/main/google-suite.json?download=true">Google JSON</a> · <a href="https://huggingface.co/spaces/cloudlabworks/agent-reliability-lab/resolve/main/aws-suite.json?download=true">AWS JSON</a> · <a href="https://huggingface.co/spaces/cloudlabworks/agent-reliability-lab/blob/main/README.md">Source and reproduction</a>. Download a run, then choose “Compare a cloud run”.</p><section id="comparison" hidden></section><section id="experiment"></section>
<section class="limits"><h2>What this proves</h2><p>A fixed official-SDK client handled these injected faults according to the recorded policy. It does not prove a production agent is reliable, or that one cloud is faster.</p><p>Clinic findings are inspection leads. Legitimate retries can still trigger its loop detector. Success comes from the fixture's checks, not from a clean-looking report.</p><p>Uploaded comparison files stay in your browser. Runtime names in a file are declarations, not independent proof of deployment.</p></section></main>
<footer><span>Cloud Lab Works · infrastructure designed to be run.</span><span id="provenance"></span></footer><script type="module" src="app.js"></script></body></html>