Download DraftMark_Local_AI_Detection_Report.html from saksornr/draftmark-roberta-int8: direct link, hf CLI and curl.
- Browser
- Download file 202 kB
-
https://huggingface.co/saksornr/draftmark-roberta-int8/resolve/main/DraftMark_Local_AI_Detection_Report.html
- Command line
-
hf download hf://saksornr/draftmark-roberta-int8/DraftMark_Local_AI_Detection_Report.html
-
curl -L -o DraftMark_Local_AI_Detection_Report.html https://huggingface.co/saksornr/draftmark-roberta-int8/resolve/main/DraftMark_Local_AI_Detection_Report.html
202 kB
| <html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en"> | |
| <head> | |
| <meta charset="utf-8" /> | |
| <meta name="generator" content="pandoc" /> | |
| <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=yes" /> | |
| <meta name="author" content="Prepared for Saksorn (Harry) Ruangtanusak" /> | |
| <meta name="dcterms.date" content="2026-11-09" /> | |
| <title>DraftMark: A Research Roadmap for Local AI Text Detection</title> | |
| <style> | |
| code{white-space: pre-wrap;} | |
| span.smallcaps{font-variant: small-caps;} | |
| div.columns{display: flex; gap: min(4vw, 1.5em);} | |
| div.column{flex: auto; overflow-x: auto;} | |
| div.hanging-indent{margin-left: 1.5em; text-indent: -1.5em;} | |
| ul.task-list[class]{list-style: none;} | |
| ul.task-list li input[type="checkbox"] { | |
| font-size: inherit; | |
| width: 0.8em; | |
| margin: 0 0.8em 0.2em -1.6em; | |
| vertical-align: middle; | |
| } | |
| div.csl-bib-body { } | |
| div.csl-entry { | |
| clear: both; | |
| margin-bottom: 0em; | |
| } | |
| .hanging-indent div.csl-entry { | |
| margin-left:2em; | |
| text-indent:-2em; | |
| } | |
| div.csl-left-margin { | |
| min-width:2em; | |
| float:left; | |
| } | |
| div.csl-right-inline { | |
| margin-left:2em; | |
| padding-left:1em; | |
| } | |
| div.csl-indent { | |
| margin-left: 2em; | |
| } </style> | |
| <style type="text/css">:root { color-scheme: light; --ink:#18212d; --muted:#556373; --blue:#1558a1; --line:#d9e1e9; } | |
| html { background:#f1f4f7; } | |
| body { margin:30px auto; padding:52px 60px; max-width:920px; background:white; color:var(--ink); font-family:Arial,'DejaVu Sans',sans-serif; font-size:14px; line-height:1.55; } | |
| header#title-block-header { border-top:7px solid var(--blue); border-bottom:1px solid var(--line); padding:34px 0 25px; margin-bottom:32px; } | |
| header h1.title { font-size:36px; line-height:1.12; letter-spacing:-1px; max-width:850px; color:#12263e; margin:0 0 17px; } | |
| header .subtitle { font-size:18px; color:var(--muted); margin:0 0 24px; } | |
| header .author,header .date { font-size:13px; color:var(--muted); margin:4px 0; } | |
| nav#TOC { border:1px solid var(--line); padding:20px 24px; margin:25px 0 36px; border-radius:5px; background:#f8fafc; } | |
| nav#TOC ul { list-style:none; margin:5px 0; padding-left:0; } | |
| nav#TOC ul ul { padding-left:22px; font-size:.94em; } | |
| nav#TOC li { margin:4px 0; } | |
| nav#TOC a { text-decoration:none; } | |
| h1,h2,h3 { color:#173655; line-height:1.25; break-after:avoid; } | |
| h1 { font-size:26px; margin:34px 0 16px; border-top:1px solid var(--line); padding-top:22px; } | |
| h2 { font-size:19px; margin:25px 0 12px; } | |
| h3 { font-size:15px; margin:20px 0 10px; } | |
| p { margin:9px 0 12px; orphans:3; widows:3; } | |
| a { color:var(--blue); text-decoration:none; overflow-wrap:anywhere; } | |
| strong { color:#142b44; } | |
| ul,ol { padding-left:22px; } | |
| li { padding-left:3px; margin:6px 0; } | |
| table { border-collapse:collapse; width:100%; margin:17px 0 21px; font-size:12px; line-height:1.45; table-layout:auto; } | |
| thead { display:table-header-group; } | |
| tr { break-inside:avoid; } | |
| th { background:#eaf0f7; text-align:left; font-weight:600; color:#153957; border-bottom:1.3px solid #aabbcd; padding:9px 9px; } | |
| td { padding:8px 9px; vertical-align:top; border-bottom:1px solid #e3e8ee; } | |
| tbody tr:nth-child(even) { background:#f8fafc; } | |
| td:first-child,th:first-child { text-align:left; } | |
| code { font-family:'DejaVu Sans Mono',monospace; font-size:.85em; overflow-wrap:anywhere; } | |
| pre { white-space:pre-wrap; background:#f3f6fa; padding:12px; border:1px solid var(--line); border-radius:4px; } | |
| figure { margin:22px 0; break-inside:avoid; } | |
| figure img { display:block; max-width:100%; height:auto; margin:auto; } | |
| figcaption { font-size:11px; line-height:1.4; color:var(--muted); margin:8px 0 4px; } | |
| math[display=block] { margin:20px auto; font-size:1.04em; } | |
| .references { font-size:11px; } | |
| .csl-entry { margin:11px 0; break-inside:avoid; overflow-wrap:anywhere; } | |
| .citation { color:#3f5e80; } | |
| @page { | |
| size:A4; margin:18mm 16mm 20mm; | |
| @top-left { content:'DraftMark / Local AI Text Detection'; font-family:Arial,sans-serif; font-size:8pt; color:#64748b; } | |
| @top-right { content:'Research review / 9 Nov 2026'; font-family:Arial,sans-serif; font-size:8pt; color:#64748b; } | |
| @bottom-right { content:counter(page); font-family:Arial,sans-serif; font-size:8pt; color:#64748b; } | |
| } | |
| @media print { | |
| html,body { background:white; } | |
| body { margin:0; padding:0; max-width:none; font-size:9.4pt; line-height:1.45; } | |
| header#title-block-header { padding-top:22px; } | |
| header h1.title { font-size:29pt; } | |
| header .subtitle { font-size:13pt; } | |
| header .author,header .date { font-size:9pt; } | |
| nav#TOC { font-size:9pt; break-after:page; border:none; border-radius:0; padding:0; background:white; } | |
| nav#TOC ul ul { font-size:8.5pt; } | |
| nav#TOC li { margin:2px 0; } | |
| h1 { font-size:18pt; margin-top:22px; padding-top:14px; } | |
| h2 { font-size:13pt; } | |
| h3 { font-size:10pt; } | |
| table { font-size:8.1pt; line-height:1.33; margin:12px 0 16px; } | |
| th,td { padding:6px 6px; } | |
| figcaption,.references { font-size:8pt; } | |
| a { color:#235c94; } | |
| pre { font-size:7.5pt; } | |
| body { -webkit-print-color-adjust:exact; print-color-adjust:exact; } | |
| } | |
| </style> | |
| </head> | |
| <body> | |
| <header id="title-block-header"> | |
| <h1 class="title">DraftMark: A Research Roadmap for Local AI Text | |
| Detection</h1> | |
| <p class="subtitle">Verified competitors, low-false-positive evaluation, | |
| H100 training, and device deployment</p> | |
| <p class="author">Prepared for Saksorn (Harry) Ruangtanusak</p> | |
| <p class="date">9 November 2026</p> | |
| </header> | |
| <nav id="TOC" role="doc-toc"> | |
| <ul> | |
| <li><a href="#executive-recommendation" id="toc-executive-recommendation"><span class="toc-section-number">1</span> Executive recommendation</a> | |
| <ul> | |
| <li><a href="#decisions-at-a-glance" id="toc-decisions-at-a-glance"><span class="toc-section-number">1.1</span> Decisions at a glance</a></li> | |
| </ul></li> | |
| <li><a href="#scope-and-evidence-protocol" id="toc-scope-and-evidence-protocol"><span class="toc-section-number">2</span> Scope and evidence protocol</a></li> | |
| <li><a href="#what-draftmark-already-achieves" id="toc-what-draftmark-already-achieves"><span class="toc-section-number">3</span> What DraftMark already achieves</a> | |
| <ul> | |
| <li><a href="#model-and-deployment-audit" id="toc-model-and-deployment-audit"><span class="toc-section-number">3.1</span> Model and deployment | |
| audit</a></li> | |
| <li><a href="#published-holdout-meaningful-progress-with-an-explicit-tradeoff" id="toc-published-holdout-meaningful-progress-with-an-explicit-tradeoff"><span class="toc-section-number">3.2</span> Published holdout: meaningful | |
| progress with an explicit tradeoff</a></li> | |
| </ul></li> | |
| <li><a href="#benchmarks-and-exact-scores-to-challenge" id="toc-benchmarks-and-exact-scores-to-challenge"><span class="toc-section-number">4</span> Benchmarks and exact scores to | |
| challenge</a> | |
| <ul> | |
| <li><a href="#official-raid-the-public-english-benchmark-gate" id="toc-official-raid-the-public-english-benchmark-gate"><span class="toc-section-number">4.1</span> Official RAID: the public English | |
| benchmark gate</a></li> | |
| <li><a href="#mage-generation-and-paraphrase-transfer" id="toc-mage-generation-and-paraphrase-transfer"><span class="toc-section-number">4.2</span> MAGE: generation and paraphrase | |
| transfer</a></li> | |
| <li><a href="#detectrl-x-multilingual-and-editing-sensitive-evaluation" id="toc-detectrl-x-multilingual-and-editing-sensitive-evaluation"><span class="toc-section-number">4.3</span> DetectRL-X: multilingual and | |
| editing-sensitive evaluation</a></li> | |
| <li><a href="#wattle-a-harder-test-of-web-rewriting" id="toc-wattle-a-harder-test-of-web-rewriting"><span class="toc-section-number">4.4</span> WATTLE: a harder test of web | |
| rewriting</a></li> | |
| <li><a href="#shared-tasks-and-supporting-benchmarks" id="toc-shared-tasks-and-supporting-benchmarks"><span class="toc-section-number">4.5</span> Shared tasks and supporting | |
| benchmarks</a></li> | |
| </ul></li> | |
| <li><a href="#model-strategy-small-students-relevant-larger-comparators" id="toc-model-strategy-small-students-relevant-larger-comparators"><span class="toc-section-number">5</span> Model strategy: small students, | |
| relevant larger comparators</a> | |
| <ul> | |
| <li><a href="#recommended-model-ladder" id="toc-recommended-model-ladder"><span class="toc-section-number">5.1</span> Recommended model ladder</a></li> | |
| <li><a href="#how-to-compare-larger-models-fairly" id="toc-how-to-compare-larger-models-fairly"><span class="toc-section-number">5.2</span> How to compare larger models | |
| fairly</a></li> | |
| </ul></li> | |
| <li><a href="#improvements-most-likely-to-matter" id="toc-improvements-most-likely-to-matter"><span class="toc-section-number">6</span> Improvements most likely to | |
| matter</a> | |
| <ul> | |
| <li><a href="#prioritize-data-coverage-before-architecture-novelty" id="toc-prioritize-data-coverage-before-architecture-novelty"><span class="toc-section-number">6.1</span> Prioritize data coverage before | |
| architecture novelty</a></li> | |
| <li><a href="#treat-editing-provenance-explicitly" id="toc-treat-editing-provenance-explicitly"><span class="toc-section-number">6.2</span> Treat editing provenance | |
| explicitly</a></li> | |
| <li><a href="#distillation-is-useful-engineering-not-sufficient-novelty" id="toc-distillation-is-useful-engineering-not-sufficient-novelty"><span class="toc-section-number">6.3</span> Distillation is useful | |
| engineering, not sufficient novelty</a></li> | |
| <li><a href="#research-candidate-stability-of-the-human-score-tail-after-export" id="toc-research-candidate-stability-of-the-human-score-tail-after-export"><span class="toc-section-number">6.4</span> Research candidate: stability of | |
| the human-score tail after export</a></li> | |
| </ul></li> | |
| <li><a href="#evaluation-protocol-for-a-publishable-result" id="toc-evaluation-protocol-for-a-publishable-result"><span class="toc-section-number">7</span> Evaluation protocol for a | |
| publishable result</a> | |
| <ul> | |
| <li><a href="#separate-four-roles-for-data" id="toc-separate-four-roles-for-data"><span class="toc-section-number">7.1</span> Separate four roles for | |
| data</a></li> | |
| <li><a href="#primary-metrics" id="toc-primary-metrics"><span class="toc-section-number">7.2</span> Primary metrics</a></li> | |
| <li><a href="#calibration-and-statistical-uncertainty" id="toc-calibration-and-statistical-uncertainty"><span class="toc-section-number">7.3</span> Calibration and statistical | |
| uncertainty</a></li> | |
| <li><a href="#concrete-success-criteria-defined-before-the-next-test" id="toc-concrete-success-criteria-defined-before-the-next-test"><span class="toc-section-number">7.4</span> Concrete success criteria, defined | |
| before the next test</a></li> | |
| </ul></li> | |
| <li><a href="#h100-training-plan" id="toc-h100-training-plan"><span class="toc-section-number">8</span> H100 training plan</a> | |
| <ul> | |
| <li><a href="#start-on-one-gpu" id="toc-start-on-one-gpu"><span class="toc-section-number">8.1</span> Start on one GPU</a></li> | |
| <li><a href="#run-sequence" id="toc-run-sequence"><span class="toc-section-number">8.2</span> Run sequence</a></li> | |
| <li><a href="#warm-start-reconstruction" id="toc-warm-start-reconstruction"><span class="toc-section-number">8.3</span> Warm-start reconstruction</a></li> | |
| <li><a href="#estimate-time-from-a-pilot" id="toc-estimate-time-from-a-pilot"><span class="toc-section-number">8.4</span> Estimate time from a | |
| pilot</a></li> | |
| </ul></li> | |
| <li><a href="#local-app-development" id="toc-local-app-development"><span class="toc-section-number">9</span> | |
| Local app development</a> | |
| <ul> | |
| <li><a href="#build-on-the-existing-browser-product" id="toc-build-on-the-existing-browser-product"><span class="toc-section-number">9.1</span> Build on the existing browser | |
| product</a></li> | |
| <li><a href="#app-rollout-sequence" id="toc-app-rollout-sequence"><span class="toc-section-number">9.2</span> App rollout sequence</a></li> | |
| </ul></li> | |
| <li><a href="#a-focused-thai-english-research-extension" id="toc-a-focused-thai-english-research-extension"><span class="toc-section-number">10</span> A focused Thai-English research | |
| extension</a></li> | |
| <li><a href="#report-and-publication-strategy" id="toc-report-and-publication-strategy"><span class="toc-section-number">11</span> Report and publication strategy</a> | |
| <ul> | |
| <li><a href="#what-can-be-published-now" id="toc-what-can-be-published-now"><span class="toc-section-number">11.1</span> What can be published | |
| now</a></li> | |
| <li><a href="#what-would-justify-a-stronger-paper" id="toc-what-would-justify-a-stronger-paper"><span class="toc-section-number">11.2</span> What would justify a stronger | |
| paper</a></li> | |
| <li><a href="#suggested-paper-outline" id="toc-suggested-paper-outline"><span class="toc-section-number">11.3</span> Suggested paper outline</a></li> | |
| </ul></li> | |
| <li><a href="#prioritized-execution-plan" id="toc-prioritized-execution-plan"><span class="toc-section-number">12</span> Prioritized execution plan</a></li> | |
| <li><a href="#references" id="toc-references"><span class="toc-section-number">13</span> References</a></li> | |
| </ul> | |
| </nav> | |
| <h1 data-number="1" id="executive-recommendation"><span class="header-section-number">1</span> Executive recommendation</h1> | |
| <p><strong>Keep DraftMarkβs existing RoBERTa release as the control. | |
| Build the next model around generalization to unseen rewrites, careful | |
| treatment of AI-assisted human writing, and low false-positive rates | |
| measured on the exact local artifact.</strong> A useful first comparison | |
| is a 68M/150M encoder student against the existing 125M model and an | |
| Ettin-1B teacher. Begin with English; add a dedicated Thai-English study | |
| as a second research track, using a multilingual backbone and locally | |
| collected evidence.</p> | |
| <p>The most relevant compact competitor found in this review is GPTTrace | |
| Ettin-150M v5e. Its official RAID submission reaches <strong>98.91% TPR | |
| at 1% FPR</strong> and <strong>99.80% TPR at 5% FPR</strong>, including | |
| attacks. The updated Ettin-1B v8s submission reaches | |
| <strong>99.54%</strong> and <strong>99.93%</strong>, respectively. A | |
| browser implementation, INT8 weights, and teacher distillation already | |
| exist for the 150M competitor. Those features alone do not constitute a | |
| new research contribution. <span class="citation" data-cites="raid150 raid1b gpttrace150">(RAID evaluation maintainers | |
| 2026a, 2026b; Nakas 2026a)</span></p> | |
| <p>The more promising question is whether a small exported detector can | |
| preserve useful recall on unfamiliar rewriting processes while keeping | |
| human false flags low across populations and devices. This is both | |
| practically meaningful and less saturated than an aggregate RAID score. | |
| This report defines the comparisons and provides a training/evaluation | |
| starter package; <strong>it does not report a new H100 training run or | |
| new detector benchmark measurements</strong>.</p> | |
| <h2 data-number="1.1" id="decisions-at-a-glance"><span class="header-section-number">1.1</span> Decisions at a glance</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 50%" /> | |
| <col style="width: 50%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Decision</th> | |
| <th style="text-align: left;">Recommendation</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Immediate model</td> | |
| <td style="text-align: left;">Continue the pinned DraftMark v3 | |
| checkpoint on one H100; compare last-four-layer training with full | |
| fine-tuning.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Main compact challenger</td> | |
| <td style="text-align: left;">Ettin-150M; add Ettin-68M to test the | |
| quality-versus-size frontier.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">English teacher</td> | |
| <td style="text-align: left;">Ettin-1B, including a frozen GPTTrace v8s | |
| comparator; teacher quality must be measured on the target data.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Multilingual starting point</td> | |
| <td style="text-align: left;">mDeBERTa-v3-base or XLM-R-base; do not | |
| assume the English checkpoint supports Thai.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Main product metric</td> | |
| <td style="text-align: left;">Recall at a threshold fixed on separate | |
| calibration data, accompanied by observed test FPR and subgroup | |
| confidence intervals.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Public benchmark gate</td> | |
| <td style="text-align: left;">Official RAID protocol, plus MAGE | |
| OOD/paraphrase, DetectRL-X, and WATTLE.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Credible contribution</td> | |
| <td style="text-align: left;">Robust transfer and calibration after | |
| quantization under a specified memory/latency budget.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Initial device budget</td> | |
| <td style="text-align: left;">Preserve the existing <160 MB | |
| model-plus-tokenizer gate; separately measure peak RAM, latency, cold | |
| start, and power.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">First app step</td> | |
| <td style="text-align: left;">Extend the existing local browser app with | |
| an auditable offline package and device benchmark suite.</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <h1 data-number="2" id="scope-and-evidence-protocol"><span class="header-section-number">2</span> Scope and evidence protocol</h1> | |
| <p>This is a targeted critical review, not a systematic claim of | |
| exhaustive coverage. The search cutoff is <strong>9 November 2026, | |
| Asia/Bangkok</strong>. The seed is | |
| <code>saksornr/draftmark-roberta-int8</code>. The review followed its | |
| base model, training notes, benchmark dependencies, and nearby detector | |
| literature, then checked recent official benchmark submissions and | |
| primary papers.</p> | |
| <p>Search lanes included the seed/model provenance; references from | |
| RAID, MAGE, DetectRL-X, MELD and WATTLE; later leaderboard updates; | |
| multilingual and mixed-authorship benchmarks; larger | |
| likelihood/representation detectors; and official runtime documentation. | |
| Searches included combinations of AI-generated text detection, local, | |
| browser, INT8, paraphrase, low false-positive rate, Thai, multilingual, | |
| calibration, and H100. Author cards were treated as author reports, not | |
| independent replications.</p> | |
| <p>Four evidence classes are used throughout:</p> | |
| <ol type="1"> | |
| <li><strong>Release evidence:</strong> DraftMarkβs public manifests and | |
| aggregate report. These were inspected, not independently rerun.</li> | |
| <li><strong>Official benchmark evidence:</strong> RAID CI results at a | |
| pinned repository revision and organizer leaderboards.</li> | |
| <li><strong>Paper or author-harness evidence:</strong> exact reported | |
| datasets, splits and metrics, with their limitations retained.</li> | |
| <li><strong>Proposed experiment or target:</strong> an | |
| engineering/research decision in this report, not an achieved | |
| result.</li> | |
| </ol> | |
| <p>The DraftMark main revision inspected was | |
| <code>068ccc7ba34f134526c9bbc72ac2033f00a40d45</code>. The RAID snapshot | |
| was <code>ac4ce10fe8fe096b0c4a01396f1e437aaf35bd53</code>, dated 5 | |
| November 2026. The companion evidence files preserve exact scores, | |
| source URLs and versions. Public artifacts do not include DraftMarkβs | |
| complete trainer, row-level evaluation predictions or full split | |
| manifests, so an independent reproduction of its original run is not yet | |
| possible from the release alone. <span class="citation" data-cites="draftmark draftmarkmetrics draftmarkhandoff raidrepo">(DraftMark | |
| project 2026b, 2026c, 2026a; Dugan et al. 2026)</span></p> | |
| <h1 data-number="3" id="what-draftmark-already-achieves"><span class="header-section-number">3</span> What DraftMark already | |
| achieves</h1> | |
| <h2 data-number="3.1" id="model-and-deployment-audit"><span class="header-section-number">3.1</span> Model and deployment audit</h2> | |
| <p>DraftMark v3 is a fine-tuned <strong>RoBERTa-base detector, | |
| approximately 125M total parameters</strong>, initialized from Oxidaneβs | |
| TMR checkpoint. It updates the last four encoder layers plus the | |
| classifier: <strong>28,943,618 trainable parameters</strong>. The | |
| published training run used Apple MPS, 6,339 training drafts, 761 | |
| calibration/development drafts and 1,120 holdout drafts. The deployed | |
| model and tokenizer occupy <strong>131,110,132 bytes</strong>, | |
| approximately 131.1 MB. <span class="citation" data-cites="draftmark draftmarkmetrics tmr">(DraftMark project 2026b, | |
| 2026c; Oxidane 2025)</span></p> | |
| <p>The release is English-only. It accepts 150-2,000 words, with a | |
| 20,000-character cap. The worker scores sentence-aligned windows of up | |
| to 512 tokens and aggregates AI scores by content-token count. The label | |
| order is human, then AI. Its production flag threshold is approximately | |
| <strong>0.9991466</strong>, with a lower limited-signal cutoff of 0.25. | |
| These are product policies fitted for this artifact; 0.9991466 is not a | |
| calibrated 99.9% probability of authorship. <span class="citation" data-cites="draftmark">(DraftMark project 2026b)</span></p> | |
| <h2 data-number="3.2" id="published-holdout-meaningful-progress-with-an-explicit-tradeoff"><span class="header-section-number">3.2</span> Published holdout: meaningful | |
| progress with an explicit tradeoff</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Cohort</th> | |
| <th style="text-align: right;">Previous DraftMark Ettin policy</th> | |
| <th style="text-align: right;">Current DraftMark v3</th> | |
| <th style="text-align: right;">Change</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Clean AI recall</td> | |
| <td style="text-align: right;">215/300 = 71.67%</td> | |
| <td style="text-align: right;">277/300 = 92.33%</td> | |
| <td style="text-align: right;">+20.67 points</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Decoder-rewritten AI recall</td> | |
| <td style="text-align: right;">18/150 = 12.00%</td> | |
| <td style="text-align: right;">89/150 = 59.33%</td> | |
| <td style="text-align: right;">+47.33 points</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Seq2seq-rewritten AI recall</td> | |
| <td style="text-align: right;">132/150 = 88.00%</td> | |
| <td style="text-align: right;">110/150 = 73.33%</td> | |
| <td style="text-align: right;">-14.67 points</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Mean of the two rewrite recalls</td> | |
| <td style="text-align: right;">50.00%</td> | |
| <td style="text-align: right;">66.33%</td> | |
| <td style="text-align: right;">+16.33 points</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">General-human false flags</td> | |
| <td style="text-align: right;">10/300 = 3.33%</td> | |
| <td style="text-align: right;">1/300 = 0.33%</td> | |
| <td style="text-align: right;">Lower is better</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Learner-essay false flags</td> | |
| <td style="text-align: right;">80/220 = 36.36%</td> | |
| <td style="text-align: right;">0/220 = 0.00%</td> | |
| <td style="text-align: right;">Lower is better</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>These are the releaseβs own matched holdout results at each shipped | |
| policy, not a comparison with todayβs best public Ettin checkpoint on an | |
| official leaderboard. Its paired 95% interval for mean rewrite | |
| improvement is +10.7 to +22.0 percentage points. The improvement is | |
| substantial on this sample, but the seq2seq regression is equally real. | |
| <span class="citation" data-cites="draftmarkmetrics">(DraftMark project | |
| 2026c)</span></p> | |
| <figure> | |
| <img src="data:image/png;base64,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" style="width:95.0%" alt="Published DraftMark holdout results at shipped thresholds. Error bars show the releaseβs Wilson 95% intervals; the two models are evaluated on the same listed cohorts. These are historical release measurements, not new runs." /> | |
| <figcaption aria-hidden="true">Published DraftMark holdout results at | |
| shipped thresholds. Error bars show the releaseβs Wilson 95% intervals; | |
| the two models are evaluated on the same listed cohorts. These are | |
| historical release measurements, not new runs.</figcaption> | |
| </figure> | |
| <h3 data-number="3.2.1" id="what-the-aggregate-numbers-do-not-establish"><span class="header-section-number">3.2.1</span> What the aggregate numbers do | |
| not establish</h3> | |
| <p>First, zero learner false flags on 220 essays is not proof of a zero | |
| or sub-1% population error rate. The releaseβs two-sided Wilson upper | |
| bound is <strong>1.72%</strong>. For 1/300 general-human false flags, | |
| the interval is approximately <strong>0.059%-1.864%</strong>. A stronger | |
| low-FPR claim needs more independent human source groups.</p> | |
| <p>Second, the release separately reports decoder AUROC 98.51% and | |
| seq2seq AUROC 98.27%, with diagnostic ROC recall around 74.50% and | |
| 76.51% at 1% FPR. These diagnostic values exclude out-of-input-limit | |
| cases and use ROC operating points. They must not replace the actual | |
| shipped recalls of 59.33% and 73.33%, whose denominators include the | |
| full cohorts. High ranking quality and a conservative production | |
| threshold can coexist.</p> | |
| <p>Third, the latency result is <strong>1.262 seconds p95</strong> for a | |
| warm 300-word scan on an Apple M4 Pro in Chromium, based on 20 samples. | |
| It does not establish mobile latency, peak RAM, energy, or a complete | |
| offline cold-start experience. <span class="citation" data-cites="draftmarkmetrics">(DraftMark project 2026c)</span></p> | |
| <p>The base TMR model was trained on RAID. DraftMark additionally used | |
| DetectRL-X-derived material and other sources. Its release documents | |
| overlap screening, including exclusions against public training sources, | |
| but acknowledges that unpublished upstream training data cannot be | |
| checked. Treat RAID as an important benchmark-distribution test and use | |
| fresh generators, rewriting methods and sources for genuine transfer | |
| claims. <span class="citation" data-cites="tmr draftmarkmetrics">(Oxidane 2025; DraftMark project | |
| 2026c)</span></p> | |
| <h1 data-number="4" id="benchmarks-and-exact-scores-to-challenge"><span class="header-section-number">4</span> Benchmarks and exact scores to | |
| challenge</h1> | |
| <h2 data-number="4.1" id="official-raid-the-public-english-benchmark-gate"><span class="header-section-number">4.1</span> Official RAID: the public | |
| English benchmark gate</h2> | |
| <p>The following rows are from the same pinned official RAID snapshot, | |
| all settings including attacks, covering 672,000 test IDs. Scores are | |
| percentages. These are selected relevant competitors, not a claim that | |
| every model is publicly runnable at the same precision. <span class="citation" data-cites="raidrepo raid150 raid1b meld">(Dugan et al. | |
| 2026; RAID evaluation maintainers 2026a, 2026b; C. Li, Wan, and Paetzold | |
| 2026)</span></p> | |
| <table> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Detector</th> | |
| <th style="text-align: right;">TPR at 1% FPR</th> | |
| <th style="text-align: right;">TPR at 5% FPR</th> | |
| <th style="text-align: right;">AUROC</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Winston AI, Sep 2026 submission</td> | |
| <td style="text-align: right;">99.89</td> | |
| <td style="text-align: right;">99.98</td> | |
| <td style="text-align: right;">99.89</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">GPTTrace Ettin-1B v8s, Nov 2026</td> | |
| <td style="text-align: right;"><strong>99.54</strong></td> | |
| <td style="text-align: right;"><strong>99.93</strong></td> | |
| <td style="text-align: right;"><strong>99.94</strong></td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">XYZ SCIENCE, Oct 31 submission</td> | |
| <td style="text-align: right;">99.35</td> | |
| <td style="text-align: right;">99.91</td> | |
| <td style="text-align: right;">99.94</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">MELD, Apr 2026 submission</td> | |
| <td style="text-align: right;">99.24</td> | |
| <td style="text-align: right;">99.78</td> | |
| <td style="text-align: right;">99.82</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">GPTTrace Ettin-150M v5e</td> | |
| <td style="text-align: right;"><strong>98.91</strong></td> | |
| <td style="text-align: right;"><strong>99.80</strong></td> | |
| <td style="text-align: right;"><strong>99.89</strong></td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">GPTZero, Oct 2025 submission</td> | |
| <td style="text-align: right;">92.93</td> | |
| <td style="text-align: right;">97.24</td> | |
| <td style="text-align: right;">99.25</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">TMR RoBERTa, DraftMarkβs ancestor</td> | |
| <td style="text-align: right;">90.17</td> | |
| <td style="text-align: right;">95.79</td> | |
| <td style="text-align: right;">99.28</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">ModernBERT AI Detection</td> | |
| <td style="text-align: right;">88.23</td> | |
| <td style="text-align: right;">94.14</td> | |
| <td style="text-align: right;">97.65</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">e5-small-LoRA</td> | |
| <td style="text-align: right;">73.08</td> | |
| <td style="text-align: right;">85.69</td> | |
| <td style="text-align: right;">96.85</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">BERT-tiny-4M</td> | |
| <td style="text-align: right;">71.90</td> | |
| <td style="text-align: right;">84.18</td> | |
| <td style="text-align: right;">96.71</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p><strong>Practical interpretation:</strong> a comparable compact | |
| detector must challenge 98.91% TPR at 1% FPR, while 99.54% is the | |
| relevant verified open 1B reference. A universal leaderboard claim must | |
| also address the strongest closed submissions. Beating legacy OpenAI | |
| GPT-2 RoBERTa detectors would not establish modern state of the art.</p> | |
| <p>The newer v8s result supersedes older pages quoting 99.42% TPR at 1% | |
| FPR for the 1B model. Avoid copying stale ranks: an October 31 | |
| submission also changed the ordering after GPTTraceβs October marketing | |
| table. The exact unrounded metrics are preserved in the evidence CSV. | |
| <span class="citation" data-cites="raid1b raidrepo">(RAID evaluation | |
| maintainers 2026b; Dugan et al. 2026)</span></p> | |
| <h3 data-number="4.1.1" id="a-critical-threshold-distinction"><span class="header-section-number">4.1.1</span> A critical threshold | |
| distinction</h3> | |
| <p>The official evaluator defaults to <strong>per-domain | |
| tuning</strong>. It chooses each domainβs threshold using unattacked | |
| human <strong>test</strong> scores, then applies that threshold to AI | |
| text and pools the resulting counts. This is the official diagnostic | |
| protocol; it is not evidence that one fixed threshold calibrated earlier | |
| will maintain 1% FPR for unknown app users. The paper should report both | |
| official RAID numbers and a second deployment track with independent | |
| calibration. <span class="citation" data-cites="raidrepo">(Dugan et al. | |
| 2026)</span></p> | |
| <p>Also, the 150M competitorβs card reports a small-sample INT8 versus | |
| PyTorch discrepancy: AUROC 0.975 versus 0.990 on 300 MAGE texts. Its | |
| official RAID number should not automatically be attributed to the exact | |
| deployed INT8 browser artifact. Re-evaluate that artifact under the same | |
| runtime and input policy as DraftMark. <span class="citation" data-cites="gpttrace150">(Nakas 2026a)</span></p> | |
| <h2 data-number="4.2" id="mage-generation-and-paraphrase-transfer"><span class="header-section-number">4.2</span> MAGE: generation and paraphrase | |
| transfer</h2> | |
| <p>MAGE supplies explicit OOD generator and paraphrase testbeds. Useful | |
| author-reported reference points are below. They are not a centralized | |
| current SOTA ranking; reproduce the exact testbed and scoring pipeline | |
| before asserting a win. <span class="citation" data-cites="mage gpttrace150 gpttrace1b">(Y. Li et al. 2024; Nakas | |
| 2026a, 2026b)</span></p> | |
| <table> | |
| <colgroup> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Testbed</th> | |
| <th style="text-align: right;">Compact GPTTrace 150M</th> | |
| <th style="text-align: right;">Larger reference</th> | |
| <th style="text-align: left;">Proposed interpretation</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Unseen GPT-4 generation, AUROC</td> | |
| <td style="text-align: right;">0.976</td> | |
| <td style="text-align: right;">0.990, earlier 1B teacher</td> | |
| <td style="text-align: left;">A compact generalization gate</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">OOD paraphrase, AUROC</td> | |
| <td style="text-align: right;"><strong>0.756</strong></td> | |
| <td style="text-align: right;"><strong>0.919</strong>, 1B v8s author | |
| run</td> | |
| <td style="text-align: left;">A more meaningful head-to-head target</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>DivEye separately reports 0.87 paraphrase AUROC, and its combination | |
| with BiScope reports 0.89. These establish that 0.756 is a useful direct | |
| compact-competitor baseline, not the best score in the literature. <span class="citation" data-cites="diveye">(Basani and Chen 2026)</span></p> | |
| <p><strong>Label policy matters.</strong> MAGEβs original labels use 0 | |
| for machine and 1 for human. Its paraphrase testbed includes | |
| machine-rewritten human text under the machine-generated category. | |
| Preserve that definition for official comparability, then separately | |
| measure human-origin writing with light AI assistance. A product that | |
| intends to protect human-origin writing is answering a different | |
| question from detecting any machine intervention. <span class="citation" data-cites="mage">(Y. Li et al. 2024)</span></p> | |
| <h2 data-number="4.3" id="detectrl-x-multilingual-and-editing-sensitive-evaluation"><span class="header-section-number">4.3</span> DetectRL-X: multilingual and | |
| editing-sensitive evaluation</h2> | |
| <p>DetectRL-X evaluates binary and ternary detection over eight transfer | |
| dimensions. Its languages do not include Thai. The table below | |
| reproduces its Table 2 averages. The column labeled <strong>macro-F1 at | |
| 1% FPR reports F1, not true-positive rate</strong>. <span class="citation" data-cites="detectrlx">(Wu et al. 2026)</span></p> | |
| <table> | |
| <colgroup> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Detector</th> | |
| <th style="text-align: right;">Binary best macro-F1</th> | |
| <th style="text-align: right;">Binary macro-F1 at 1% FPR</th> | |
| <th style="text-align: right;">Ternary best macro-F1</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">XLM-R-base classifier</td> | |
| <td style="text-align: right;"><strong>95.58</strong></td> | |
| <td style="text-align: right;">91.31</td> | |
| <td style="text-align: right;">82.78</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">mDeBERTa-v3-base classifier</td> | |
| <td style="text-align: right;">95.48</td> | |
| <td style="text-align: right;"><strong>93.20</strong></td> | |
| <td style="text-align: right;"><strong>87.68</strong></td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">BiScope, Qwen2.5-7B-based</td> | |
| <td style="text-align: right;">80.06</td> | |
| <td style="text-align: right;">63.62</td> | |
| <td style="text-align: right;">59.69</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Binoculars</td> | |
| <td style="text-align: right;">68.17</td> | |
| <td style="text-align: right;">54.18</td> | |
| <td style="text-align: right;">41.62</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Fast-DetectGPT</td> | |
| <td style="text-align: right;">48.27</td> | |
| <td style="text-align: right;">40.50</td> | |
| <td style="text-align: right;">28.12</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>The cross-paraphrase binary slice is much harder: XLM-R reaches 77.32 | |
| best macro-F1 and 64.13 macro-F1 at 1% FPR. For the ternary task, | |
| mDeBERTaβs macro-F1 at 1% FPR is 81.10. These are useful challenge | |
| scores. Larger decoder-based systems do not automatically win, but the | |
| encoder systems receive supervised training: this is not a controlled | |
| estimate of model-size effects. <span class="citation" data-cites="detectrlx">(Wu et al. 2026)</span></p> | |
| <p>Because DraftMark already draws from DetectRL-X, evaluate only | |
| documented held-out groups for regression and reserve genuinely unseen | |
| generator/rewriter/source combinations for transfer. Do not call the | |
| complete dataset unseen merely because a particular test file was not | |
| used.</p> | |
| <h2 data-number="4.4" id="wattle-a-harder-test-of-web-rewriting"><span class="header-section-number">4.4</span> WATTLE: a harder test of web | |
| rewriting</h2> | |
| <p>WATTLE is a recent source-matched web benchmark. Its results make the | |
| research gap clear: good conventional benchmark scores coexist with poor | |
| detection at fixed low-FPR operating points. Table 4 reports the | |
| following spin/rewrite results; each method uses its own calibrated | |
| operating point. <span class="citation" data-cites="wattle">(Chernikov, | |
| Ptitsyn, and Burlova 2026)</span></p> | |
| <table> | |
| <colgroup> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Method</th> | |
| <th style="text-align: right;">Rewrite AUROC</th> | |
| <th style="text-align: right;">Rewrite recall at calibrated 1% FPR</th> | |
| <th style="text-align: right;">Frontier rewrite AUROC</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">RoBERTa trained on MAGE</td> | |
| <td style="text-align: right;">73.5</td> | |
| <td style="text-align: right;">6.9%</td> | |
| <td style="text-align: right;">67.3</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">RADAR</td> | |
| <td style="text-align: right;"><strong>80.3</strong></td> | |
| <td style="text-align: right;">1.5%</td> | |
| <td style="text-align: right;"><strong>75.4</strong></td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Binoculars</td> | |
| <td style="text-align: right;">67.9</td> | |
| <td style="text-align: right;">11.7%</td> | |
| <td style="text-align: right;">38.2</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">PAWN++, two frozen 1B probes</td> | |
| <td style="text-align: right;">70.1</td> | |
| <td style="text-align: right;">8.6%</td> | |
| <td style="text-align: right;">58.1</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>A useful experiment should challenge the strongest relevant baseline | |
| on each metric, not only the proposed method in the paper. On this | |
| comparison, surpassing PAWN++βs 58.1 frontier AUROC is insufficient | |
| because RADAR already scores 75.4. Re-run current GPTTrace and | |
| MELD/Vanguard comparators here before describing a new result as the | |
| best available. The paperβs clean RAID-train results are a separate | |
| protocol and are deliberately omitted from the official RAID table. | |
| <span class="citation" data-cites="wattle">(Chernikov, Ptitsyn, and | |
| Burlova 2026)</span></p> | |
| <h2 data-number="4.5" id="shared-tasks-and-supporting-benchmarks"><span class="header-section-number">4.5</span> Shared tasks and supporting | |
| benchmarks</h2> | |
| <p>PAN-CLEF 2026 provides an adversarial software-submission setting. | |
| The winning systemβs composite score is 0.975, with F1 0.976 and FPR | |
| 5.3%; the second-place ModernBERT system has AUROC 0.993 but FPR 10.8% | |
| at its submitted decision rule. Its ranking is a composite, not a | |
| low-FPR guarantee. The taskβs data page restricts redistribution and | |
| commercial reuse; use it within its stated research terms. <span class="citation" data-cites="pan2026">(PAN organizers 2026)</span></p> | |
| <p>COLING 2025βs GenAI detection shared task offers additional transfer | |
| checks: 83.07 English test macro-F1 and 79.16 multilingual test macro-F1 | |
| are organizer-reported winning scores. These are distinct from | |
| development accuracy. MULTITuDE is useful for broader cross-language | |
| diagnosis but does not include Thai. Legacy HC3, Ghostbuster and | |
| original DetectRL are supporting tests, not sufficient evidence of | |
| current frontier robustness. <span class="citation" data-cites="coling2025 multitude">(Wang et al. 2025; Macko et al. | |
| 2023)</span></p> | |
| <h1 data-number="5" id="model-strategy-small-students-relevant-larger-comparators"><span class="header-section-number">5</span> Model strategy: small students, | |
| relevant larger comparators</h1> | |
| <h2 data-number="5.1" id="recommended-model-ladder"><span class="header-section-number">5.1</span> Recommended model ladder</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| <col style="width: 25%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Model or family</th> | |
| <th style="text-align: left;">Role</th> | |
| <th style="text-align: left;">Why include it</th> | |
| <th style="text-align: left;">Main limitation</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">DraftMark v3, about 125M</td> | |
| <td style="text-align: left;">Frozen product control</td> | |
| <td style="text-align: left;">Actual existing product, published | |
| baseline and device policy</td> | |
| <td style="text-align: left;">English; rewrite-family imbalance</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Same RoBERTa, all layers trainable</td> | |
| <td style="text-align: left;">First controlled candidate</td> | |
| <td style="text-align: left;">Isolates adaptation capacity without | |
| changing tokenizer/export architecture</td> | |
| <td style="text-align: left;">More training freedom can overfit</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Ettin-68M</td> | |
| <td style="text-align: left;">Small local student</td> | |
| <td style="text-align: left;">Tests whether a substantially smaller | |
| encoder retains useful low-FPR recall</td> | |
| <td style="text-align: left;">Quality and export must be measured</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Ettin-150M</td> | |
| <td style="text-align: left;">Main local student</td> | |
| <td style="text-align: left;">Direct comparison with a strong | |
| browser-sized competitor</td> | |
| <td style="text-align: left;">Existing distillation prior art</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Ettin-1B / GPTTrace v8s</td> | |
| <td style="text-align: left;">English teacher and larger comparator</td> | |
| <td style="text-align: left;">Strong verified official score and more | |
| recent rewrite training</td> | |
| <td style="text-align: left;">Around 1B weights; unsuitable as the | |
| default low-memory tier</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">MELD, version pinned</td> | |
| <td style="text-align: left;">Robust training baseline</td> | |
| <td style="text-align: left;">Multi-task, consistency and ranking | |
| recipe</td> | |
| <td style="text-align: left;">Paperβs 396M version and current | |
| repository differ</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Vanguard ModernBERT-large</td> | |
| <td style="text-align: left;">Additional encoder comparator</td> | |
| <td style="text-align: left;">Recent OOD author evaluation and | |
| shared-task lineage</td> | |
| <td style="text-align: left;">Author-harness results need | |
| replication</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">mDeBERTa-v3-base / XLM-R-base</td> | |
| <td style="text-align: left;">Multilingual teacher and baseline</td> | |
| <td style="text-align: left;">Strong matched DetectRL-X results; | |
| suitable multilingual starting point</td> | |
| <td style="text-align: left;">Large vocabulary affects total size</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Binoculars / Fast-DetectGPT</td> | |
| <td style="text-align: left;">Larger statistical baselines</td> | |
| <td style="text-align: left;">Different inductive bias and training-data | |
| requirements</td> | |
| <td style="text-align: left;">Multiple large forward passes and | |
| distribution sensitivity</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>Ettin exposes 17M, 32M, 68M, 150M, 400M and 1B encoder scales. Begin | |
| with 68M and 150M rather than running the entire ladder. Microsoftβs | |
| mDeBERTa card counts 86M backbone plus about 190M embedding parameters, | |
| approximately <strong>276M total</strong>, not 86M. Its multilingual | |
| pretraining and Thai evaluation make it a reasonable Thai starting | |
| point, but they do not establish Thai AI-detection accuracy. <span class="citation" data-cites="ettin mdeberta">(JHU CLSP 2025; Microsoft | |
| 2021)</span></p> | |
| <p>MELDβs paper describes a 396M trainable Ettin-400M configuration. Its | |
| current model repository differs, so compare a pinned paper-era artifact | |
| or explicitly label a fresh reproduction. Vanguard reports 0.9475 mean | |
| OOD AUROC across seven datasets, but this author-harness mean cannot be | |
| ranked against RAID or DetectRL-X aggregates. <span class="citation" data-cites="meld vanguard">(C. Li, Wan, and Paetzold 2026; Thorat | |
| 2026)</span></p> | |
| <h2 data-number="5.2" id="how-to-compare-larger-models-fairly"><span class="header-section-number">5.2</span> How to compare larger models | |
| fairly</h2> | |
| <p>Use two experiments. In the <strong>released-system | |
| comparison</strong>, run each published detector using its documented | |
| preprocessing, then fit operating thresholds on the same permitted | |
| calibration population. In the <strong>controlled architecture | |
| comparison</strong>, train the selected student and larger encoder on | |
| the same groups, labels, sampling distribution and tuning budget. Report | |
| both; they answer different questions.</p> | |
| <p>For a decoder comparator, specify the exact observer/scoring pair, | |
| dtype, sequence limit and feature extraction. Binoculars uses paired | |
| model likelihoods; Fast-DetectGPTβs maintained repository recommends | |
| newer model pairs than its original experiment. Count all resident | |
| models and inference passes, not only the learned classification head. A | |
| 2.7M head atop two frozen 1B probes is still roughly a 2B inference | |
| system. <span class="citation" data-cites="binoculars fastdetect wattle">(Hans et al. 2024; Bao et al. | |
| 2026; Chernikov, Ptitsyn, and Burlova 2026)</span></p> | |
| <p>Do not compare H100 latency for a large model with phone CPU latency | |
| for the student as though the hardware were matched. Report server | |
| throughput separately from device batch-one latency. For each target | |
| device, list either measured performance or the concrete reason a model | |
| cannot run under the selected memory/runtime budget.</p> | |
| <h1 data-number="6" id="improvements-most-likely-to-matter"><span class="header-section-number">6</span> Improvements most likely to | |
| matter</h1> | |
| <h2 data-number="6.1" id="prioritize-data-coverage-before-architecture-novelty"><span class="header-section-number">6.1</span> Prioritize data coverage before | |
| architecture novelty</h2> | |
| <p>The present failure pattern suggests a targeted first intervention: | |
| expand decoder-style rewriting while retaining enough seq2seq examples | |
| and difficult human negatives to avoid another recall tradeoff. Start | |
| with a controlled 50k-100k-example experiment, then test 200k-300k only | |
| if the held-out transfer curve justifies it. Those counts are proposed | |
| experiment sizes, not requirements or demonstrated optima.</p> | |
| <p>Create matched groups containing the original human source, | |
| independent AI drafts on the same topic, multiple rewrite styles and any | |
| subsequent edits. Hold out entire source families, generator families | |
| and rewriting methods in separate test tracks. Match length, topic and | |
| formatting across classes so the classifier cannot solve the task by | |
| recognizing different corpora.</p> | |
| <p>Human negatives should include learner prose, formal technical | |
| writing, templated business text, short but valid passages, translated | |
| writing and repeated-domain vocabulary. Recruit or license a distinct | |
| learner-writing corpus for final evaluation. Reusing ASAP-derived | |
| patterns in both development and the only learner test gives a narrow | |
| estimate of real-world fairness.</p> | |
| <p>Keep the current v3 holdout as a regression suite. Once its results | |
| guide multiple decisions, it is no longer an untouched final test. | |
| Repeatedly changing thresholds or examples based on that holdout would | |
| make later headline gains less convincing.</p> | |
| <h2 data-number="6.2" id="treat-editing-provenance-explicitly"><span class="header-section-number">6.2</span> Treat editing provenance | |
| explicitly</h2> | |
| <p>Use four documented provenance groups: human originals; human-origin | |
| drafts lightly edited by AI; AI-origin drafts rewritten or human-edited; | |
| and unedited AI drafts. Where available, retain edit operations and | |
| aligned spans. Derive benchmark labels from each benchmarkβs stated | |
| definition rather than forcing one universal label mapping.</p> | |
| <p>For product evaluation, separate <strong>origin</strong> from | |
| <strong>assistance</strong>. A detector may recognize machine polishing | |
| without establishing that the underlying ideas or draft came from AI. A | |
| third assisted/uncertain outcome can be useful, but only when trained | |
| and evaluated with those labels. Document-level supervision alone does | |
| not justify presenting sentence highlights as verified machine-authored | |
| spans.</p> | |
| <p>This construct distinction is central to the proposed publication. It | |
| prevents a model from appearing more robust simply because it flags all | |
| polished human writing as AI.</p> | |
| <h2 data-number="6.3" id="distillation-is-useful-engineering-not-sufficient-novelty"><span class="header-section-number">6.3</span> Distillation is useful | |
| engineering, not sufficient novelty</h2> | |
| <p>After measuring the English teacher on training/development groups, | |
| train a compact student with hard labels and cached teacher logits. A | |
| reasonable starting objective is:</p> | |
| <p><math display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>β</mi><mrow><mi mathvariant="normal">b</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">s</mi><mi mathvariant="normal">e</mi></mrow></msub><mo>=</mo><mrow><mo stretchy="true" form="prefix">(</mo><mn>1</mn><mo>β</mo><mi>Ξ±</mi><mo stretchy="true" form="postfix">)</mo></mrow><mspace width="0.167em"></mspace><mrow><mi mathvariant="normal">C</mi><mi mathvariant="normal">E</mi></mrow><mrow><mo stretchy="true" form="prefix">(</mo><mi>y</mi><mo>,</mo><msub><mi>p</mi><mi>s</mi></msub><mo stretchy="true" form="postfix">)</mo></mrow><mo>+</mo><mi>Ξ±</mi><msup><mi>T</mi><mn>2</mn></msup><mspace width="0.167em"></mspace><mrow><mi mathvariant="normal">K</mi><mi mathvariant="normal">L</mi></mrow><mrow><mo stretchy="true" form="prefix">(</mo><msubsup><mi>p</mi><mi>t</mi><mrow><mo stretchy="true" form="prefix">(</mo><mi>T</mi><mo stretchy="true" form="postfix">)</mo></mrow></msubsup><mspace width="0.167em"></mspace><mo stretchy="false" form="postfix">β₯</mo><mspace width="0.167em"></mspace><msubsup><mi>p</mi><mi>s</mi><mrow><mo stretchy="true" form="prefix">(</mo><mi>T</mi><mo stretchy="true" form="postfix">)</mo></mrow></msubsup><mo stretchy="true" form="postfix">)</mo></mrow><mi>.</mi></mrow><annotation encoding="application/x-tex"> | |
| \mathcal{L}_{\mathrm{base}}=(1-\alpha)\,\mathrm{CE}(y,p_s) | |
| +\alpha T^2\,\mathrm{KL}(p_t^{(T)}\,\|\,p_s^{(T)}). | |
| </annotation></semantics></math></p> | |
| <p>Try | |
| <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mi>T</mi><mo>=</mo><mn>2</mn></mrow><annotation encoding="application/x-tex">T=2</annotation></semantics></math> | |
| and a small grid for | |
| <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>Ξ±</mi><annotation encoding="application/x-tex">\alpha</annotation></semantics></math>, | |
| selected only on development data. Preserve hard provenance labels when | |
| a teacher is confidently wrong. Cache logits with model revision, | |
| tokenizer, text checksum and class mapping. Teacher use of test data is | |
| still training contamination after distillation.</p> | |
| <p>Then ablate one extra mechanism at a time: clean/rewrite consistency | |
| for transformations that preserve the task label; ranking against hard | |
| human negatives; and group-balanced sampling or group-aware validation. | |
| MELD already combines auxiliary heads, an EMA consistency teacher and | |
| hard-negative ranking. Reproducing those components is a strong | |
| baseline, not a new method claim. <span class="citation" data-cites="meld">(C. Li, Wan, and Paetzold 2026)</span></p> | |
| <h2 data-number="6.4" id="research-candidate-stability-of-the-human-score-tail-after-export"><span class="header-section-number">6.4</span> Research candidate: stability | |
| of the human-score tail after export</h2> | |
| <p>The proposed method-level question is whether training and | |
| calibration designed around the <em>exported</em> score distribution | |
| improve low-FPR transfer more than ordinary distillation plus | |
| post-training quantization.</p> | |
| <p>Evaluate FP32/BF16, weight-only INT8, and any candidate | |
| activation-quantized version on identical calibration and test groups. | |
| Track the change in human-tail quantiles, decision flips, worst-group | |
| FPR and rewrite recall. If export hurts tail separation, investigate a | |
| quantization-aware consistency term or selective higher precision for | |
| sensitive layers, keeping training data and tuning budgets matched.</p> | |
| <p>This is a hypothesis to test, not an established first. A negative | |
| result is informative: post-export calibration alone may be sufficient, | |
| or a different student architecture may be more effective. Quantization | |
| is not lossless, and runtime documentation explicitly recommends | |
| inspecting where numerical changes arise. <span class="citation" data-cites="ortquant">(ONNX Runtime contributors 2026c)</span></p> | |
| <p>Temperature scaling can improve probability calibration on a matching | |
| population; it cannot create missing class separation or guarantee FPR | |
| under distribution shift. Use it as a calibration baseline, not as the | |
| central explanation for OOD gains.</p> | |
| <h1 data-number="7" id="evaluation-protocol-for-a-publishable-result"><span class="header-section-number">7</span> Evaluation protocol for a | |
| publishable result</h1> | |
| <h2 data-number="7.1" id="separate-four-roles-for-data"><span class="header-section-number">7.1</span> Separate four roles for | |
| data</h2> | |
| <p>Maintain distinct <strong>training</strong>, <strong>model-selection | |
| development</strong>, <strong>threshold calibration</strong>, and | |
| <strong>final test</strong> partitions. Split before augmentation, | |
| grouping originals, descendants, near duplicates and author/source | |
| families when available. Keep generator/rewriter holdouts distinct from | |
| ordinary random source-group splits.</p> | |
| <p>Each row should include a stable ID, source-group ID, text checksum, | |
| language, domain, provenance label, generator and revision, edit method, | |
| generation date, source/license information and split. Manifest joins | |
| must require a prediction or an explicit invalid result for every | |
| expected ID. Missing outputs must not silently improve recall or | |
| false-positive estimates.</p> | |
| <p>Report training-data exposure for every released comparator. A | |
| teacher or baseline may have seen benchmark training sources or | |
| unavailable private data. Separate verified disjoint tests from tests | |
| where exposure is unknown; do not claim that one modelβs data advantage | |
| demonstrates an architectural advantage.</p> | |
| <h2 data-number="7.2" id="primary-metrics"><span class="header-section-number">7.2</span> Primary metrics</h2> | |
| <p>The main deployment table should report, per dataset and | |
| subgroup:</p> | |
| <ul> | |
| <li>TPR/recall at the calibration-selected fixed threshold and the | |
| <strong>actual test FPR</strong>.</li> | |
| <li>Human false-flag counts, denominators and 95% confidence | |
| intervals.</li> | |
| <li>AUROC and partial AUROC over the low-FPR region as | |
| threshold-independent diagnostics.</li> | |
| <li>Precision at declared base rates, not only the balanced test-set | |
| precision.</li> | |
| <li>Abstention/unsupported-input rates and risk-versus-coverage | |
| curves.</li> | |
| <li>Model bytes, peak runtime memory, cold/warm p50/p95 latency and | |
| supported runtime/device.</li> | |
| </ul> | |
| <p>If 1% of checked documents are actually AI-origin, a detector with | |
| 90% recall and 1% FPR has illustrative positive predictive value:</p> | |
| <p><math display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mfrac><mrow><mn>0.90</mn><mo>Γ</mo><mn>0.01</mn></mrow><mrow><mn>0.90</mn><mo>Γ</mo><mn>0.01</mn><mo>+</mo><mn>0.01</mn><mo>Γ</mo><mn>0.99</mn></mrow></mfrac><mo>=</mo><mn>47.6</mn><mi>%</mi><mi>.</mi></mrow><annotation encoding="application/x-tex">\frac{0.90\times0.01}{0.90\times0.01+0.01\times0.99}=47.6\%.</annotation></semantics></math></p> | |
| <p>This is a base-rate calculation, not a measured DraftMark deployment | |
| result. It explains why a seemingly small FPR still matters for a | |
| writing tool.</p> | |
| <h2 data-number="7.3" id="calibration-and-statistical-uncertainty"><span class="header-section-number">7.3</span> Calibration and statistical | |
| uncertainty</h2> | |
| <p>Select a threshold on calibration human scores, with a documented | |
| strict/non-strict tie rule. Freeze it before test evaluation. For | |
| multiple supported populations, predeclare whether the policy uses one | |
| global threshold or observable language/length buckets. Do not silently | |
| select a more favorable threshold after observing a test domain.</p> | |
| <p>For initial development, collect several thousand independent human | |
| source groups. For a credible 1% deployment claim, aim for roughly | |
| <strong>10,000 human examples per priority population</strong> where | |
| practical, reporting both sample and independent-group counts. At a true | |
| 1% rate, a simple independent-binomial standard error with 10,000 cases | |
| is about 0.10 percentage points; correlated documents reduce effective | |
| information. For a 0.1% target, the necessary human test scale is | |
| substantially larger.</p> | |
| <p>Zero events provide a useful sanity check: the approximate one-sided | |
| 95% rule-of-three upper bound is | |
| <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mn>3</mn><mi>/</mi><mi>n</mi></mrow><annotation encoding="application/x-tex">3/n</annotation></semantics></math>. | |
| With 220 essays, that is about 1.36%; the two-sided Wilson interval used | |
| by the release is wider. State the interval convention rather than | |
| mixing bounds.</p> | |
| <p>Use paired source-group bootstrap intervals for differences between | |
| models on the same examples. Report seed variation separately from test | |
| sampling uncertainty. Frozen-calibrator test bootstrap intervals are | |
| conditional on the chosen threshold; if making a stronger claim about | |
| calibration stability, also resample the calibration groups and refit | |
| thresholds. When many subgroup constraints are asserted simultaneously, | |
| account for multiple comparisons or label intervals as exploratory.</p> | |
| <h2 data-number="7.4" id="concrete-success-criteria-defined-before-the-next-test"><span class="header-section-number">7.4</span> Concrete success criteria, | |
| defined before the next test</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 50%" /> | |
| <col style="width: 50%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Goal</th> | |
| <th style="text-align: left;">Proposed gate, not an achieved result</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Existing product regression</td> | |
| <td style="text-align: left;">Improve both rewrite families; target at | |
| least 80% recall per family while retaining approximately 92% clean | |
| recall.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Human false flags</td> | |
| <td style="text-align: left;">Target <=1% observed FPR on independent | |
| general and learner sets; publish intervals. Use an | |
| upper-confidence-bound criterion for a statistically supported bound on | |
| the evaluated population, with independence/clustering assumptions | |
| stated.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Compact public benchmark</td> | |
| <td style="text-align: left;">Match or exceed 98.91% official RAID TPR | |
| at 1% FPR, with the exact export evaluated separately.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Paraphrase transfer</td> | |
| <td style="text-align: left;">Beat the compact comparator on identical | |
| MAGE inputs; aim toward the 1B reference of 0.919 AUROC while improving | |
| fixed-threshold recall.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Hard web rewriting</td> | |
| <td style="text-align: left;">Beat the strongest rerun comparator on | |
| WATTLE at matched actual FPR; published RADAR 80.3/75.4 AUROC values are | |
| starting references.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Device size</td> | |
| <td style="text-align: left;">Model plus tokenizer below 160 MB; | |
| separately publish measured peak memory.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Desktop latency</td> | |
| <td style="text-align: left;">Preserve or improve the current appβs warm | |
| p95 on the same device and text distribution.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Mobile</td> | |
| <td style="text-align: left;">Establish a measured budget on named | |
| iPhone and Android models; do not infer it from M4 Pro results.</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <h1 data-number="8" id="h100-training-plan"><span class="header-section-number">8</span> H100 training plan</h1> | |
| <h2 data-number="8.1" id="start-on-one-gpu"><span class="header-section-number">8.1</span> Start on one GPU</h2> | |
| <p>The current 125M detector is small enough that a single H100 is the | |
| appropriate first allocation. Training with BF16 autocast while | |
| retaining FP32 optimizer states is a straightforward starting point. A | |
| rough full-Adam parameter-state budget of 16 bytes per parameter implies | |
| about 2 GB for 125M parameters and 16 GB for 1B, before activations, | |
| temporary buffers and framework overhead. These are arithmetic memory | |
| estimates, not measured allocations; the exact software stack and | |
| batch/sequence sizes determine actual usage. <span class="citation" data-cites="torchamp">(PyTorch contributors 2026)</span></p> | |
| <p>Do not begin by distributing the original 6,339-example experiment | |
| across four H100s. First establish correctness and throughput on one. | |
| Multi-GPU training becomes useful for larger data, teacher sweeps or | |
| long sequences after profiling.</p> | |
| <h2 data-number="8.2" id="run-sequence"><span class="header-section-number">8.2</span> Run sequence</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 33%" /> | |
| <col style="width: 33%" /> | |
| <col style="width: 33%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Run</th> | |
| <th style="text-align: left;">Change</th> | |
| <th style="text-align: left;">Purpose</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">R0</td> | |
| <td style="text-align: left;">Frozen released INT8 artifact</td> | |
| <td style="text-align: left;">Establish the application baseline and | |
| score policy.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">R1</td> | |
| <td style="text-align: left;">Pinned TMR base, original four-layer | |
| recipe</td> | |
| <td style="text-align: left;">Reproduce the original adaptation as | |
| closely as available source data permit.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">R2</td> | |
| <td style="text-align: left;">DraftMark v3 warm start, added | |
| rewrite/human groups</td> | |
| <td style="text-align: left;">Measure the value of better data without | |
| changing architecture.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">R3</td> | |
| <td style="text-align: left;">Same data, all layers trainable</td> | |
| <td style="text-align: left;">Isolate adaptation capacity.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">R4</td> | |
| <td style="text-align: left;">Same data, 384 versus 512 training | |
| tokens</td> | |
| <td style="text-align: left;">Test the train/inference context | |
| mismatch.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">R5</td> | |
| <td style="text-align: left;">Same data, Ettin-150M and selected larger | |
| teacher</td> | |
| <td style="text-align: left;">Establish a controlled architecture | |
| comparison.</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">R6</td> | |
| <td style="text-align: left;">Distilled 150M and 68M students</td> | |
| <td style="text-align: left;">Measure the quality-size frontier.</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">R7</td> | |
| <td style="text-align: left;">Export/calibration ablations</td> | |
| <td style="text-align: left;">Test whether tail stability survives | |
| quantization and runtime changes.</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>The original published recipe is last four layers plus classifier, | |
| 384-token training cap, global batch 16, AdamW learning rate | |
| <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mn>2</mn><mo>Γ</mo><msup><mn>10</mn><mrow><mi>β</mi><mn>5</mn></mrow></msup></mrow><annotation encoding="application/x-tex">2\times10^{-5}</annotation></semantics></math>, | |
| weight decay 0.01, clipping 1.0, three epochs and seed 36513. Its cohort | |
| weights are learner 1.45, decoder 1.30, seq2seq 0.90 and encoder 1.0. | |
| Match these first; increasing batch size is a new experiment, not a | |
| transparent acceleration. <span class="citation" data-cites="draftmarkhandoff">(DraftMark project 2026a)</span></p> | |
| <p>For the expanded-data run, profile sequence lengths and try a small | |
| development-only grid: learning rates | |
| <math display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mn>4</mn><mo>Γ</mo><msup><mn>10</mn><mrow><mi>β</mi><mn>5</mn></mrow></msup></mrow><annotation encoding="application/x-tex">1,2,4\times10^{-5}</annotation></semantics></math>, | |
| 384/512 tokens, and one to three epochs. Use microbatches and | |
| accumulation to reach a specified effective batch without changing it | |
| accidentally. Record all decisions before final-test access.</p> | |
| <h2 data-number="8.3" id="warm-start-reconstruction"><span class="header-section-number">8.3</span> Warm-start reconstruction</h2> | |
| <p>The public training branch contains replacement tensors for the | |
| trained layers. They must <strong>replace</strong> the corresponding | |
| tensors in the pinned TMR base, not be added as a LoRA delta. Validate | |
| the expected key set, label order, parameter count and SHA-256, then | |
| initialize a new optimizer for continued training. The supplied training | |
| kit implements this reconstruction and a new single-GPU entry point. It | |
| is not the unpublished original trainer and cannot promise bitwise | |
| reproduction. <span class="citation" data-cites="draftmarkhandoff">(DraftMark project 2026a)</span></p> | |
| <p>The kit supports source-grouped local JSONL inputs, weighted loss, | |
| optional cached teacher logits, checkpointing/resume and run provenance. | |
| The separate evaluator operates on prediction files from the <em>actual | |
| deployment runtime</em>, so native training validation cannot | |
| accidentally substitute for exported-browser evaluation.</p> | |
| <h2 data-number="8.4" id="estimate-time-from-a-pilot"><span class="header-section-number">8.4</span> Estimate time from a pilot</h2> | |
| <p>Measure examples/second and tokens/second after warm-up on the | |
| intended batch and sequence distribution. Estimate training time as:</p> | |
| <p><math display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mrow><msub><mi>t</mi><mrow><mi mathvariant="normal">h</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">u</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">s</mi></mrow></msub><mo>β</mo><mfrac><mrow><msub><mi>N</mi><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">m</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi></mrow></msub><mspace width="0.167em"></mspace><mi>E</mi></mrow><mrow><mn>3600</mn><mspace width="0.167em"></mspace><msub><mi>r</mi><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">m</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">s</mi><mi mathvariant="normal">/</mi><mi mathvariant="normal">s</mi><mi mathvariant="normal">e</mi><mi mathvariant="normal">c</mi></mrow></msub></mrow></mfrac><mo>+</mo><msub><mi>t</mi><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">v</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">l</mi><mi mathvariant="normal">u</mi><mi mathvariant="normal">a</mi><mi mathvariant="normal">t</mi><mi mathvariant="normal">i</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">n</mi></mrow></msub><mo>+</mo><msub><mi>t</mi><mrow><mi mathvariant="normal">e</mi><mi mathvariant="normal">x</mi><mi mathvariant="normal">p</mi><mi mathvariant="normal">o</mi><mi mathvariant="normal">r</mi><mi mathvariant="normal">t</mi></mrow></msub><mi>.</mi></mrow><annotation encoding="application/x-tex">t_{\mathrm{hours}}\approx\frac{N_{\mathrm{examples}}\,E}{3600\,r_{\mathrm{examples/sec}}}+t_{\mathrm{evaluation}}+t_{\mathrm{export}}.</annotation></semantics></math></p> | |
| <p>For illustration only, 300,000 examples across three epochs at 50 | |
| examples/second would require five hours of training computation before | |
| evaluation/export. The throughput is a hypothetical input, not an H100 | |
| benchmark. Teacher scoring, dataset generation and repeated app | |
| evaluation may dominate the overall research schedule. Avoid committing | |
| to an H100-hour budget until the pilot is measured.</p> | |
| <h1 data-number="9" id="local-app-development"><span class="header-section-number">9</span> Local app development</h1> | |
| <h2 data-number="9.1" id="build-on-the-existing-browser-product"><span class="header-section-number">9.1</span> Build on the existing browser | |
| product</h2> | |
| <p>DraftMark already has a local browser application. The next app | |
| milestone should make the model lifecycle and offline behavior reliable: | |
| package the runtime, tokenizer, model and decision policy under | |
| versioned hashes; cache them; verify a full reload and scan in airplane | |
| mode after installation; and keep processing off the UI thread. | |
| Downloading assets once is compatible with local inference, but it is | |
| different from working offline before the first installation. <span class="citation" data-cites="draftmark ortweb">(DraftMark project 2026b; | |
| ONNX Runtime contributors 2026b)</span></p> | |
| <p>Keep the existing weight-only INT8 path as the compatibility | |
| baseline. The handoff states that activation-quantized trials produced | |
| browser/native score drift. Do not replace it with generic dynamic | |
| quantization just because another model uses that format. After any | |
| export or runtime change, refit calibration on the shipped graph and | |
| inspect threshold flips, especially among high-scoring human texts. | |
| <span class="citation" data-cites="draftmarkhandoff ortquant">(DraftMark | |
| project 2026a; ONNX Runtime contributors 2026c)</span></p> | |
| <p>ONNX Runtime Web can use WASM and other execution providers. | |
| Multithreaded WASM requires the browserβs support and cross-origin | |
| isolation. WebGPU should be an optional measured backend, with operator | |
| compatibility and parity verified, rather than a promised speedup. | |
| Runtime-native mobile providers also have device-specific behavior and | |
| may fall back or partition graphs. <span class="citation" data-cites="ortflags ortmobile">(ONNX Runtime contributors 2026d, | |
| 2026a)</span></p> | |
| <h2 data-number="9.2" id="app-rollout-sequence"><span class="header-section-number">9.2</span> App rollout sequence</h2> | |
| <table> | |
| <colgroup> | |
| <col style="width: 33%" /> | |
| <col style="width: 33%" /> | |
| <col style="width: 33%" /> | |
| </colgroup> | |
| <thead> | |
| <tr class="header"> | |
| <th style="text-align: left;">Stage</th> | |
| <th style="text-align: left;">Deliverable</th> | |
| <th style="text-align: left;">Acceptance evidence</th> | |
| </tr> | |
| </thead> | |
| <tbody> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Browser/PWA</td> | |
| <td style="text-align: left;">Existing UI plus versioned offline | |
| assets</td> | |
| <td style="text-align: left;">Reload and inference without network; no | |
| text-upload requests; safe cache update/recovery</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Desktop</td> | |
| <td style="text-align: left;">Lightweight wrapper or native ONNX | |
| runner</td> | |
| <td style="text-align: left;">Same tokenizer/windowing/policy; offline | |
| installer; memory and latency measurements</td> | |
| </tr> | |
| <tr class="odd"> | |
| <td style="text-align: left;">Mobile</td> | |
| <td style="text-align: left;">iOS/Android packaging after browser | |
| profiling</td> | |
| <td style="text-align: left;">Device-specific memory, thermal and power | |
| results; foreground/background recovery</td> | |
| </tr> | |
| <tr class="even"> | |
| <td style="text-align: left;">Integration</td> | |
| <td style="text-align: left;">Selection/share extension or local | |
| CLI</td> | |
| <td style="text-align: left;">Explicit user action; versioned reports; | |
| no hidden inference API fallback</td> | |
| </tr> | |
| </tbody> | |
| </table> | |
| <p>The minimal user flow is paste/import, check locally, and review a | |
| cautious evidence result. Use states such as strong AI-like signal, | |
| inconclusive and outside the supported range. A detector score is not | |
| proof of misconduct. Show model version and supported-language/length | |
| information where useful; do not present an uncalibrated softmax as an | |
| authorship percentage.</p> | |
| <p>Measure at least short, medium and long supported drafts. Report | |
| download/install separately from cold model load and warm inference, | |
| plus UI responsiveness, batch-one latency, peak RAM and repeated-scan | |
| thermal behavior. Model file size alone is not RAM usage. On-device | |
| optimization should minimize cost at matched error rates; a smaller | |
| graph that sharply increases learner false flags is not a successful | |
| product optimization. <span class="citation" data-cites="ortmobile">(ONNX Runtime contributors 2026a)</span></p> | |
| <h1 data-number="10" id="a-focused-thai-english-research-extension"><span class="header-section-number">10</span> A focused Thai-English research | |
| extension</h1> | |
| <p>This extension fits a multilingual research program, but it should | |
| not delay the first rigorous English release. DetectRL-X and MULTITuDE | |
| leave Thai outside their language sets. That is a coverage gap in those | |
| benchmarks, not proof that no Thai detector or dataset exists. <span class="citation" data-cites="detectrlx multitude">(Wu et al. 2026; Macko | |
| et al. 2023)</span></p> | |
| <p>Start from a Thai-capable multilingual teacher. Build a | |
| provenance-controlled paired set with Thai originals, English originals | |
| and naturally code-switched text. Include locally written academic, | |
| professional and everyday prose; translated text; Thai speakersβ | |
| English; grammar assistance; heavier rewriting; and mixed human/AI | |
| drafts. Preserve authors, sources and rewriting lineage across | |
| splits.</p> | |
| <p>Thai should not use an English whitespace word-count gate. Specify | |
| character or tokenizer-based length buckets and audit tokenization | |
| efficiency, sentence segmentation, Unicode normalization and code | |
| switching. Avoid normalizers that erase meaningful Thai marks. Test | |
| whether a compact shared student, separate language models, or a | |
| language-routed design gives the best quality/memory tradeoff. | |
| Calibration buckets must be based on observable supported language | |
| information and tested under code switching.</p> | |
| <p>A useful publication claim would be: <strong>a provenance-controlled | |
| evaluation of compact local detectors under Thai-English editing and | |
| generator shift, with uncertainty and device cost measured | |
| together</strong>. Claiming the first Thai AI detector would require a | |
| dedicated novelty search beyond this review.</p> | |
| <h1 data-number="11" id="report-and-publication-strategy"><span class="header-section-number">11</span> Report and publication | |
| strategy</h1> | |
| <h2 data-number="11.1" id="what-can-be-published-now"><span class="header-section-number">11.1</span> What can be published now</h2> | |
| <p>The current public model supports a transparent engineering report: | |
| its training provenance, exact shipped artifact, cohort-level regression | |
| results, confidence intervals and reference-device latency. The report | |
| should acknowledge the seq2seq recall loss and limited human-test scale. | |
| Its defensible claim is improved performance over the previous DraftMark | |
| policy on that holdout, not state of the art across AI-text detection. | |
| <span class="citation" data-cites="draftmarkmetrics">(DraftMark project | |
| 2026c)</span></p> | |
| <p>This document can accompany that release as a research roadmap and | |
| benchmark audit. It intentionally distinguishes reported evidence from | |
| experiments still to run.</p> | |
| <h2 data-number="11.2" id="what-would-justify-a-stronger-paper"><span class="header-section-number">11.2</span> What would justify a stronger | |
| paper</h2> | |
| <p>Frame the next paper around one main hypothesis: <strong>quantized | |
| compact encoders can retain useful low-FPR generalization under unseen | |
| rewriting and editing while reducing device cost relative to larger | |
| detectors</strong>. A bilingual extension can test whether this survives | |
| under language and code-switching shifts.</p> | |
| <p>The minimum convincing experiment set is:</p> | |
| <ol type="1"> | |
| <li>A matched comparison of current released detectors on public | |
| testbeds and a fresh untouched transfer test.</li> | |
| <li>A same-data student/teacher study separating architecture, | |
| additional data and distillation effects.</li> | |
| <li>Ablations of rewrite balancing, hard human negatives and the | |
| proposed export-aware method.</li> | |
| <li>Calibration on separate data, with subgroup FPR counts/intervals and | |
| a clearly defined abstention policy.</li> | |
| <li>Native-versus-exported quality, browser/device latency and | |
| peak-memory measurements using identical documents.</li> | |
| <li>Transparent limitations, source-exposure audit, task-label mappings, | |
| and enough manifests/predictions to reproduce the findings where | |
| licenses allow.</li> | |
| </ol> | |
| <p>Report paired intervals for the principal improvements and three | |
| training seeds for the final selected recipe when feasible. Choose the | |
| primary benchmark and metric before searching many variants. A 0.1-point | |
| gain on a saturated aggregate with a weaker human-error tail is not the | |
| intended success.</p> | |
| <p>Potential formats include a technical report first, then an NLP | |
| system-demonstration or applied/industry paper if deployment is the main | |
| contribution, or a methods/benchmark paper if transfer and the | |
| Thai-English study provide substantial new evidence. Venue eligibility, | |
| deadlines and disclosure requirements must be checked when a submission | |
| is selected; none is assumed here.</p> | |
| <h2 data-number="11.3" id="suggested-paper-outline"><span class="header-section-number">11.3</span> Suggested paper outline</h2> | |
| <p>Use an introduction grounded in real local-writing use cases, | |
| followed by task/provenance definitions, related work, data/splits, | |
| method, calibration protocol, main comparisons, quantization/device | |
| analysis and error analysis. The main results table should include both | |
| quality and measured cost. Place exact benchmark adapters, prompts, | |
| source licensing and checkpoint/runtime hashes in the appendix and | |
| repository.</p> | |
| <p>The release package should contain model weights, tokenizer, label | |
| mapping, runtime/policy versions, calibration provenance, a benchmark | |
| table with evidence status, reproducible commands and device results. | |
| Where raw text cannot be redistributed, release permitted IDs/hashes and | |
| reconstruction instructions instead of pretending an aggregate JSON | |
| alone is fully reproducible.</p> | |
| <h1 data-number="12" id="prioritized-execution-plan"><span class="header-section-number">12</span> Prioritized execution plan</h1> | |
| <p><strong>First, reproduce the current productβs behavior and establish | |
| the comparison harness.</strong> Recover the original source checkout | |
| and dataset manifests; restore the pinned model; generate | |
| browser-runtime predictions; rerun the compact and 1B competitors on the | |
| same independent calibration/test sets. This determines whether the | |
| biggest deficit is model separation, threshold transfer, input handling | |
| or the chosen text populations.</p> | |
| <p><strong>Second, run the one-H100 controlled adaptation.</strong> | |
| Expand decoder rewrites and difficult human negatives, retain seq2seq | |
| coverage, compare partial/full unfreezing, and evaluate with a fresh | |
| final set. Only advance to a new encoder after the data and calibration | |
| baselines are understood.</p> | |
| <p><strong>Third, train the compact student and validate the | |
| export.</strong> Distill into 150M/68M candidates, compare export | |
| formats, freeze calibration, and measure on the target desktop and | |
| mobile devices. Use these results to decide which model belongs in each | |
| local tier.</p> | |
| <p><strong>Fourth, publish the verified result and extend the | |
| study.</strong> Release the English evidence and app update when the | |
| gates are met. Develop the Thai-English provenance-controlled benchmark | |
| as a distinct research contribution, keeping the public paperβs claims | |
| tied to the actual measured populations and devices.</p> | |
| <h1 data-number="13" id="references"><span class="header-section-number">13</span> References</h1> | |
| <div id="refs" class="references csl-bib-body hanging-indent" data-entry-spacing="0" role="list"> | |
| <div id="ref-fastdetect" class="csl-entry" role="listitem"> | |
| Bao, Guangsheng et al. 2026. <span>β<span class="nocase">Fast-DetectGPT: | |
| Official Implementation and Model-Pair Updates</span>.β</span> <a href="https://github.com/baoguangsheng/fast-detect-gpt">https://github.com/baoguangsheng/fast-detect-gpt</a>. | |
| </div> | |
| <div id="ref-diveye" class="csl-entry" role="listitem"> | |
| Basani, Advik Raj, and Pin-Yu Chen. 2026. <span>β<span>Diversity Boosts | |
| AI-Generated Text Detection</span>.β</span> <em>arXiv Preprint | |
| arXiv:2509.18880v3</em>. <a href="https://arxiv.org/html/2509.18880v3">https://arxiv.org/html/2509.18880v3</a>. | |
| </div> | |
| <div id="ref-wattle" class="csl-entry" role="listitem"> | |
| Chernikov, Oleg, Evgenii Ptitsyn, and Albina Burlova. 2026. <span>β<span class="nocase">PAWN++ in the Wild: Evaluating AI-Generated Text | |
| Detection on the Web</span>.β</span> In <em>Proceedings of the 13th | |
| Web-as-Corpus Workshop</em>, 282β96. <a href="https://doi.org/10.18653/2026.wac-1.25">https://doi.org/10.18653/2026.wac-1.25</a>. | |
| </div> | |
| <div id="ref-draftmarkhandoff" class="csl-entry" role="listitem"> | |
| DraftMark project. 2026a. <span>β<span>DraftMark GPU Training | |
| Handoff</span>.β</span> <a href="https://huggingface.co/saksornr/draftmark-roberta-int8/blob/068ccc7ba34f134526c9bbc72ac2033f00a40d45/handoff.md">https://huggingface.co/saksornr/draftmark-roberta-int8/blob/068ccc7ba34f134526c9bbc72ac2033f00a40d45/handoff.md</a>. | |
| </div> | |
| <div id="ref-draftmark" class="csl-entry" role="listitem"> | |
| βββ. 2026b. <span>β<span class="nocase">DraftMark RoBERTa FT INT8: Model | |
| Card and Release Manifest</span>.β</span> <a href="https://huggingface.co/saksornr/draftmark-roberta-int8/tree/068ccc7ba34f134526c9bbc72ac2033f00a40d45">https://huggingface.co/saksornr/draftmark-roberta-int8/tree/068ccc7ba34f134526c9bbc72ac2033f00a40d45</a>. | |
| </div> | |
| <div id="ref-draftmarkmetrics" class="csl-entry" role="listitem"> | |
| βββ. 2026c. <span>β<span class="nocase">DraftMark v3 Aggregate Training, | |
| Calibration, Holdout and Runtime Report</span>.β</span> <a href="https://huggingface.co/saksornr/draftmark-roberta-int8/blob/068ccc7ba34f134526c9bbc72ac2033f00a40d45/finetune-v3.json">https://huggingface.co/saksornr/draftmark-roberta-int8/blob/068ccc7ba34f134526c9bbc72ac2033f00a40d45/finetune-v3.json</a>. | |
| </div> | |
| <div id="ref-raidrepo" class="csl-entry" role="listitem"> | |
| Dugan, Liam et al. 2026. <span>β<span class="nocase">RAID Official | |
| Repository, Evaluator and Leaderboard Snapshot</span>.β</span> <a href="https://github.com/liamdugan/raid/tree/ac4ce10fe8fe096b0c4a01396f1e437aaf35bd53">https://github.com/liamdugan/raid/tree/ac4ce10fe8fe096b0c4a01396f1e437aaf35bd53</a>. | |
| </div> | |
| <div id="ref-binoculars" class="csl-entry" role="listitem"> | |
| Hans, Abhimanyu et al. 2024. <span>β<span class="nocase">Binoculars: | |
| Zero-Shot Detection of LLM-Generated Text</span>.β</span> <a href="https://github.com/ahans30/Binoculars">https://github.com/ahans30/Binoculars</a>. | |
| </div> | |
| <div id="ref-ettin" class="csl-entry" role="listitem"> | |
| JHU CLSP. 2025. <span>β<span>Ettin Encoder Model Family: Official Model | |
| Card</span>.β</span> <a href="https://huggingface.co/jhu-clsp/ettin-encoder-150m">https://huggingface.co/jhu-clsp/ettin-encoder-150m</a>. | |
| </div> | |
| <div id="ref-meld" class="csl-entry" role="listitem"> | |
| Li, Chenjun, Cheng Wan, and Johannes C. Paetzold. 2026. <span>β<span class="nocase">MELD: Multi-Task Equilibrated Learning Detector for | |
| AI-Generated Text</span>.β</span> <em>arXiv Preprint | |
| arXiv:2605.06903v1</em>. <a href="https://arxiv.org/html/2605.06903v1">https://arxiv.org/html/2605.06903v1</a>. | |
| </div> | |
| <div id="ref-mage" class="csl-entry" role="listitem"> | |
| Li, Yafu et al. 2024. <span>β<span class="nocase">MAGE: | |
| Machine-Generated Text Detection in the Wild</span>.β</span> <a href="https://github.com/yafuly/MAGE">https://github.com/yafuly/MAGE</a>. | |
| </div> | |
| <div id="ref-multitude" class="csl-entry" role="listitem"> | |
| Macko, Dominik et al. 2023. <span>β<span>MULTITuDE: Large-Scale | |
| Multilingual Machine-Generated Text Detection Benchmark</span>.β</span> | |
| In <em>Proceedings of EMNLP</em>, 9960β87. <a href="https://doi.org/10.18653/v1/2023.emnlp-main.616">https://doi.org/10.18653/v1/2023.emnlp-main.616</a>. | |
| </div> | |
| <div id="ref-mdeberta" class="csl-entry" role="listitem"> | |
| Microsoft. 2021. <span>β<span class="nocase">mDeBERTa-v3-base: Official | |
| Model Card</span>.β</span> <a href="https://huggingface.co/microsoft/mdeberta-v3-base">https://huggingface.co/microsoft/mdeberta-v3-base</a>. | |
| </div> | |
| <div id="ref-gpttrace150" class="csl-entry" role="listitem"> | |
| Nakas, Andrew. 2026a. <span>β<span class="nocase">GPTTrace Text | |
| Detector: Ettin-150M v5e, INT8 ONNX</span>.β</span> <a href="https://huggingface.co/nakas/gpttrace-ettin150-text-v5e">https://huggingface.co/nakas/gpttrace-ettin150-text-v5e</a>. | |
| </div> | |
| <div id="ref-gpttrace1b" class="csl-entry" role="listitem"> | |
| βββ. 2026b. <span>β<span class="nocase">GPTTrace Text Detector: Ettin-1B | |
| v8s</span>.β</span> <a href="https://huggingface.co/nakas/gpttrace-v8s-1b">https://huggingface.co/nakas/gpttrace-v8s-1b</a>. | |
| </div> | |
| <div id="ref-ortmobile" class="csl-entry" role="listitem"> | |
| ONNX Runtime contributors. 2026a. <span>β<span class="nocase">Deploy on | |
| Mobile</span>.β</span> <a href="https://onnxruntime.ai/docs/tutorials/mobile/">https://onnxruntime.ai/docs/tutorials/mobile/</a>. | |
| </div> | |
| <div id="ref-ortweb" class="csl-entry" role="listitem"> | |
| βββ. 2026b. <span>β<span>Deploying ONNX Runtime Web</span>.β</span> <a href="https://onnxruntime.ai/docs/tutorials/web/deploy.html">https://onnxruntime.ai/docs/tutorials/web/deploy.html</a>. | |
| </div> | |
| <div id="ref-ortquant" class="csl-entry" role="listitem"> | |
| βββ. 2026c. <span>β<span>Quantize ONNX Models</span>.β</span> <a href="https://onnxruntime.ai/docs/performance/model-optimizations/quantization.html">https://onnxruntime.ai/docs/performance/model-optimizations/quantization.html</a>. | |
| </div> | |
| <div id="ref-ortflags" class="csl-entry" role="listitem"> | |
| βββ. 2026d. <span>β<span class="nocase">Web Environment Flags and | |
| Session Options</span>.β</span> <a href="https://onnxruntime.ai/docs/tutorials/web/env-flags-and-session-options.html">https://onnxruntime.ai/docs/tutorials/web/env-flags-and-session-options.html</a>. | |
| </div> | |
| <div id="ref-tmr" class="csl-entry" role="listitem"> | |
| Oxidane. 2025. <span>β<span class="nocase">TMR: Target Mining RoBERTa | |
| for AI Text Detection</span>.β</span> <a href="https://huggingface.co/Oxidane/tmr-ai-text-detector">https://huggingface.co/Oxidane/tmr-ai-text-detector</a>. | |
| </div> | |
| <div id="ref-pan2026" class="csl-entry" role="listitem"> | |
| PAN organizers. 2026. <span>β<span class="nocase">Voight-Kampff | |
| Generative AI Detection 2026: Task, Metrics and Final | |
| Leaderboard</span>.β</span> <a href="https://pan.webis.de/clef26/pan26-web/generated-content-analysis.html">https://pan.webis.de/clef26/pan26-web/generated-content-analysis.html</a>. | |
| </div> | |
| <div id="ref-torchamp" class="csl-entry" role="listitem"> | |
| PyTorch contributors. 2026. <span>β<span>Automatic Mixed Precision | |
| Documentation</span>.β</span> <a href="https://docs.pytorch.org/docs/stable/amp.html">https://docs.pytorch.org/docs/stable/amp.html</a>. | |
| </div> | |
| <div id="ref-raid150" class="csl-entry" role="listitem"> | |
| RAID evaluation maintainers. 2026a. <span>β<span class="nocase">Official | |
| CI Evaluation: GPTTrace Ettin-150M v5e, Pull Request 225</span>.β</span> | |
| <a href="https://github.com/liamdugan/raid/pull/225">https://github.com/liamdugan/raid/pull/225</a>. | |
| </div> | |
| <div id="ref-raid1b" class="csl-entry" role="listitem"> | |
| βββ. 2026b. <span>β<span class="nocase">Official CI Evaluation: GPTTrace | |
| Ettin-1B v8s, Pull Request 233</span>.β</span> <a href="https://github.com/liamdugan/raid/pull/233">https://github.com/liamdugan/raid/pull/233</a>. | |
| </div> | |
| <div id="ref-vanguard" class="csl-entry" role="listitem"> | |
| Thorat, Shantanu. 2026. <span>β<span>Vanguard AI-Generated Text | |
| Detector: Model Card</span>.β</span> <a href="https://huggingface.co/ShantanuT01/vanguard-ai-text-detector">https://huggingface.co/ShantanuT01/vanguard-ai-text-detector</a>. | |
| </div> | |
| <div id="ref-coling2025" class="csl-entry" role="listitem"> | |
| Wang, Yuxia et al. 2025. <span>β<span class="nocase">GenAI Content | |
| Detection Task 1: English and Multilingual Machine-Generated Text | |
| Detection: AI vs. Human</span>.β</span> In <em>Proceedings of the 1st | |
| Workshop on GenAI Content Detection</em>, 244β61. <a href="https://aclanthology.org/2025.genaidetect-1.27.pdf">https://aclanthology.org/2025.genaidetect-1.27.pdf</a>. | |
| </div> | |
| <div id="ref-detectrlx" class="csl-entry" role="listitem"> | |
| Wu, Junchao et al. 2026. <span>β<span class="nocase">DetectRL-X: Towards | |
| Reliable Multilingual and Real-World LLM-Generated Text | |
| Detection</span>.β</span> In <em>Proceedings of ACL</em>, 38247β94. <a href="https://doi.org/10.18653/v1/2026.acl-long.1773">https://doi.org/10.18653/v1/2026.acl-long.1773</a>. | |
| </div> | |
| </div> | |
| </body> | |
| </html> | |