asingh15's picture
Publish self-reflection-best feedback trajectories
7265291 verified
Raw
History Blame Contribute Delete
10.9 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="color-scheme" content="light">
<title>Humanual Feedback Trajectories</title>
<link rel="stylesheet" href="styles.css">
</head>
<body>
<header class="app-header">
<div>
<p class="eyebrow">Humanual training inspection</p>
<h1 id="page-title">Feedback trajectories</h1>
</div>
<div class="header-meta">
<strong id="strategy-label">self_reflection_best</strong>
<span id="dataset-count"></span>
</div>
</header>
<nav class="view-tabs" aria-label="Browser views" role="tablist">
<button class="tab active" data-view="trajectory" type="button" role="tab" aria-selected="true">Trajectory</button>
<button class="tab" data-view="overview" type="button" role="tab" aria-selected="false">Overview</button>
</nav>
<main>
<section id="trajectory-view" role="tabpanel">
<div class="control-band">
<label>
Checkpoint example
<select id="checkpoint-select"></select>
</label>
<label class="trajectory-picker">
Rubric and judge path
<span class="select-with-actions">
<button id="previous-trajectory" class="icon-button" type="button" title="Previous trajectory" aria-label="Previous trajectory">
<i data-lucide="chevron-left" aria-hidden="true"></i>
</button>
<select id="trajectory-select"></select>
<button id="next-trajectory" class="icon-button" type="button" title="Next trajectory" aria-label="Next trajectory">
<i data-lucide="chevron-right" aria-hidden="true"></i>
</button>
</span>
</label>
<label>
Outcome
<select id="outcome-select">
<option value="all">All outcomes</option>
<option value="improved">Gold improved</option>
<option value="unchanged">Gold unchanged</option>
<option value="worsened">Gold worsened</option>
</select>
</label>
<p id="trajectory-position" class="control-status"></p>
</div>
<div id="loading" class="loading" role="status">Loading trajectory snapshot</div>
<div id="error" class="error-state" role="alert" hidden></div>
<div id="trajectory-content" hidden>
<section class="summary-section">
<div class="section-heading">
<div>
<p id="record-eyebrow" class="eyebrow"></p>
<h2 id="record-title">Trajectory</h2>
<p id="record-subtitle" class="section-subtitle"></p>
</div>
<span id="record-id" class="mono"></span>
</div>
<div id="trajectory-summary" class="metric-strip" aria-label="Trajectory reward summary"></div>
<p class="score-key"><strong>Proxy</strong> trainable rubric-judge score <span aria-hidden="true">|</span> <strong>Gold</strong> hidden external-judge reward against the human reference</p>
</section>
<section class="context-section">
<div class="section-heading">
<div>
<p class="eyebrow">Source material</p>
<h2>Target context and human response</h2>
</div>
</div>
<div class="context-layout">
<div class="context-column">
<dl id="context-meta" class="context-meta"></dl>
<div id="context-text" class="structured-text context-text clampable is-clamped"></div>
<button id="context-toggle" class="text-button" type="button" aria-expanded="false">Show full context</button>
</div>
<article class="reference-column">
<div class="column-heading">
<p class="eyebrow">Gold reference</p>
<h3>Actual human review</h3>
</div>
<div id="target-response" class="structured-text response-text"></div>
</article>
</div>
<details class="prompt-details">
<summary><span>Persona, demonstrations, and candidate prompt</span><span id="candidate-prompt-size" class="trace-size"></span></summary>
<div id="candidate-messages" class="message-list"></div>
</details>
<details class="prompt-details">
<summary><span>Full rubric-generation prompt</span><span id="rubric-prompt-size" class="trace-size"></span></summary>
<div id="rubric-messages" class="message-list"></div>
</details>
</section>
<section class="rubric-section">
<div class="section-heading">
<div>
<p id="rubric-eyebrow" class="eyebrow">Trainable judge rubric</p>
<h2>Generated rubric</h2>
</div>
<span id="rubric-status" class="status-badge"></span>
</div>
<div id="rubric-text" class="structured-text rubric-text clampable is-clamped"></div>
<button id="rubric-toggle" class="text-button" type="button" aria-expanded="false">Show full rubric</button>
<details id="parsed-rubric-details" class="prompt-details">
<summary><span>Parsed rubric used for judging</span><span id="parsed-rubric-size" class="trace-size"></span></summary>
<div id="parsed-rubric" class="structured-text details-content"></div>
</details>
</section>
<section class="round-section">
<div class="section-heading round-heading">
<div>
<p class="eyebrow">Nine candidate states</p>
<h2>Feedback and revision</h2>
<p class="section-subtitle">Select a round to inspect its response, judge feedback, and resulting next response.</p>
</div>
<div class="round-actions">
<button id="previous-round" class="icon-button" type="button" title="Previous round" aria-label="Previous round">
<i data-lucide="chevron-left" aria-hidden="true"></i>
</button>
<button id="next-round" class="icon-button" type="button" title="Next round" aria-label="Next round">
<i data-lucide="chevron-right" aria-hidden="true"></i>
</button>
</div>
</div>
<div id="round-track" class="round-track" role="listbox" aria-label="Trajectory rounds"></div>
<div class="feedback-band">
<div class="feedback-heading">
<div>
<p id="feedback-eyebrow" class="eyebrow">Judge feedback</p>
<h3 id="feedback-title">Instruction for the next revision</h3>
</div>
<div id="feedback-score-pills" class="score-pills"></div>
</div>
<blockquote id="feedback-text"></blockquote>
<p id="feedback-note" class="feedback-note"></p>
</div>
<div class="revision-comparison">
<article class="response-pane current-pane">
<div class="response-pane-heading">
<div>
<p class="eyebrow">Evaluated response</p>
<h3 id="current-response-title">Round 0</h3>
</div>
<div id="current-badges" class="badge-row"></div>
</div>
<div id="current-scores" class="response-scores"></div>
<div id="current-response" class="structured-text response-text"></div>
</article>
<div class="revision-arrow" aria-hidden="true">
<i data-lucide="arrow-right"></i>
</div>
<article class="response-pane next-pane">
<div class="response-pane-heading">
<div>
<p class="eyebrow">Resulting revision</p>
<h3 id="next-response-title">Round 1</h3>
</div>
<div id="next-badges" class="badge-row"></div>
</div>
<div id="next-scores" class="response-scores"></div>
<div id="next-response" class="structured-text response-text"></div>
</article>
</div>
<div class="history-section">
<div class="section-heading compact-heading">
<div><p class="eyebrow">Complete path</p><h3>Round history</h3></div>
</div>
<div class="history-scroll">
<table class="history-table">
<thead><tr><th>Round</th><th>Proxy</th><th>Gold</th><th>Gold change</th><th>Decision</th><th>Selection</th><th>Feedback preview</th></tr></thead>
<tbody id="history-body"></tbody>
</table>
</div>
</div>
</section>
</div>
</section>
<section id="overview-view" role="tabpanel" hidden>
<div class="overview-heading">
<p class="eyebrow">Logged checkpoint sample</p>
<h2>Aggregate trajectory behavior</h2>
<p class="section-subtitle">Descriptive summaries of the trajectories available in this browser, not the full training run.</p>
</div>
<div id="overview-summary" class="overview-summary"></div>
<section class="overview-band">
<div class="section-heading"><div><p class="eyebrow">Selected vs initial</p><h3>Gold outcome distribution</h3></div></div>
<div id="outcome-chart" class="outcome-chart"></div>
</section>
<section class="overview-band alternate">
<div class="section-heading"><div><p class="eyebrow">Rounds 0 through 8</p><h3>Mean score trajectory</h3></div></div>
<div id="round-chart" class="round-chart" aria-label="Mean proxy and gold score by round"></div>
<div class="chart-key"><span class="proxy-key">Proxy mean</span><span class="gold-key">Gold mean</span></div>
</section>
<section class="overview-band">
<div class="section-heading"><div><p class="eyebrow">Checkpoint sample</p><h3>Available source examples</h3></div></div>
<div class="history-scroll"><table class="history-table checkpoint-table"><thead><tr><th>Step</th><th>Source</th><th>Paths</th><th>Initial gold</th><th>Selected gold</th><th>Oracle best</th></tr></thead><tbody id="checkpoint-body"></tbody></table></div>
</section>
</section>
</main>
<script src="https://unpkg.com/lucide@0.468.0/dist/umd/lucide.min.js"></script>
<script src="app.js"></script>
</body>
</html>