CoderDoge commited on
Commit
b5053ae
·
verified ·
1 Parent(s): 5627466

Show WAI results on a labeled 50–100% axis

Browse files
Files changed (1) hide show
  1. index.html +9 -9
index.html CHANGED
@@ -11,9 +11,9 @@
11
  <link rel="icon" href="data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 64 64'%3E%3Crect width='64' height='64' rx='16' fill='%23173d34'/%3E%3Cpath d='M18 33l9 9 20-22' fill='none' stroke='%23c9f2cf' stroke-width='6'/%3E%3C/svg%3E">
12
  <style>
13
  :root{--ink:#17352f;--muted:#5b6964;--green:#287655;--light:#edf4ed;--paper:#fbfcf8;--line:#dce4db;--mono:ui-monospace,SFMono-Regular,Consolas,monospace;--sans:Inter,-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif}
14
- *{box-sizing:border-box}html{scroll-behavior:smooth;scroll-padding-top:80px}body{margin:0;background:var(--paper);color:var(--ink);font-family:var(--sans);font-size:16px;line-height:1.65}a{color:inherit;text-underline-offset:4px}button{font:inherit}button,a{-webkit-tap-highlight-color:transparent}a:focus-visible,button:focus-visible,summary:focus-visible{outline:3px solid #399365;outline-offset:5px}svg{display:block}p,h1,h2,h3{margin-top:0}h1,h2,h3{line-height:1.15;letter-spacing:-.045em}p{color:var(--muted)}.wrap{max-width:1180px;margin:auto;padding:0 40px}.nav{border-bottom:1px solid var(--line);background:var(--paper)}.nav-inner{height:76px;display:flex;align-items:center;justify-content:space-between;gap:24px}.brand{display:flex;align-items:center;gap:10px;font-size:20px;font-weight:750;letter-spacing:-.04em;text-decoration:none}.brand-mark{display:grid;place-items:center;width:30px;height:30px;background:var(--ink);border-radius:8px;color:#d8f1d4}.nav-links{display:flex;align-items:center;gap:28px}.nav-links a{text-decoration:none;font-size:13px;color:var(--muted)}.nav-links a:hover{color:var(--green)}.external{font-family:var(--mono)}.eyebrow{display:flex;align-items:center;gap:9px;color:var(--green);font-family:var(--mono);font-size:11px;text-transform:uppercase;letter-spacing:.13em}.dot{height:6px;width:6px;border-radius:50%;background:var(--green)}.hero{padding:66px 0 45px}.hero-top{display:grid;grid-template-columns:1.08fr 1fr;gap:52px;align-items:center}.wordmark{font-family:Georgia,"Times New Roman",serif;font-size:100px;font-weight:400;letter-spacing:-.07em;line-height:1;margin:24px 0 16px}.hero h1{font-size:39px;line-height:1.14;max-width:570px;font-weight:550;margin-bottom:21px}.hero-copy{font-size:17px;line-height:1.7;max-width:480px}.hero-art{padding:28px 20px 18px;background:var(--light);border:1px solid #d8e4d7;border-radius:18px;position:relative}.art-label{font:10px var(--mono);text-transform:uppercase;letter-spacing:.13em;color:#64816c;margin-bottom:25px;display:flex;justify-content:space-between}.hero-art img{display:block;width:100%;height:auto;mix-blend-mode:multiply}.art-caption{padding-top:20px;border-top:1px solid #d4e1d1;margin-top:22px;display:flex;justify-content:space-between;gap:10px;color:#3b634d;font-size:11px}.resources{display:flex;gap:10px;flex-wrap:wrap;margin-top:28px}.button{display:inline-flex;align-items:center;gap:12px;justify-content:center;padding:10px 18px;border:1px solid var(--line);border-radius:7px;background:transparent;font-size:13px;font-weight:600;text-decoration:none;cursor:pointer;transition:background .15s}.button:hover{background:#e6ede3}.button.primary{background:var(--ink);border-color:var(--ink);color:#fff}.button.primary:hover{background:#285044}.authors{margin-top:38px;padding-top:26px;border-top:1px solid var(--line);font-size:13px;color:#42564b;line-height:1.9}.author-list{display:flex;gap:6px 21px;flex-wrap:wrap}.affiliation{font-family:var(--mono);font-size:11px;letter-spacing:.03em;margin-top:6px;color:#7b857b}.metrics{display:grid;grid-template-columns:repeat(3,1fr);border-top:1px solid var(--line);border-bottom:1px solid var(--line);padding:29px 0 27px;margin:0 0 68px}.metric{padding:0 30px;border-left:1px solid var(--line)}.metric:first-child{padding-left:0;border-left:0}.metric-label{font-size:12px;color:var(--muted);margin-bottom:9px}.metric-value{font-size:45px;letter-spacing:-.06em;line-height:1.05;font-weight:500}.metric-value span{font-size:25px}.metric-note{font-size:11px;color:var(--muted);margin-top:9px}.gain{color:var(--green);font-weight:650}.section{padding:0 0 72px}.section-head{display:grid;grid-template-columns:.67fr 1fr;gap:60px;margin-bottom:32px;align-items:start}.section-number{font:11px var(--mono);letter-spacing:.08em;color:var(--green);margin-bottom:14px}.section-head h2{font-size:34px;font-weight:550;margin-bottom:0}.section-head p{font-size:15px;margin-bottom:0;line-height:1.8}.method-figure{margin:0;border:1px solid var(--line);border-radius:10px;background:#fff;padding:22px}.method-figure img{width:100%;height:auto;display:block}.method-figure figcaption{font-size:11px;line-height:1.6;color:var(--muted);border-top:1px solid var(--line);padding-top:15px;margin-top:18px}.steps{display:grid;grid-template-columns:repeat(3,1fr);gap:30px;margin-top:27px}.step-top{display:flex;align-items:center;gap:10px;margin-bottom:10px}.step-n{font:11px var(--mono);color:var(--green);border:1px solid #c9d9c8;border-radius:50%;height:25px;width:25px;display:grid;place-items:center;flex:none}.step h3{font-size:17px;margin:0;font-weight:600;letter-spacing:-.025em}.step p{font-size:13px;line-height:1.8;margin:0}.results-panel{border:1px solid var(--line);border-radius:10px;background:white;overflow:hidden}.tabs{padding:15px 18px 0;display:flex;gap:8px;border-bottom:1px solid var(--line);background:#f5f7f1;overflow-x:auto}.tab{white-space:nowrap;border:0;border-bottom:3px solid transparent;padding:11px 13px 14px;cursor:pointer;font-size:12px;color:var(--muted);background:transparent}.tab[aria-selected="true"]{border-bottom-color:var(--green);color:var(--ink);font-weight:650}.result-content{padding:28px 30px 25px}.result-topline{display:flex;align-items:start;justify-content:space-between;gap:25px;margin-bottom:24px}.result-topline h3{font-size:21px;letter-spacing:-.025em;margin-bottom:7px}.result-topline p{font-size:12px;max-width:650px;margin-bottom:0}.pill{font:10px var(--mono);background:var(--light);border-radius:5px;color:var(--green);padding:6px 9px;white-space:nowrap}.chart-axis{margin-left:255px;display:flex;justify-content:space-between;font:10px var(--mono);color:#899087;padding-bottom:12px}.bar-row{display:grid;grid-template-columns:235px 1fr 50px;gap:20px;align-items:center;margin-bottom:16px}.bar-label{font-size:12px;color:var(--muted)}.bar-track{height:22px;background:#f3f5f0;border-radius:3px;overflow:hidden}.bar-fill{height:100%;background:#c5d1c1;border-radius:3px}.bar-row.featured .bar-fill{background:var(--green)}.bar-row.featured .bar-label,.bar-row.featured .bar-value{color:var(--ink);font-weight:650}.bar-value{font:12px var(--mono);color:var(--muted)}.axis-label{font-size:10px;color:#7e877c;text-align:right;margin:2px 0 21px}.result-foot{padding-top:16px;border-top:1px solid var(--line);font-size:11px;color:var(--muted);line-height:1.8}.result-foot a{color:var(--green)}.insight{margin-top:24px;display:flex;gap:17px;align-items:baseline}.insight strong{font-size:25px;letter-spacing:-.04em;color:var(--green);white-space:nowrap}.insight p{font-size:13px;margin:0}.abstract{border-top:1px solid var(--line);padding-top:38px;display:grid;grid-template-columns:.45fr 1fr;gap:60px}.abstract h2{font-size:26px;font-weight:550}.abstract p{font-size:14px;line-height:1.95;margin:0}.citation{background:var(--light);padding:30px;border-radius:10px;border:1px solid #d9e5d6}.citation-top{display:flex;align-items:center;justify-content:space-between;gap:16px;margin-bottom:20px}.citation h2{font-size:24px;margin:0;font-weight:550}.citation pre{font:11px/1.9 var(--mono);color:#43594b;white-space:pre-wrap;overflow-wrap:anywhere;margin:0}.copy-button{font:11px var(--mono);border:1px solid #c3d3be;border-radius:5px;padding:7px 11px;background:var(--paper);color:var(--ink);cursor:pointer}.copy-status{font-size:12px;min-height:20px;color:var(--green);margin:8px 0 0}.footer{border-top:1px solid var(--line);padding:23px 0 30px}.footer-inner{display:flex;justify-content:space-between;gap:20px;align-items:center;font-size:11px;color:var(--muted)}.footer-links{display:flex;gap:20px}[hidden]{display:none!important}.sr-only{position:absolute;width:1px;height:1px;overflow:hidden;clip:rect(0,0,0,0)}
15
  @media(min-width:1450px){.wrap{max-width:1260px}.hero{padding-top:78px}.hero h1{font-size:43px}}
16
- @media(max-width:850px){.wrap{padding:0 25px}.hero-top{gap:25px}.hero h1{font-size:29px}.wordmark{font-size:85px}.hero-art{padding:20px 14px}.art-label{font-size:8px}.section-head{gap:25px;grid-template-columns:.8fr 1fr}.metric{padding:0 18px}.metric-value{font-size:38px}.section-head h2{font-size:29px}.bar-row{grid-template-columns:185px 1fr 42px;gap:12px}.chart-axis{margin-left:197px}.abstract{gap:35px}}
17
  @media(max-width:620px){.wrap{padding:0 21px}.nav-inner{height:65px}.nav-links{gap:17px}.nav-links a{font-size:11px}.nav-links .nav-cite{display:none}.hero{padding-top:35px}.hero-top{grid-template-columns:1fr;gap:32px}.wordmark{font-size:88px;margin-top:20px}.hero h1{font-size:33px;max-width:430px}.hero-copy{font-size:15px}.hero-art{padding:24px 20px}.art-label{font-size:9px}.authors{margin-top:28px;font-size:12px}.author-list{gap:4px 16px}.metrics{padding:22px 0;margin-bottom:45px;gap:16px;grid-template-columns:1fr}.metric,.metric:first-child{border:0;padding:0;display:grid;grid-template-columns:1fr auto;column-gap:12px}.metric-label{margin-bottom:0;align-self:end}.metric-value{grid-column:2;grid-row:1 / span 2;font-size:34px;align-self:center}.metric-value span{font-size:22px}.metric-note{margin-top:3px}.section{padding-bottom:45px}.section-head{grid-template-columns:1fr;gap:18px;margin-bottom:25px}.section-head h2{font-size:29px}.section-head p{font-size:14px}.steps{grid-template-columns:1fr;gap:23px}.method-figure{padding:10px;overflow:hidden}.method-figure img{min-width:680px}.method-scroll{overflow:auto}.method-figure figcaption{font-size:10px;margin-top:12px}.tabs{padding:9px 8px 0;gap:0}.tab{font-size:10px;padding:10px 9px 13px}.result-content{padding:22px 17px}.result-topline{display:block}.result-topline h3{font-size:19px}.pill{display:inline-block;margin-top:12px}.bar-row{grid-template-columns:1fr 40px;gap:7px 10px;margin-bottom:18px}.bar-label{grid-column:1;font-size:11px}.bar-track{grid-column:1;grid-row:2;height:18px}.bar-value{grid-column:2;grid-row:2;font-size:11px}.chart-axis{margin-left:0;padding-right:50px}.insight{gap:12px;align-items:start}.insight strong{font-size:23px}.insight p{font-size:12px}.abstract{grid-template-columns:1fr;gap:15px;padding-top:28px}.abstract p{font-size:13px}.citation{padding:22px 18px}.citation pre{font-size:10px}.footer-inner{align-items:start;flex-direction:column;gap:10px}.footer-links{gap:17px}}
18
  @media(prefers-reduced-motion:reduce){html{scroll-behavior:auto}.button{transition:none}}
19
  </style>
@@ -61,13 +61,13 @@
61
  <div class="tabs" role="tablist" aria-label="Benchmark results"><button class="tab" id="tab-wai" role="tab" aria-selected="true" aria-controls="panel-wai" tabindex="0">WebArena Infinity</button><button class="tab" id="tab-vwa" role="tab" aria-selected="false" aria-controls="panel-vwa" tabindex="-1">VisualWebArena</button><button class="tab" id="tab-om2w" role="tab" aria-selected="false" aria-controls="panel-om2w" tabindex="-1">Online Mind2Web</button></div>
62
  <div class="result-content" id="panel-wai" role="tabpanel" aria-labelledby="tab-wai">
63
  <div class="result-topline"><div><h3>Stronger training on an open backbone</h3><p>Task-weighted success rate on the public 9-app comparison set. CLIFT trains Gemma-4 31B with LoRA; CTS adds inference-time selection.</p></div><span class="pill">OPEN MODEL · TRAIN + SELECT</span></div>
64
- <div class="chart" role="img" aria-label="WAI success rates: Gemma-4 base 61.8%, vanilla GRPO 64.7%, CLIFT 72.6%, CLIFT plus CTS K=3 74.6%.">
65
- <div class="chart-axis" aria-hidden="true"><span>0</span><span>50</span><span>100%</span></div>
66
- <div class="bar-row"><div class="bar-label">Gemma-4 base</div><div class="bar-track"><div class="bar-fill" style="width:61.8%"></div></div><div class="bar-value">61.8</div></div>
67
- <div class="bar-row"><div class="bar-label">Vanilla GRPO</div><div class="bar-track"><div class="bar-fill" style="width:64.7%"></div></div><div class="bar-value">64.7</div></div>
68
- <div class="bar-row"><div class="bar-label">CLIFT</div><div class="bar-track"><div class="bar-fill" style="width:72.6%"></div></div><div class="bar-value">72.6</div></div>
69
- <div class="bar-row featured"><div class="bar-label">CLIFT + CTS (K = 3)</div><div class="bar-track"><div class="bar-fill" style="width:74.6%"></div></div><div class="bar-value">74.6</div></div>
70
- </div><div class="axis-label">Success rate (%) · higher is better</div>
71
  <div class="result-foot">Source: <a href="https://arxiv.org/html/2610.06829v1#S4.T6" target="_blank" rel="noopener">arXiv v1, Table 6</a>. Training rows share the comparative judge and browser-use backbone. Training contributes +10.8 pp; CTS adds +2.0 pp.</div>
72
  </div>
73
  <div class="result-content" id="panel-vwa" role="tabpanel" aria-labelledby="tab-vwa" hidden>
 
11
  <link rel="icon" href="data:image/svg+xml,%3Csvg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 64 64'%3E%3Crect width='64' height='64' rx='16' fill='%23173d34'/%3E%3Cpath d='M18 33l9 9 20-22' fill='none' stroke='%23c9f2cf' stroke-width='6'/%3E%3C/svg%3E">
12
  <style>
13
  :root{--ink:#17352f;--muted:#5b6964;--green:#287655;--light:#edf4ed;--paper:#fbfcf8;--line:#dce4db;--mono:ui-monospace,SFMono-Regular,Consolas,monospace;--sans:Inter,-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif}
14
+ *{box-sizing:border-box}html{scroll-behavior:smooth;scroll-padding-top:80px}body{margin:0;background:var(--paper);color:var(--ink);font-family:var(--sans);font-size:16px;line-height:1.65}a{color:inherit;text-underline-offset:4px}button{font:inherit}button,a{-webkit-tap-highlight-color:transparent}a:focus-visible,button:focus-visible,summary:focus-visible{outline:3px solid #399365;outline-offset:5px}svg{display:block}p,h1,h2,h3{margin-top:0}h1,h2,h3{line-height:1.15;letter-spacing:-.045em}p{color:var(--muted)}.wrap{max-width:1180px;margin:auto;padding:0 40px}.nav{border-bottom:1px solid var(--line);background:var(--paper)}.nav-inner{height:76px;display:flex;align-items:center;justify-content:space-between;gap:24px}.brand{display:flex;align-items:center;gap:10px;font-size:20px;font-weight:750;letter-spacing:-.04em;text-decoration:none}.brand-mark{display:grid;place-items:center;width:30px;height:30px;background:var(--ink);border-radius:8px;color:#d8f1d4}.nav-links{display:flex;align-items:center;gap:28px}.nav-links a{text-decoration:none;font-size:13px;color:var(--muted)}.nav-links a:hover{color:var(--green)}.external{font-family:var(--mono)}.eyebrow{display:flex;align-items:center;gap:9px;color:var(--green);font-family:var(--mono);font-size:11px;text-transform:uppercase;letter-spacing:.13em}.dot{height:6px;width:6px;border-radius:50%;background:var(--green)}.hero{padding:66px 0 45px}.hero-top{display:grid;grid-template-columns:1.08fr 1fr;gap:52px;align-items:center}.wordmark{font-family:Georgia,"Times New Roman",serif;font-size:100px;font-weight:400;letter-spacing:-.07em;line-height:1;margin:24px 0 16px}.hero h1{font-size:39px;line-height:1.14;max-width:570px;font-weight:550;margin-bottom:21px}.hero-copy{font-size:17px;line-height:1.7;max-width:480px}.hero-art{padding:28px 20px 18px;background:var(--light);border:1px solid #d8e4d7;border-radius:18px;position:relative}.art-label{font:10px var(--mono);text-transform:uppercase;letter-spacing:.13em;color:#64816c;margin-bottom:25px;display:flex;justify-content:space-between}.hero-art img{display:block;width:100%;height:auto;mix-blend-mode:multiply}.art-caption{padding-top:20px;border-top:1px solid #d4e1d1;margin-top:22px;display:flex;justify-content:space-between;gap:10px;color:#3b634d;font-size:11px}.resources{display:flex;gap:10px;flex-wrap:wrap;margin-top:28px}.button{display:inline-flex;align-items:center;gap:12px;justify-content:center;padding:10px 18px;border:1px solid var(--line);border-radius:7px;background:transparent;font-size:13px;font-weight:600;text-decoration:none;cursor:pointer;transition:background .15s}.button:hover{background:#e6ede3}.button.primary{background:var(--ink);border-color:var(--ink);color:#fff}.button.primary:hover{background:#285044}.authors{margin-top:38px;padding-top:26px;border-top:1px solid var(--line);font-size:13px;color:#42564b;line-height:1.9}.author-list{display:flex;gap:6px 21px;flex-wrap:wrap}.affiliation{font-family:var(--mono);font-size:11px;letter-spacing:.03em;margin-top:6px;color:#7b857b}.metrics{display:grid;grid-template-columns:repeat(3,1fr);border-top:1px solid var(--line);border-bottom:1px solid var(--line);padding:29px 0 27px;margin:0 0 68px}.metric{padding:0 30px;border-left:1px solid var(--line)}.metric:first-child{padding-left:0;border-left:0}.metric-label{font-size:12px;color:var(--muted);margin-bottom:9px}.metric-value{font-size:45px;letter-spacing:-.06em;line-height:1.05;font-weight:500}.metric-value span{font-size:25px}.metric-note{font-size:11px;color:var(--muted);margin-top:9px}.gain{color:var(--green);font-weight:650}.section{padding:0 0 72px}.section-head{display:grid;grid-template-columns:.67fr 1fr;gap:60px;margin-bottom:32px;align-items:start}.section-number{font:11px var(--mono);letter-spacing:.08em;color:var(--green);margin-bottom:14px}.section-head h2{font-size:34px;font-weight:550;margin-bottom:0}.section-head p{font-size:15px;margin-bottom:0;line-height:1.8}.method-figure{margin:0;border:1px solid var(--line);border-radius:10px;background:#fff;padding:22px}.method-figure img{width:100%;height:auto;display:block}.method-figure figcaption{font-size:11px;line-height:1.6;color:var(--muted);border-top:1px solid var(--line);padding-top:15px;margin-top:18px}.steps{display:grid;grid-template-columns:repeat(3,1fr);gap:30px;margin-top:27px}.step-top{display:flex;align-items:center;gap:10px;margin-bottom:10px}.step-n{font:11px var(--mono);color:var(--green);border:1px solid #c9d9c8;border-radius:50%;height:25px;width:25px;display:grid;place-items:center;flex:none}.step h3{font-size:17px;margin:0;font-weight:600;letter-spacing:-.025em}.step p{font-size:13px;line-height:1.8;margin:0}.results-panel{border:1px solid var(--line);border-radius:10px;background:white;overflow:hidden}.tabs{padding:15px 18px 0;display:flex;gap:8px;border-bottom:1px solid var(--line);background:#f5f7f1;overflow-x:auto}.tab{white-space:nowrap;border:0;border-bottom:3px solid transparent;padding:11px 13px 14px;cursor:pointer;font-size:12px;color:var(--muted);background:transparent}.tab[aria-selected="true"]{border-bottom-color:var(--green);color:var(--ink);font-weight:650}.result-content{padding:28px 30px 25px}.result-topline{display:flex;align-items:start;justify-content:space-between;gap:25px;margin-bottom:24px}.result-topline h3{font-size:21px;letter-spacing:-.025em;margin-bottom:7px}.result-topline p{font-size:12px;max-width:650px;margin-bottom:0}.pill{font:10px var(--mono);background:var(--light);border-radius:5px;color:var(--green);padding:6px 9px;white-space:nowrap}.chart-axis{margin-left:255px;padding-right:70px;display:flex;justify-content:space-between;font:10px var(--mono);color:#899087;padding-bottom:12px}.bar-row{display:grid;grid-template-columns:235px 1fr 50px;gap:20px;align-items:center;margin-bottom:16px}.bar-label{font-size:12px;color:var(--muted)}.bar-track{height:22px;background:#f3f5f0;border-radius:3px;overflow:hidden}.bar-fill{height:100%;background:#c5d1c1;border-radius:3px}.bar-row.featured .bar-fill{background:var(--green)}.bar-row.featured .bar-label,.bar-row.featured .bar-value{color:var(--ink);font-weight:650}.bar-value{font:12px var(--mono);color:var(--muted)}.axis-label{font-size:10px;color:#7e877c;text-align:right;margin:2px 0 21px}.result-foot{padding-top:16px;border-top:1px solid var(--line);font-size:11px;color:var(--muted);line-height:1.8}.result-foot a{color:var(--green)}.insight{margin-top:24px;display:flex;gap:17px;align-items:baseline}.insight strong{font-size:25px;letter-spacing:-.04em;color:var(--green);white-space:nowrap}.insight p{font-size:13px;margin:0}.abstract{border-top:1px solid var(--line);padding-top:38px;display:grid;grid-template-columns:.45fr 1fr;gap:60px}.abstract h2{font-size:26px;font-weight:550}.abstract p{font-size:14px;line-height:1.95;margin:0}.citation{background:var(--light);padding:30px;border-radius:10px;border:1px solid #d9e5d6}.citation-top{display:flex;align-items:center;justify-content:space-between;gap:16px;margin-bottom:20px}.citation h2{font-size:24px;margin:0;font-weight:550}.citation pre{font:11px/1.9 var(--mono);color:#43594b;white-space:pre-wrap;overflow-wrap:anywhere;margin:0}.copy-button{font:11px var(--mono);border:1px solid #c3d3be;border-radius:5px;padding:7px 11px;background:var(--paper);color:var(--ink);cursor:pointer}.copy-status{font-size:12px;min-height:20px;color:var(--green);margin:8px 0 0}.footer{border-top:1px solid var(--line);padding:23px 0 30px}.footer-inner{display:flex;justify-content:space-between;gap:20px;align-items:center;font-size:11px;color:var(--muted)}.footer-links{display:flex;gap:20px}[hidden]{display:none!important}.sr-only{position:absolute;width:1px;height:1px;overflow:hidden;clip:rect(0,0,0,0)}
15
  @media(min-width:1450px){.wrap{max-width:1260px}.hero{padding-top:78px}.hero h1{font-size:43px}}
16
+ @media(max-width:850px){.wrap{padding:0 25px}.hero-top{gap:25px}.hero h1{font-size:29px}.wordmark{font-size:85px}.hero-art{padding:20px 14px}.art-label{font-size:8px}.section-head{gap:25px;grid-template-columns:.8fr 1fr}.metric{padding:0 18px}.metric-value{font-size:38px}.section-head h2{font-size:29px}.bar-row{grid-template-columns:185px 1fr 42px;gap:12px}.chart-axis{margin-left:197px;padding-right:54px}.abstract{gap:35px}}
17
  @media(max-width:620px){.wrap{padding:0 21px}.nav-inner{height:65px}.nav-links{gap:17px}.nav-links a{font-size:11px}.nav-links .nav-cite{display:none}.hero{padding-top:35px}.hero-top{grid-template-columns:1fr;gap:32px}.wordmark{font-size:88px;margin-top:20px}.hero h1{font-size:33px;max-width:430px}.hero-copy{font-size:15px}.hero-art{padding:24px 20px}.art-label{font-size:9px}.authors{margin-top:28px;font-size:12px}.author-list{gap:4px 16px}.metrics{padding:22px 0;margin-bottom:45px;gap:16px;grid-template-columns:1fr}.metric,.metric:first-child{border:0;padding:0;display:grid;grid-template-columns:1fr auto;column-gap:12px}.metric-label{margin-bottom:0;align-self:end}.metric-value{grid-column:2;grid-row:1 / span 2;font-size:34px;align-self:center}.metric-value span{font-size:22px}.metric-note{margin-top:3px}.section{padding-bottom:45px}.section-head{grid-template-columns:1fr;gap:18px;margin-bottom:25px}.section-head h2{font-size:29px}.section-head p{font-size:14px}.steps{grid-template-columns:1fr;gap:23px}.method-figure{padding:10px;overflow:hidden}.method-figure img{min-width:680px}.method-scroll{overflow:auto}.method-figure figcaption{font-size:10px;margin-top:12px}.tabs{padding:9px 8px 0;gap:0}.tab{font-size:10px;padding:10px 9px 13px}.result-content{padding:22px 17px}.result-topline{display:block}.result-topline h3{font-size:19px}.pill{display:inline-block;margin-top:12px}.bar-row{grid-template-columns:1fr 40px;gap:7px 10px;margin-bottom:18px}.bar-label{grid-column:1;font-size:11px}.bar-track{grid-column:1;grid-row:2;height:18px}.bar-value{grid-column:2;grid-row:2;font-size:11px}.chart-axis{margin-left:0;padding-right:50px}.insight{gap:12px;align-items:start}.insight strong{font-size:23px}.insight p{font-size:12px}.abstract{grid-template-columns:1fr;gap:15px;padding-top:28px}.abstract p{font-size:13px}.citation{padding:22px 18px}.citation pre{font-size:10px}.footer-inner{align-items:start;flex-direction:column;gap:10px}.footer-links{gap:17px}}
18
  @media(prefers-reduced-motion:reduce){html{scroll-behavior:auto}.button{transition:none}}
19
  </style>
 
61
  <div class="tabs" role="tablist" aria-label="Benchmark results"><button class="tab" id="tab-wai" role="tab" aria-selected="true" aria-controls="panel-wai" tabindex="0">WebArena Infinity</button><button class="tab" id="tab-vwa" role="tab" aria-selected="false" aria-controls="panel-vwa" tabindex="-1">VisualWebArena</button><button class="tab" id="tab-om2w" role="tab" aria-selected="false" aria-controls="panel-om2w" tabindex="-1">Online Mind2Web</button></div>
62
  <div class="result-content" id="panel-wai" role="tabpanel" aria-labelledby="tab-wai">
63
  <div class="result-topline"><div><h3>Stronger training on an open backbone</h3><p>Task-weighted success rate on the public 9-app comparison set. CLIFT trains Gemma-4 31B with LoRA; CTS adds inference-time selection.</p></div><span class="pill">OPEN MODEL · TRAIN + SELECT</span></div>
64
+ <div class="chart" role="img" aria-label="WAI success rates on a 50–100% axis: Gemma-4 base 61.8%, vanilla GRPO 64.7%, CLIFT 72.6%, CLIFT plus CTS K=3 74.6%.">
65
+ <div class="chart-axis" aria-hidden="true"><span>50%</span><span>75%</span><span>100%</span></div>
66
+ <div class="bar-row"><div class="bar-label">Gemma-4 base</div><div class="bar-track"><div class="bar-fill" style="width:23.6%"></div></div><div class="bar-value">61.8</div></div>
67
+ <div class="bar-row"><div class="bar-label">Vanilla GRPO</div><div class="bar-track"><div class="bar-fill" style="width:29.4%"></div></div><div class="bar-value">64.7</div></div>
68
+ <div class="bar-row"><div class="bar-label">CLIFT</div><div class="bar-track"><div class="bar-fill" style="width:45.2%"></div></div><div class="bar-value">72.6</div></div>
69
+ <div class="bar-row featured"><div class="bar-label">CLIFT + CTS (K = 3)</div><div class="bar-track"><div class="bar-fill" style="width:49.2%"></div></div><div class="bar-value">74.6</div></div>
70
+ </div><div class="axis-label">Success rate (%) · axis starts at 50% · higher is better</div>
71
  <div class="result-foot">Source: <a href="https://arxiv.org/html/2610.06829v1#S4.T6" target="_blank" rel="noopener">arXiv v1, Table 6</a>. Training rows share the comparative judge and browser-use backbone. Training contributes +10.8 pp; CTS adds +2.0 pp.</div>
72
  </div>
73
  <div class="result-content" id="panel-vwa" role="tabpanel" aria-labelledby="tab-vwa" hidden>