Sync v7 demo reel and revised benchmark charts

index.html

@@ -22,9 +22,9 @@
   <header><div class="topbar"><a class="brand" href="https://autotrust.ai" target="_blank" rel="noopener noreferrer" aria-label="AutoTrust website"><img src="autotrust-mark.svg" width="39" height="39" alt=""><span>AutoTrust</span></a><nav aria-label="Main navigation"><a class="nav-secondary" href="#recordings">Demo reel</a><a class="nav-secondary" href="#scorecard">Scorecard</a><a class="nav-live" href="/play">Try live demo <span aria-hidden="true">↗</span></a></nav></div></header>
   <main>
     <section class="hero" aria-labelledby="page-title"><div><p class="eyebrow">The JEV-27B experience</p><h1 id="page-title">JEV-27B.<br><span>See it in action.</span></h1></div><div class="hero-copy"><p>Explore interactive demos and ten recorded workflows across games, browser tasks, and code.</p><div class="actions"><a class="button primary" href="/play">Try live demo <span class="arrow" aria-hidden="true">↗</span></a><a class="button" href="https://huggingface.co/autotrust/JEV-27B" target="_blank" rel="noopener noreferrer">Get the model <span class="arrow" aria-hidden="true">↗</span></a></div></div></section>
-    <section id="recordings" aria-label="Recorded demonstration reel"><div class="section-label"><strong class="recorded">Recorded demo reel</strong><span>10 workflows · 1080p · 5× demo footage</span></div><div class="player-shell"><video id="reel" controls playsinline preload="metadata" poster="poster.png?v=benchmark-bars-6" aria-label="Ten recorded JEV-27B demonstrations with instrumental music"><source src="reel.mp4?v=benchmark-bars-6" type="video/mp4">Your browser does not support this video. <a href="reel.mp4?v=benchmark-bars-6">Download the MP4.</a></video><p class="video-error" id="video-error" hidden>The video could not be loaded. <a href="reel.mp4?v=benchmark-bars-6">Open the MP4</a> to watch the recording.</p><div class="player-footer"><span>NOW SHOWING <strong id="now-playing">Introduction</strong></span><span class="timer"><span id="current-time">0:00</span> / <span id="total-time">0:41</span></span></div></div><div class="reel-note"><p>Recorded experiments with self-hosted JEV-27B. Edited highlights, with on-screen text and instrumental music.</p><a href="reel.mp4?v=benchmark-bars-6" download="JEV-27B-demo-reel.mp4">Download reel ↓</a></div></section>
+    <section id="recordings" aria-label="Recorded demonstration reel"><div class="section-label"><strong class="recorded">Recorded demo reel</strong><span>10 workflows · 1080p · 5× demo footage</span></div><div class="player-shell"><video id="reel" controls playsinline preload="metadata" poster="poster.png?v=benchmark-bars-7" aria-label="Ten recorded JEV-27B demonstrations with instrumental music"><source src="reel.mp4?v=benchmark-bars-7" type="video/mp4">Your browser does not support this video. <a href="reel.mp4?v=benchmark-bars-7">Download the MP4.</a></video><p class="video-error" id="video-error" hidden>The video could not be loaded. <a href="reel.mp4?v=benchmark-bars-7">Open the MP4</a> to watch the recording.</p><div class="player-footer"><span>NOW SHOWING <strong id="now-playing">Introduction</strong></span><span class="timer"><span id="current-time">0:00</span> / <span id="total-time">0:41</span></span></div></div><div class="reel-note"><p>Recorded experiments with self-hosted JEV-27B. Edited highlights, with on-screen text and instrumental music.</p><a href="reel.mp4?v=benchmark-bars-7" download="JEV-27B-demo-reel.mp4">Download reel ↓</a></div></section>
     <section class="scorecard-section" id="scorecard" aria-labelledby="scorecard-title"><div class="section-heading"><div><p class="eyebrow">A closer look</p><h2 id="scorecard-title">The model scorecard</h2></div><a href="model-scorecard.png" target="_blank" rel="noopener noreferrer">Open full size ↗</a></div><figure><a href="model-scorecard.png" target="_blank" rel="noopener noreferrer" aria-label="Open the full JEV-27B model scorecard"><img src="model-scorecard.png" width="1920" height="1080" loading="lazy" alt="AutoTrust JEV-27B model scorecard. System 1 fidelity: mean KL 0.017, yes/no AUROC 0.995, top-1 agreement 90.5%, rating error 0.098. Calibration error 0.0009, option-shuffle flips 2.9%, unseen-task KL 0.104. System 2 HumanEval pass at 1: 78.0%. See full-size image for methods and qualifications."></a><figcaption>Provided by AutoTrust. Metrics, evaluation settings, and comparison notes are included in the scorecard.</figcaption></figure></section>
-    <section class="benchmark-section" aria-labelledby="benchmarks-title"><div class="benchmark-head"><div><h2 id="benchmarks-title">Across six benchmarks</h2><p>Six-model comparison · September 27, 2026</p></div></div><figure class="benchmark-chart"><a href="benchmark-comparison.png?v=benchmark-bars-6" target="_blank" rel="noopener noreferrer" aria-label="Open the full-size benchmark comparison chart"><img src="benchmark-comparison.svg?v=benchmark-bars-6" width="1920" height="1080" loading="lazy" alt="Six models compared across JevBench, Kev, OpenJev text, Nimble, VitaminC, and MASSIVE-en. JEV-27B is highlighted in gold with an 84.07% six-benchmark average. Each panel uses its labeled axis range."></a><figcaption><a href="benchmark-comparison.png?v=benchmark-bars-6" target="_blank" rel="noopener noreferrer">Open full size ↗</a> · <a href="benchmark-comparison.svg?v=benchmark-bars-6" download="JEV-27B-benchmark-comparison.svg">Download SVG ↓</a></figcaption></figure><p class="benchmark-note">JevBench, Kev, and OpenJev text use averages across task families, splits, and components respectively. Nimble, VitaminC, and MASSIVE-en report per-example accuracy. These results use separate evaluation settings from the scorecard above. <a href="benchmark-notes.md">Read the evaluation notes ↗</a></p></section>
+    <section class="benchmark-section" aria-labelledby="benchmarks-title"><div class="benchmark-head"><div><h2 id="benchmarks-title">Across six benchmarks</h2><p>Six-model comparison · September 27, 2026</p></div></div><figure class="benchmark-chart"><a href="benchmark-comparison.png?v=benchmark-bars-7" target="_blank" rel="noopener noreferrer" aria-label="Open the full-size benchmark comparison chart"><img src="benchmark-comparison.svg?v=benchmark-bars-7" width="1920" height="1080" loading="lazy" alt="Six models compared across JevBench, Kev, OpenJev text, Nimble, VitaminC, and MASSIVE-en. JEV-27B is highlighted in gold with an 84.07% six-benchmark average. Each panel uses its labeled axis range."></a><figcaption><a href="benchmark-comparison.png?v=benchmark-bars-7" target="_blank" rel="noopener noreferrer">Open full size ↗</a> · <a href="benchmark-comparison.svg?v=benchmark-bars-7" download="JEV-27B-benchmark-comparison.svg">Download SVG ↓</a></figcaption></figure><p class="benchmark-note">JevBench, Kev, and OpenJev text use averages across task families, splits, and components respectively. Nimble, VitaminC, and MASSIVE-en report per-example accuracy. These results use separate evaluation settings from the scorecard above. <a href="benchmark-notes.md">Read the evaluation notes ↗</a></p></section>
     <section class="closing" aria-labelledby="closing-title"><div><h2 id="closing-title">Your turn to explore.</h2><p>Try an interactive workflow, or take a closer look at the model.</p></div><div class="actions"><a class="button primary" href="/play">Try live demo <span class="arrow" aria-hidden="true">↗</span></a><a class="button" href="https://huggingface.co/autotrust/JEV-27B" target="_blank" rel="noopener noreferrer">Model on Hugging Face ↗</a></div></section>
     <p class="credits">Reel soundtrack: “Tech Live” by Kevin MacLeod (<a href="https://incompetech.com/music/royalty-free/index.html?isrc=USUAN1700030" target="_blank" rel="noopener noreferrer">incompetech.com</a>), licensed under <a href="https://creativecommons.org/licenses/by/4.0/" target="_blank" rel="noopener noreferrer">Creative Commons Attribution 4.0</a>. Edited excerpt. <a href="THIRD_PARTY_NOTICES.md">Third-party notices</a>.</p>
   </main>
@@ -35,7 +35,7 @@
     let chapters=[];
     const formatTime=value=>{if(!Number.isFinite(value)||value<0)return '—:—';const n=Math.floor(value);return Math.floor(n/60)+':'+String(n%60).padStart(2,'0')};
     function update(){document.getElementById('current-time').textContent=formatTime(player.currentTime);if(Number.isFinite(player.duration))document.getElementById('total-time').textContent=formatTime(player.duration);let active=-1;chapters.forEach((chapter,index)=>{if(player.currentTime>=chapter.start-.01)active=index});if(active>=0)document.getElementById('now-playing').textContent=chapters[active].title}
-    async function loadChapters(){try{const response=await fetch('chapters.json?v=benchmark-bars-6');if(!response.ok)return;const data=await response.json();if(!Array.isArray(data))return;chapters=data.filter(x=>typeof x.title==='string'&&typeof x.start==='number'&&Number.isFinite(x.start)&&x.start>=0).sort((a,b)=>a.start-b.start);update()}catch(error){/* Playback remains available without chapter labels. */}}
+    async function loadChapters(){try{const response=await fetch('chapters.json?v=benchmark-bars-7');if(!response.ok)return;const data=await response.json();if(!Array.isArray(data))return;chapters=data.filter(x=>typeof x.title==='string'&&typeof x.start==='number'&&Number.isFinite(x.start)&&x.start>=0).sort((a,b)=>a.start-b.start);update()}catch(error){/* Playback remains available without chapter labels. */}}
     player.addEventListener('timeupdate',update);player.addEventListener('loadedmetadata',update);player.addEventListener('durationchange',update);player.addEventListener('error',()=>document.getElementById('video-error').hidden=false);loadChapters();
   </script>
 </body>

reel.mp4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07f02985bd964368e9382ba07aedfc4252fe11cb8b7c2ab055ca62d20bbac0bb
-size 8510069
+oid sha256:4b1248959c531276178b712f410a4019ec254ef203332e1003a5230155947da6
+size 8502397