Datasets:

Humanswith-ai
/

contentos-preprint

@@ -4,7 +4,7 @@
   <meta charset="utf-8" />
   <meta name="generator" content="pandoc" />
   <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes" />
-  <title>ContentOS Preprint v1.0.1</title>
   <style>
     /* Default styles provided by pandoc.
     ** See https://pandoc.org/MANUAL.html#variables-for-html for config info.
@@ -238,7 +238,7 @@
 </head>
 <body>
 <header id="title-block-header">
-<h1 class="title">ContentOS Preprint v1.0.1</h1>
 </header>
 <h1
 id="contentos-a-reproducible-bilingual-ai-text-detection-ensemble-with-adversarial-robustness-evaluation">ContentOS:
@@ -270,8 +270,12 @@ Wrong-rate of 4% and median latency of 1.2 seconds on commodity 8-vCPU
 hardware. Earlier 44-text hand-curated smoke (v1.0 paper measurement)
 reported 0.821 EN / 0.837 RU; the 4× expanded battery with proper class
 balance per (lang, genre) cell stabilized the numbers upward.</p>
-<p>On the 300-sample adversarial paired set, ensemble AUROC reaches
-<strong>0.985</strong> (in-distribution human baseline).</p>
 <p>The contribution of this work is <strong>field-leading
 reproducibility</strong>, not state-of-the-art absolute AUROC. Anyone
 can clone the repository, run the regression test in 0.05 seconds, and
@@ -323,8 +327,9 @@ hardware ($25/month) within 90 minutes:</p>
 expanded smoke, 2026-04-29)</li>
 <li><strong>Russian ensemble OOD AUROC: 0.846</strong> (176-sample
 expanded smoke, 2026-04-29)</li>
-<li><strong>English ensemble adversarial AUROC: 0.984</strong> on
-300-sample paraphrase-paired set</li>
 <li><strong>English ensemble p50 latency: 1.2 seconds</strong> (8-core
 CPU, no GPU)</li>
 </ul>

   <meta charset="utf-8" />
   <meta name="generator" content="pandoc" />
   <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes" />
+  <title>ContentOS Preprint v1.0.2</title>
   <style>
     /* Default styles provided by pandoc.
     ** See https://pandoc.org/MANUAL.html#variables-for-html for config info.
 </head>
 <body>
 <header id="title-block-header">
+<h1 class="title">ContentOS Preprint v1.0.2</h1>
 </header>
 <h1
 id="contentos-a-reproducible-bilingual-ai-text-detection-ensemble-with-adversarial-robustness-evaluation">ContentOS:
 hardware. Earlier 44-text hand-curated smoke (v1.0 paper measurement)
 reported 0.821 EN / 0.837 RU; the 4× expanded battery with proper class
 balance per (lang, genre) cell stabilized the numbers upward.</p>
+<p>On the 300-sample adversarial paired set (cross-model paraphrasing
+attack, OOD-augmented baseline), ensemble AUROC reaches
+<strong>0.998</strong> (re-measured 2026-04-29 with current
+calibration). Earlier v1.0 paper measurement was 0.985 — the slight
+increase reflects the intervening calibration tuning between Gap-7 and
+current state.</p>
 <p>The contribution of this work is <strong>field-leading
 reproducibility</strong>, not state-of-the-art absolute AUROC. Anyone
 can clone the repository, run the regression test in 0.05 seconds, and
 expanded smoke, 2026-04-29)</li>
 <li><strong>Russian ensemble OOD AUROC: 0.846</strong> (176-sample
 expanded smoke, 2026-04-29)</li>
+<li><strong>English ensemble adversarial AUROC: 0.998</strong> on
+300-sample paraphrase-paired OOD-augmented set (re-measured
+2026-04-29)</li>
 <li><strong>English ensemble p50 latency: 1.2 seconds</strong> (8-core
 CPU, no GPU)</li>
 </ul>