index.html

<!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 dark">
  <title>ScriptBench Leaderboard</title>
  <meta name="description" content="Which model writes My Weird Prompts, and how every candidate scored on ScriptBench, the My Weird Prompts script-writer benchmark.">
  <link rel="stylesheet" href="style.css">
</head>
<body>
  <header>
    <h1>ScriptBench — the My Weird Prompts script-writer benchmark</h1>
    <p class="disclaimer">ScriptBench is our own internal benchmark, open-sourced. It measures one workload: a ~28k-token brief turned into a ~3,800-word, two-host comedic dialogue for text-to-speech. It probably isn't the right benchmark for your use case. Fork it and tailor the task card, rubric and measures to yours.</p>
    <p class="lede">
      <a href="https://myweirdprompts.com">My Weird Prompts</a> is a podcast written entirely by language models.
      This page shows which model writes it now, how candidates scored on
      <a href="https://github.com/My-Weird-Prompts/eval">ScriptBench</a> (formerly MWP-Script; one ~3,800-word,
      two-host script in a single pass), which ones failed on censorship or refusals, and informal
      verdicts on models that wrote aired episodes. The data is read live from the eval repo.
    </p>
    <ul class="links">
      <li><a href="https://myweirdprompts.com/benchmark">Benchmark hub on the site</a></li>
      <li><a href="https://github.com/My-Weird-Prompts/eval">Eval repo (GitHub)</a></li>
      <li><a href="https://huggingface.co/spaces/My-Weird-Prompts/transcripts-explorer">Transcript explorer</a></li>
      <li><a href="https://huggingface.co/datasets/My-Weird-Prompts/model-chronicle">Model chronicle dataset</a></li>
      <li><a href="https://huggingface.co/datasets/My-Weird-Prompts/mwp-script-results">Per-sample results</a></li>
    </ul>
    <p class="small muted related">Related work: <a href="https://arxiv.org/abs/2601.14903">PodBench</a> (Tencent, ACL 2026,
      arXiv 2601.14903; <a href="https://github.com/xucncn/PodBench">code</a>, MIT) is the closest public benchmark in this
      space: 800 samples, inputs up to 21k tokens, multi-speaker podcast scripts, deterministic checks plus an LLM judge.</p>
  </header>

  <main>
    <section aria-labelledby="h-current">
      <h2 id="h-current">Current writer</h2>
      <div id="current" class="card"><span class="muted">Loading…</span></div>
    </section>

    <section aria-labelledby="h-board">
      <h2 id="h-board">Leaderboard</h2>
      <div class="controls">
        <label class="field grow">Run
          <select id="run-picker"></select>
        </label>
      </div>
      <div id="run-head" class="run-head"></div>
      <div class="table-wrap"><table id="board"></table></div>
      <p class="muted small" id="board-foot"></p>
    </section>

    <section aria-labelledby="h-fails">
      <h2 id="h-fails">Technical fails and exclusions</h2>
      <p class="muted small">A <b>technical fail</b> is any refusal, by the model or by a hosting provider's filter
        cutting the output off. A technical fail is never ranked and is out of the production round robin.</p>
      <div class="table-wrap"><table id="fails"></table></div>
      <h3>Exclusions</h3>
      <p class="muted small">Models not screened again, with the reason: technical fails, superseded generations
        (only the newest generation of a family is evaluated), geo-blocks, already aired.</p>
      <div class="controls">
        <label class="field grow">Filter
          <input type="search" id="excl-filter" placeholder="model id or reason">
        </label>
      </div>
      <div class="table-wrap"><table id="exclusions"></table></div>
    </section>

    <section aria-labelledby="h-informal">
      <h2 id="h-informal">Informal evaluations</h2>
      <p class="muted small">Claude reads one or two aired scripts per model and gives a light verdict; Daniel's own
        impressions are quoted where he gave them. Production-trial evidence, not a benchmark.</p>
      <div id="informal"><span class="muted">Loading…</span></div>
    </section>
  </main>

  <footer>
    <p>Source: <a href="https://github.com/My-Weird-Prompts/eval/blob/main/results/leaderboard.json">results/leaderboard.json</a>
      and <a href="https://github.com/My-Weird-Prompts/eval/blob/main/results/informal-evaluations.json">results/informal-evaluations.json</a>,
      rebuilt after every run; <code>git log</code> on them is the history. <span id="generated"></span></p>
  </footer>
  <script src="app.js"></script>
</body>
</html>