edward-izgorodin's picture
Release Evaluation Map 0.2.0
281ef6a verified
Raw History Blame Contribute Delete
8.89 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="description" content="Map an agent-memory evaluation need to public papers, benchmarks, frameworks, and reproducible questions.">
<meta http-equiv="Content-Security-Policy" content="default-src 'self'; script-src 'self'; style-src 'self'; img-src 'self' data:; connect-src 'self'; object-src 'none'; base-uri 'none'; form-action 'none'">
<title>Agent Memory Evaluation Map</title>
<link rel="icon" href="favicon.svg" type="image/svg+xml">
<link rel="stylesheet" href="styles.css">
<script src="app.js" defer></script>
</head>
<body>
<a class="skip-link" href="#evidence">Skip to evidence</a>
<header class="site-header">
<a class="brand" href="#top" aria-label="Agent Memory Evaluation Map home">
<span class="brand-mark" aria-hidden="true">M</span>
<span>Memory Evaluation Map</span>
</a>
<nav aria-label="Primary">
<a href="#evidence">Evidence</a>
<a href="#compare">Compare</a>
<a href="#recipe">Recipe</a>
<a href="SOURCES.md">Sources</a>
</nav>
</header>
<main id="top">
<section class="hero" aria-labelledby="hero-title">
<div class="hero-copy">
<p class="eyebrow">Public evidence · no ranking</p>
<h1 id="hero-title">What should your agent memory actually prove?</h1>
<p class="hero-lede">Start with the capability, not the architecture. Filter public artifacts, compare their evidence boundaries, then export the questions your evaluation must answer.</p>
<div class="hero-actions">
<a class="button primary" href="#evidence">Map the evidence</a>
<a class="button quiet" href="#recipe">Build a recipe</a>
</div>
</div>
<aside class="principle-card" aria-label="Core principle">
<span class="signal" aria-hidden="true"></span>
<p class="principle-label">The core distinction</p>
<p class="principle">“Not evaluated” is not “failed.” “Unknown” is not “absent.”</p>
<p class="principle-note">The map keeps evidence gaps visible instead of converting them into a score.</p>
</aside>
</section>
<section class="metrics" aria-label="Map summary">
<div><strong id="artifact-count">7</strong><span>public artifacts</span></div>
<div><strong id="target-count">8</strong><span>evaluation targets</span></div>
<div><strong>0</strong><span>composite rankings</span></div>
<div><strong id="verified-date">2026-09-20</strong><span>latest source update</span></div>
</section>
<section class="workspace" id="evidence" aria-labelledby="evidence-title" tabindex="-1">
<div class="section-heading">
<div>
<p class="eyebrow">01 · Find useful evidence</p>
<h2 id="evidence-title">Choose what you need to measure</h2>
</div>
<p>Results show scope and source boundaries, not recommendations.</p>
</div>
<div class="controls" aria-label="Evidence filters">
<label class="search-control" for="search">
<span>Search the map</span>
<input id="search" type="search" placeholder="Try temporal, graph, abstention…" autocomplete="off">
</label>
<label for="target-filter">
<span>Evaluation target</span>
<select id="target-filter">
<option value="all">All public evidence</option>
</select>
</label>
<label for="outcome-filter">
<span>Outcome level</span>
<select id="outcome-filter">
<option value="all">All outcome levels</option>
<option value="retrieval">Retrieval</option>
<option value="answer">Answer</option>
<option value="agent_task">Agent task</option>
<option value="mixed">Mixed</option>
<option value="unknown">Unknown / not evaluated</option>
</select>
</label>
<button class="button quiet reset" id="reset-filters" type="button">Reset filters</button>
</div>
<div class="result-toolbar">
<p id="result-summary" aria-live="polite">Loading public evidence…</p>
<div class="view-switch" role="group" aria-label="Evidence view">
<button class="view-button active" type="button" data-view="cards" aria-pressed="true">Cards</button>
<button class="view-button" type="button" data-view="matrix" aria-pressed="false">Matrix</button>
</div>
</div>
<div id="card-view" class="evidence-grid" aria-label="Evidence cards"></div>
<div id="matrix-view" class="matrix-wrap" hidden>
<table>
<caption>Source-backed comparison matrix. Scroll horizontally for all fields.</caption>
<thead>
<tr>
<th scope="col">Artifact</th>
<th scope="col">Memory unit</th>
<th scope="col">Time model</th>
<th scope="col">Update / conflict model</th>
<th scope="col">Outcome</th>
<th scope="col">Code</th>
<th scope="col">Dataset</th>
<th scope="col">License</th>
</tr>
</thead>
<tbody id="matrix-body"></tbody>
</table>
</div>
<div class="empty-state" id="empty-state" hidden>
<p>No artifact matches these filters.</p>
<button class="button quiet" type="button" id="empty-reset">Clear filters</button>
</div>
</section>
<section class="compare-section" id="compare" aria-labelledby="compare-title">
<div class="section-heading">
<div>
<p class="eyebrow">02 · Compare without flattening</p>
<h2 id="compare-title">Put two evidence boundaries side by side</h2>
</div>
<p id="compare-status" aria-live="polite">Choose up to two artifacts from the evidence cards.</p>
</div>
<div id="compare-grid" class="compare-grid">
<div class="compare-placeholder">Select an artifact</div>
<div class="compare-placeholder">Select a second artifact</div>
</div>
</section>
<section class="recipe-section" id="recipe" aria-labelledby="recipe-title">
<div class="section-heading">
<div>
<p class="eyebrow">03 · Make the evaluation reproducible</p>
<h2 id="recipe-title">Build an evaluation recipe</h2>
</div>
<p>Deterministic questions only. No generated vendor recommendation.</p>
</div>
<div class="recipe-layout">
<div class="recipe-builder">
<fieldset>
<legend>Capabilities in scope</legend>
<p class="field-help">Select one or more. The checklist updates immediately.</p>
<div id="recipe-targets" class="check-grid"></div>
</fieldset>
<label for="recipe-name">
<span>Evaluation name <small>(optional)</small></span>
<input id="recipe-name" type="text" maxlength="80" placeholder="e.g. Cross-session support agent">
</label>
<div class="export-actions">
<a class="button primary" id="download-markdown" href="#" download="agent-memory-evaluation.md">Download Markdown</a>
<a class="button quiet" id="download-json" href="#" download="agent-memory-evaluation.json">Download JSON</a>
</div>
</div>
<div class="recipe-output" aria-live="polite">
<div class="recipe-output-header">
<span>Evaluation checklist</span>
<span id="question-count">0 questions</span>
</div>
<ol id="recipe-questions"></ol>
</div>
</div>
</section>
<section class="method-note" aria-labelledby="method-title">
<p class="eyebrow">Interpretation guardrail</p>
<h2 id="method-title">Use the map to plan what to test.</h2>
<p>Reported outcomes remain attached to their original task, data, model, prompts, and system design. Follow the primary sources before citing or comparing a result.</p>
</section>
</main>
<footer>
<p>Public research map · source-check dates are shown on each card</p>
<p><a href="data/artifacts.json">Manifest</a> · <a href="data/schema.json">Schema</a> · <a href="SOURCES.md">Sources</a></p>
</footer>
<noscript>This map needs JavaScript for filtering, comparison, and recipe exports. The source manifest remains available at data/artifacts.json.</noscript>
</body>
</html>