Spaces:
Running
Running
| <html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1"><meta name="theme-color" content="#081321"><meta name="description" content="Transparent multilingual NLP robustness, reliability, and abstention research using synthetic evaluation data."><title>RobustnessLab AI | Research Reliability Studio</title><link rel="stylesheet" href="styles.css"></head> | |
| <body> | |
| <div class="layout"> | |
| <aside class="sidebar" aria-label="Navigation"><a class="brand" href="#overview"><span class="brand-mark">◈</span><span><strong>ROBUSTNESS<span class="brand-accent">LAB</span></strong><small>Research reliability studio</small></span></a> | |
| <div class="side-section">RESEARCH WORKSPACE</div> | |
| <nav id="sideNav" class="navigation" aria-label="Research modules"> | |
| <button type="button" class="nav-btn active" data-view="overview"><span class="nav-glyph">▦</span><span>01 / Command center</span></button> | |
| <button type="button" class="nav-btn" data-view="matrix"><span class="nav-glyph">▤</span><span>02 / Stress matrix</span></button> | |
| <button type="button" class="nav-btn" data-view="cases"><span class="nav-glyph">⌕</span><span>03 / Paired examples</span></button> | |
| <button type="button" class="nav-btn" data-view="selective"><span class="nav-glyph">◌</span><span>04 / Risk & abstention</span></button> | |
| <button type="button" class="nav-btn" data-view="reliability"><span class="nav-glyph">◉</span><span>05 / Reliability</span></button> | |
| <button type="button" class="nav-btn" data-view="slices"><span class="nav-glyph">▥</span><span>06 / Language slices</span></button> | |
| <button type="button" class="nav-btn" data-view="gates"><span class="nav-glyph">◇</span><span>07 / Research gates</span></button> | |
| <button type="button" class="nav-btn" data-view="sandbox"><span class="nav-glyph">⚙</span><span>08 / Sandbox & export</span></button> | |
| </nav> | |
| <div class="side-info"><span class="signal"></span> <b>OFFLINE-FIRST</b><p>Transparent lexical baselines. No remote inference or paid services.</p><div class="small-pair"><span>Evidence</span><span>SYNTHETIC</span></div><div class="small-pair"><span>Models</span><span>4 baselines</span></div><div class="small-pair"><span>Runtime</span><span>Static · local</span></div></div> | |
| <div class="side-footer">Rodrigo Pantoja · Applied AI Research<br>v1.0 / 2026</div> | |
| </aside> | |
| <main class="main" id="main"> | |
| <header class="topbar"><div class="breadcrumbs">RESEARCH ENGINEERING <span>/</span> MODEL RELIABILITY <span>/</span> <strong>STRESS TESTS</strong></div><div class="top-right"><span class="micro-dot"></span><span id="status">LOADING EVIDENCE</span><span class="build-chip">R&D / 016</span></div></header> | |
| <section class="hero"><div class="hero-copy"><div class="eyebrow">PORTFOLIO RESEARCH SERIES · STUDY 016</div><h1>Robustness<span>Lab AI</span></h1><p>Does a classifier still behave the same when the wording changes? Study multilingual prediction stability, uncertainty, data perturbations and research release gates with fully traceable evidence.</p><div class="hero-badges"><span>EN · ES · PT</span><span>4 lexical baselines</span><span>Paired stress tests</span><span>No paid API</span></div></div><div class="hero-right"><div class="snapshot-head"><span class="signal"></span> EXPERIMENT SNAPSHOT <span class="version">v1.0</span></div><div class="snapshot-grid"><div><b id="snapN">—</b><span>Evaluation records</span></div><div><b>6</b><span>Test conditions</span></div><div><b>3</b><span>Languages</span></div><div><b>2</b><span>Held-out splits</span></div></div><div class="snapshot-foot">Source: Academic Rhetorical Intent Mini · synthetic-curated · no real-world validation</div></div></section> | |
| <section class="content"><div class="toolbar"><div class="control"><label for="methodSelect">CLASSIFIER</label><select id="methodSelect"><option value="hybrid">Hybrid TF-IDF</option><option value="nb">Multinomial Naive Bayes</option><option value="word">Word TF-IDF</option><option value="char">Character TF-IDF</option></select></div><div class="control"><label for="splitSelect">EVALUATION SPLIT</label><select id="splitSelect"><option value="challenge">Novel-pattern · 54</option><option value="iid">Controlled IID · 130</option></select></div><div class="control grow"><label for="attackSelect">STRESS CONDITION</label><select id="attackSelect"><option value="typo">Deterministic typo</option><option value="context">Neutral context sentence</option><option value="synonym">Dictionary substitution</option><option value="format">Case & punctuation</option><option value="truncate">Information removal</option><option value="original">Unchanged original</option></select></div><button type="button" class="toolbar-export" id="exportSummary">↓ Report .md</button></div> | |
| <div class="notice" id="notice"><span class="notice-icon">i</span><span><strong>Synthetic research warning.</strong> Transformation labels are inherited for exploratory diagnostics only. Some changes may alter meaning. A flipped prediction is not automatically evidence of real-world model failure.</span></div> | |
| <div class="view-title"><span id="viewId">01 / COMMAND CENTER</span><h2 id="viewTitle">Robustness overview</h2><p id="viewSubtitle">Paired baseline vs. perturbed evaluation on identical example IDs.</p></div> | |
| <div id="panel" aria-live="polite"><div class="load-state">Loading reproducible evidence…</div></div> | |
| <footer class="footer"><span>◈ RobustnessLab AI · Rodrigo Pantoja · 2026</span><span>Code MIT · Source dataset CC BY 4.0 · Offline research</span></footer> | |
| </section></main></div> | |
| <script src="source_modelops_engine.js"></script><script src="robustness_engine.js"></script><script src="app.js"></script></body></html> | |