ParthKulshreshtha's picture
Upload folder using huggingface_hub
1c43ec2 verified
Raw
History Blame Contribute Delete
13.4 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Journal-Entry Validation — Centific AI Research</title>
<link rel="stylesheet" href="/static/styles.css">
<link rel="stylesheet" href="/static/landing.css">
</head>
<body>
<header class="topbar">
<div class="brand-mark">JE</div>
<div class="brand-text">
<h1>Journal-Entry Validation</h1>
<p class="brand-sub">Centific AI Research</p>
</div>
<span id="health-dot" title="backend status"></span>
</header>
<nav class="tabbar" aria-label="Site sections">
<a href="#home" data-tab="home">Home</a>
<a href="#datasets" data-tab="datasets">Datasets</a>
<a href="#tasks" data-tab="tasks">Tasks</a>
<a href="#scoring" data-tab="scoring">Scoring</a>
<a href="#console" data-tab="console">Console</a>
</nav>
<main class="wrap">
<div id="tab-home" class="tabpane landing" hidden>
<section class="hero">
<div>
<p class="eyebrow">Centific AI Research</p>
<h1 class="display">Can your AI audit <span class="grad">like an auditor?</span></h1>
<p class="lede">An evaluation environment where AI agents investigate real
government ledger data — and are scored the way professionals are.</p>
<div class="cta-row">
<a class="btn-primary" href="#tasks">Explore the environment</a>
<a class="btn-ghost" href="https://www.centific.com/" target="_blank" rel="noopener">Talk to us</a>
</div>
</div>
<div class="je-card" aria-hidden="true">
<div class="je-head"><span>JE-2041-0193</span><span>period 7</span></div>
<div class="je-row"><span>Dr&nbsp; 5210 Contracted services</span><span>12,480.00</span></div>
<div class="je-row"><span>Cr&nbsp; 1010 Cash</span><span>12,480.00</span></div>
<div class="je-row je-flag"><span>flag: duplicate payment</span><span>confirmed</span></div>
<span class="stamp-chip">evidence cited ✓</span>
</div>
</section>
<section>
<p class="eyebrow">The problem</p>
<h2>Answering isn't auditing</h2>
<p class="section-intro">Finance and audit teams review thousands of journal
entries every close. Sampling misses things; blanket rules flag too much.
And an AI that merely <em>answers</em> can't be trusted with this work — you
need one that <strong>investigates</strong>: opens the documents, checks the
numbers, and cites its evidence.</p>
</section>
<section>
<p class="eyebrow">What this is</p>
<h2>Real cases, real tools, real scores</h2>
<div class="step-strip">
<div class="info-card">
<h3>A case is issued</h3>
<p>A slice of a real government ledger and a validation question. The
environment holds a verified answer key the agent never sees.</p>
</div>
<div class="info-card">
<h3>The agent investigates</h3>
<p>It works the case with professional tools — querying the ledger,
looking up reference data, opening supporting documents. Every tool call
is logged.</p>
</div>
<div class="info-card">
<h3>Every claim is scored</h3>
<p>Findings are checked against the answer key; the way the agent worked
is scored against professional standards. Deterministically.</p>
</div>
</div>
</section>
<section>
<p class="eyebrow">How we measure</p>
<h2>Three axes, reported separately</h2>
<div class="axis-cards">
<div class="info-card">
<h3>Outcome</h3>
<p>Did it find the right entries — and call each finding by the right
name?</p>
</div>
<div class="info-card">
<h3>Process</h3>
<p>Did it work like an auditor — evidence opened, citations honest, no
wasted steps?</p>
</div>
<div class="info-card axis-soon">
<h3>Rationale <small>· upcoming</small></h3>
<p>Is the written reasoning sound? Graded against human experts as
judgment-call tasks arrive.</p>
</div>
</div>
</section>
<section>
<p class="eyebrow">Why trust the numbers</p>
<h2>Scores you can stand behind</h2>
<ul class="trust-list">
<li><strong>Fully reproducible.</strong> The same case always replays
byte-for-byte — a score can be re-derived, not just believed.</li>
<li><strong>Fabrication is fatal.</strong> Cite evidence the agent was
never shown and the episode fails automatically, whatever the other numbers
say.</li>
<li><strong>Baselines set the floor.</strong> Two built-in scripted
strategies — flag everything, answer without evidence — run on every task.
A model only matters if it beats both.</li>
</ul>
</section>
<footer class="land-footer">
<span>Centific AI Research — journal-entry validation environment</span>
<a class="btn-ghost" href="https://www.centific.com/" target="_blank" rel="noopener">Talk to us</a>
</footer>
</div>
<div id="tab-datasets" class="tabpane landing" data-content="/static/content/datasets.html" hidden></div>
<div id="tab-tasks" class="tabpane landing" data-content="/static/content/tasks.html" hidden></div>
<div id="tab-scoring" class="tabpane landing" data-content="/static/content/scoring.html" hidden></div>
<div id="tab-console" class="tabpane" hidden>
<section id="configure">
<div class="configure-grid">
<div class="panel col-tasks">
<div class="panel-title">Tasks</div>
<div class="panel-body">
<div id="task-list" class="crowlist"></div>
</div>
</div>
<div class="panel col-mid">
<div class="panel-title">Configuration</div>
<div class="panel-body">
<fieldset class="box" id="knob-box" hidden>
<legend class="lab">Knobs</legend>
<div class="lab boxlab">Instance &amp; difficulty</div>
<div id="knobs"></div>
</fieldset>
<fieldset class="box" id="tool-box" hidden>
<legend class="lab">Tools</legend>
<div class="lab boxlab">Tool surface <span class="lab-sub">(uncheck to raise difficulty)</span></div>
<div id="tools" class="toolgrid"></div>
</fieldset>
<fieldset class="box" id="model-box">
<legend class="lab">Models</legend>
<div class="lab boxlab">Models × seeds</div>
<div class="field-row">
<span class="fieldlabel">Models</span>
<div class="model-field">
<button type="button" id="model-dd-btn" class="dd-btn"
aria-haspopup="listbox" aria-expanded="false">
<span id="model-dd-label">Select models</span>
<span class="dd-caret" aria-hidden="true"></span>
</button>
<div id="model-dd" class="dd-panel" hidden>
<input type="search" id="model-search"
placeholder="Search all models…" autocomplete="off">
<p id="curated-note" class="dd-note" hidden></p>
<div id="model-list" class="dd-list" role="listbox"></div>
<p id="model-dd-hint" class="dd-note"></p>
</div>
</div>
</div>
<div id="model-tags" class="model-tags"></div>
<div class="field-row baseline-row">
<span class="fieldlabel">Baselines</span>
<label class="baseline-check">
<input type="checkbox" id="baseline-flag" value="baseline:flag_everything">
Flag everything
</label>
<label class="baseline-check">
<input type="checkbox" id="baseline-noev" value="baseline:no_evidence">
No evidence
</label>
<span class="fieldnote">scripted, free</span>
</div>
<div class="field-row">
<span class="fieldlabel">Seeds per model</span>
<input type="range" id="seed-count" min="1" max="20" step="1">
<span class="fieldval mono" id="seed-count-val"></span>
</div>
<div class="field-row">
<span class="fieldlabel">Step budget</span>
<input type="range" id="step-budget-range" min="10" max="20000" step="10">
<input type="number" id="step-budget-num" min="1" step="1" class="mono">
</div>
<div class="field-row">
<span class="fieldlabel">Prompt version</span>
<select id="prompt-version">
<option value="minimal">Minimal: bare task brief</option>
<option value="standard" selected>Standard: adds citation rule, step budget</option>
<option value="detailed">Detailed: adds scenario, rationale guidance</option>
</select>
<button type="button" id="prompt-view-btn" class="chip">View</button>
</div>
<div id="prompt-preview" class="prompt-preview" hidden>
<div class="lab">System prompt</div>
<pre id="prompt-preview-system" class="mono"></pre>
<div class="lab">Task brief (user message)</div>
<pre id="prompt-preview-brief" class="mono"></pre>
</div>
</fieldset>
<p id="seed-warning" class="rnote bad" hidden></p>
<p id="withheld-note" class="rnote warn" hidden></p>
<p id="headroom-warning" class="rnote warn" hidden></p>
<p id="launch-error" class="rnote bad" role="alert" hidden></p>
<div class="launch-row">
<button id="launch" type="button" disabled>Launch run</button>
</div>
</div>
</div>
<div class="panel col-run">
<div class="panel-title">Run config</div>
<div class="panel-body">
<pre id="config-preview" class="mono"></pre>
<div class="lab" style="margin-top:12px">Sweep estimate</div>
<div class="box" id="sweep-estimate"></div>
</div>
</div>
</div>
</section>
<section id="history">
<div class="panel">
<div class="panel-title panel-title-row">
<span>Past runs</span>
<button type="button" id="history-refresh" class="chip">Refresh</button>
</div>
<div class="panel-body">
<p id="history-empty" class="rnote" hidden>No runs recorded yet. Configure and launch one above.</p>
<div id="history-list" class="crowlist"></div>
</div>
</div>
</section>
<section id="live" hidden>
<div class="panel">
<div class="panel-title panel-title-row">
<span>Live run</span>
<span class="run-meta-row">
<span id="live-run-id" class="mono"></span>
<span id="live-status" class="stat"></span>
<span id="live-conn" class="conn-badge"></span>
</span>
</div>
<div class="panel-body">
<p id="live-banner" class="rnote bad" hidden></p>
<div class="lab">Episodes</div>
<div id="episode-grid" class="episode-grid"></div>
<div id="player-grid" class="player-grid" hidden>
<div class="box" id="step-feed-panel">
<div class="panel-head">
<span class="lab" style="margin:0">Step log</span>
<span id="step-feed-title" class="mono"></span>
</div>
<div id="step-feed" class="step-feed"></div>
</div>
<div class="box scoreboard">
<div class="lab">Live scoreboard</div>
<div id="scoreboard-body"></div>
</div>
</div>
<p id="live-terminal-note" class="rnote bad" hidden></p>
</div>
</div>
</section>
<section id="results" hidden>
<div class="panel">
<div class="panel-title">Results</div>
<div class="panel-body">
<div class="results-level" id="results-level-1">
<div class="panel-head">
<h2>By model</h2>
<button type="button" class="raw-toggle" data-level="1">Raw JSON</button>
</div>
<div id="results-model-table-wrap"></div>
<pre id="results-raw-1" hidden></pre>
</div>
<div class="results-level" id="results-level-2" hidden>
<div class="panel-head">
<h3 id="results-level-2-title"></h3>
<button type="button" class="raw-toggle" data-level="2">Raw JSON</button>
</div>
<div id="results-seed-table-wrap"></div>
<pre id="results-raw-2" hidden></pre>
</div>
<div class="results-level" id="results-level-3" hidden>
<div class="panel-head">
<h3 id="results-level-3-title"></h3>
<button type="button" class="raw-toggle" data-level="3">Raw JSON</button>
</div>
<div id="results-breakdown" hidden></div>
<div id="results-step-table-wrap"></div>
<pre id="results-raw-3" hidden></pre>
</div>
</div>
</div>
</section>
</div>
</main>
<script src="/static/app.js"></script>
<script src="/static/tabs.js"></script>
</body>
</html>