-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
693 lines (651 loc) · 29.9 KB
/
Copy pathindex.html
File metadata and controls
693 lines (651 loc) · 29.9 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>WebHarbor: Docking Real Websites for Evolving GUI Agent Environments</title>
<meta name="description" content="WebHarbor docks popular websites into local, stable, Docker-based mirrors with full auth, database, and multimodal content. Environments evolve with agent capability. Starting with 15 sites, scaling to 100." />
<link rel="icon" type="image/x-icon" href="assets/icons/favicon.ico" />
<link rel="icon" type="image/png" sizes="32x32" href="assets/icons/favicon-32x32.png" />
<link rel="icon" type="image/png" sizes="16x16" href="assets/icons/favicon-16x16.png" />
<link rel="apple-touch-icon" sizes="180x180" href="assets/icons/apple-touch-icon.png" />
<meta property="og:image" content="assets/icons/og-image.jpg" />
<meta property="og:title" content="WebHarbor: Docking Real Websites for Evolving GUI Agent Environments" />
<meta property="og:type" content="website" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link href="https://fonts.googleapis.com/css2?family=DM+Serif+Display:ital@0;1&family=Instrument+Sans:ital,wght@0,400;0,500;0,600;0,700;1,400&family=IBM+Plex+Mono:wght@400;500;600&display=swap" rel="stylesheet" />
<link rel="stylesheet" href="assets/style.css" />
<script src="https://cdn.plot.ly/plotly-2.35.2.min.js" defer></script>
</head>
<body>
<header class="site-nav">
<div class="container nav-inner">
<a href="#top" class="brand">
<span class="brand-mark">⚓</span>
<span class="brand-name">WebHarbor</span>
</a>
<nav>
<a href="#gallery">Live Demos</a>
<a href="#motivation">Motivation</a>
<a href="#evolving">Our Approach</a>
<a href="#validation">Validation</a>
<a href="#quickstart">Quickstart</a>
<a href="#contribute">Contribute</a>
</nav>
<div class="nav-cta-group">
<a class="nav-cta nav-cta-ghost" href="guide-create.html">Contribution Guide</a>
<a class="nav-cta nav-cta-ghost" href="guide-review.html">Review Guide</a>
</div>
</div>
</header>
<main id="top">
<!-- =================== HERO =================== -->
<section class="hero">
<div class="container hero-inner">
<div class="eyebrow">
<span class="dot"></span>
Evolving GUI Web Agent Environments
</div>
<h1 class="hero-title">
WebHarbor: Docking Real Websites for<br />
<span class="accent">Evolving GUI Agent Environments</span>
</h1>
<p class="hero-sub">
Live websites are often noisy, blocked by reCAPTCHA, and hide their most useful
features behind login walls. WebHarbor <strong>docks</strong> them into
local Docker mirrors: stable, reproducible, resettable in
milliseconds, and <strong>evolves</strong> environments with more functionalities and data as agents grow stronger.
</p>
<ul class="hero-highlights">
<li><strong>Stable & reproducible</strong> no network noise, no content drift, no geo-blocks</li>
<li><strong>Deep features unlocked</strong> carts, checkouts, and accounts, all testable</li>
<li><strong>Evolving</strong> harder tasks drive richer mirrors, environment evolves</li>
<li><strong>RL-ready</strong> sub-second database resets for lightweight environments</li>
<li><strong>Community-driven</strong> 15 sites today, let's scale to 100 and beyond, together</li>
</ul>
<div class="hero-quickstart">
<p class="hero-qs-label">Start deploying 15 <a href="https://github.com/MinorJerry/WebVoyager/blob/main/data/WebVoyager_data.jsonl" target="_blank" rel="noopener">WebVoyager</a> sites locally with one command:</p>
<div class="hero-qs-cmd">
<code>docker run -p 40000-40014:40000-40014 battalion7244/webharbor:latest</code>
<button class="copy-btn" data-copy="docker run -p 40000-40014:40000-40014 battalion7244/webharbor:latest">Copy</button>
</div>
<div class="hero-qs-badge-row">
<a href="#validation" class="hero-qs-badge">
<span class="hero-qs-badge-mark">✓</span>
Validated — model rankings on the docked WebVoyager sites match live-web benchmarks
<span class="hero-qs-badge-arrow">→</span>
</a>
</div>
</div>
<div class="hero-ctas">
<a class="btn btn-primary" href="https://github.com/aiming-lab/WebHarbor" target="_blank" rel="noopener">
<span>💻 Code on GitHub</span><span class="arr">↗</span>
</a>
<a class="btn btn-ghost" href="https://huggingface.co/datasets/ChilleD/WebHarbor" target="_blank" rel="noopener">🤗 Dataset</a>
<a class="btn btn-ghost" href="https://docs.google.com/spreadsheets/d/1vZsrQjy9nJKze58fx4kbQtFi85NjVXIWCFyu3ShD7gk/edit?gid=0#gid=0" target="_blank" rel="noopener">📊 Track Sheet</a>
<a class="btn btn-ghost" href="https://forms.gle/ngcD1rzAfUEphNmRA" target="_blank" rel="noopener">📝 Request Form</a>
</div>
<div class="hero-strip" id="hero-strip">
<div class="strip-track"></div>
</div>
</div>
</section>
<!-- =================== SECTION OUTLINE =================== -->
<section id="outline" class="section outline">
<div class="container">
<div class="section-head">
<span class="tag">Contents</span>
<h2>Page outline</h2>
</div>
<ol class="outline-list">
<li>
<a href="#gallery">
<span class="ol-num">01</span>
<span class="ol-body">
<span class="ol-title">15 live demos</span>
<span class="ol-desc">15 live websites running in Docker containers.</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
<li>
<a href="#motivation">
<span class="ol-num">02</span>
<span class="ol-body">
<span class="ol-title">Motivation</span>
<span class="ol-desc">Why current environments (live, offline, or toy) can't evaluate the real web.</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
<li>
<a href="#evolving">
<span class="ol-num">03</span>
<span class="ol-body">
<span class="ol-title">Evolving Environments</span>
<span class="ol-desc">Coding agents build, humans verify, tasks drive environment evolution.</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
<li>
<a href="#validation">
<span class="ol-num">04</span>
<span class="ol-body">
<span class="ol-title">Validation</span>
<span class="ol-desc">Our simulated benchmark preserves the model ranking observed on live-web benchmarks.</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
<li>
<a href="#quickstart">
<span class="ol-num">05</span>
<span class="ol-body">
<span class="ol-title">Quickstart</span>
<span class="ol-desc">One command to deploy live website environments, locally.</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
<li>
<a href="#contribute">
<span class="ol-num">06</span>
<span class="ol-body">
<span class="ol-title">Contribution</span>
<span class="ol-desc">Scale with community power. Contribute now!</span>
</span>
<span class="ol-arrow">↓</span>
</a>
</li>
</ol>
</div>
</section>
<!-- =================== GALLERY (15 LIVE DEMOS) =================== -->
<section id="gallery" class="section bordered">
<div class="container">
<div class="section-head">
<span class="tag">01 · 15 live demos</span>
<h2>Real websites, running locally, with every feature unlocked.</h2>
<p class="section-lede">
Each clip below is a <em>real interaction</em> captured against a
running WebHarbor mirror: shopping carts, flight search,
map directions, issue tracking, meal planning, and scoreboard stats.
These are the deep, login-gated features that live websites can never reach.
</p>
</div>
<div class="gallery" id="gallery-grid">
<!-- populated from JS -->
</div>
</div>
</section>
<!-- =================== MOTIVATION =================== -->
<section id="motivation" class="section bordered">
<div class="container">
<div class="section-head">
<span class="tag">02 · Motivation</span>
<h2>The web is multimodal, deep, and gated. Current benchmarks and environments are none of these.</h2>
<p class="section-lede">
Real websites have rich visual layouts, product images, interactive
maps, and deeply nested functionality behind login walls. Existing
web agent benchmarks either fight the live web’s instability or
retreat into toy environments that don’t resemble it.
</p>
</div>
<div class="problem-grid">
<article class="problem-card">
<div class="problem-head">
<span class="prob-id">a</span>
<h3>Live-site eval</h3>
<span class="prob-sub">WebVoyager, Online-Mind2Web</span>
</div>
<p>
reCAPTCHA, geo-blocks, network flakiness, and content drift inflate
eval noise. Multiple WebVoyager sites are effectively inaccessible.
Tasks never require login, so coverage is limited to surface-level
fact lookups.
</p>
</article>
<article class="problem-card">
<div class="problem-head">
<span class="prob-id">b</span>
<h3>Offline traces</h3>
<span class="prob-sub">Mind2Web</span>
</div>
<p>
No interactive environment, no state, no consequence of a wrong
click. Useful for supervised pre-training but fundamentally
unsuited to evaluating agents that must act, not just predict.
</p>
</article>
<article class="problem-card">
<div class="problem-head">
<span class="prob-id">c</span>
<h3>Synthetic Environments</h3>
<span class="prob-sub">WebArena</span>
</div>
<p>
Some synthetic browser environments built from coloured rectangles instead of real images; a
handful of sites with no multimodal richness. WebArena is
content-rich but heavyweight, and slow resets make RL-scale
rollouts impractical.
</p>
</article>
</div>
<div class="thesis">
<span class="thesis-kicker">Core insight</span>
<p>
The bottleneck isn’t the agent. It’s
<strong>the environment</strong>. We need web environments that are
<em>stable</em> (no eval noise), <em>multimodal</em> (real images
and layouts), <em>deep</em> (auth-gated flows), and
<em>evolving</em> (growing with agent capability).
</p>
</div>
<!-- comparison table -->
<div class="compare-wrap">
<h3 class="compare-title">How WebHarbor compares</h3>
<div class="compare-table-scroll">
<table class="compare-table">
<thead>
<tr>
<th class="compare-dim"></th>
<th>WebArena</th>
<th>WebVoyager</th>
<th>Online-Mind2Web</th>
<th class="compare-ours">WebHarbor</th>
</tr>
</thead>
<tbody>
<tr>
<td class="compare-dim">Environment</td>
<td>Self-hosted (synthetic)</td>
<td>Live web</td>
<td>Live web</td>
<td class="compare-ours">Self-hosted (synthetic)</td>
</tr>
<tr>
<td class="compare-dim"># of Sites</td>
<td>5</td>
<td>15</td>
<td>100+</td>
<td class="compare-ours">15 (100+ planned)</td>
</tr>
<tr>
<td class="compare-dim">Eval Stability</td>
<td><span class="cmp-ok">Stable</span></td>
<td><span class="cmp-bad">Noisy</span> (reCAPTCHA, geo-blocks, tasks outdated)</td>
<td><span class="cmp-bad">Noisy</span> (reCAPTCHA, geo-blocks)</td>
<td><span class="cmp-ok">Stable</span></td>
</tr>
<tr>
<td class="compare-dim">Login & Deep Features</td>
<td class="compare-ours"><span class="cmp-good">Fully unlocked</span></td>
<td><span class="cmp-bad">Not supported</span></td>
<td><span class="cmp-bad">Not supported</span></td>
<td class="compare-ours"><span class="cmp-good">Fully unlocked</span></td>
</tr>
<tr>
<td class="compare-dim">Reset Speed</td>
<td><span class="cmp-bad">Slow</span> (Docker restart)</td>
<td>N/A (live)</td>
<td>N/A (live)</td>
<td class="compare-ours"><span class="cmp-good">Fast</span> (SQLite reset)</td>
</tr>
<tr>
<td class="compare-dim">Evolution</td>
<td>Static, frozen</td>
<td><span class="cmp-bad">Tasks outdated</span></td>
<td><span class="cmp-ok">Tasks updated</span></td>
<td class="compare-ours"><span class="cmp-good">Evolving</span>env grows with tasks</td>
</tr>
<tr>
<td class="compare-dim">RL Compatible</td>
<td><span class="cmp-bad">Impractical</span> (slow reset)</td>
<td><span class="cmp-bad">No</span> (noisy, live)</td>
<td><span class="cmp-bad">No</span> (noisy, live)</td>
<td class="compare-ours"><span class="cmp-good">Yes</span> (fast reset, deterministic)</td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</section>
<!-- =================== EVOLVING ENVIRONMENTS (OUR APPROACH) =================== -->
<section id="evolving" class="section bordered">
<div class="container">
<div class="section-head">
<span class="tag">03 · Evolving Environments: Our Approach</span>
<h2>Coding agents build. Humans verify. Tasks drive evolution.</h2>
<p class="section-lede">
WebHarbor environments are built by coding agents (e.g. Claude Code with
specialized skills), then rigorously reviewed by human annotators to
ensure multimodal fidelity and functional depth. Tasks from existing
benchmarks (starting with
<a href="https://github.com/MinorJerry/WebVoyager" target="_blank" rel="noopener">WebVoyager</a>)
define what the environment support and look like.
</p>
</div>
<!-- build pipeline: 4 steps -->
<div class="evo-flow">
<div class="evo-step">
<div class="evo-num">1</div>
<h4>Coding agent builds</h4>
<p>A coding agent with specialized skills scrapes the real site's structure, assets, and catalog, then generates a full-stack mirror — SQLite database, REST backend, and frontend — with auth and CRUD flows.</p>
</div>
<div class="evo-arrow">→</div>
<div class="evo-step">
<div class="evo-num">2</div>
<h4>Human-in-the-loop review</h4>
<p>Human reviewers verify every mirror for visual fidelity, functional correctness, and data quality. This step is essential: we observe that agents frequently take shortcuts, produce superficially correct but broken pages, or skip edge cases.</p>
</div>
<div class="evo-arrow">→</div>
<div class="evo-step">
<div class="evo-num">3</div>
<h4>Task-driven scoping</h4>
<p>Cloning an entire live website is impractical. Instead, tasks define what features the mirror must support. Every task is verified to run against the environment with grounded truth checks.</p>
</div>
<div class="evo-arrow">→</div>
<div class="evo-step">
<div class="evo-num">4</div>
<h4>Environment evolves</h4>
<p>Once agents master existing tasks, we can have harder ones that require deeper site features. The coding agent extends the mirror on demand. The environment is never "done"; it grows alongside agent capability.</p>
</div>
</div>
<!-- why humans matter -->
<div class="thesis" style="margin-top:46px;">
<span class="thesis-kicker">Why humans are essential</span>
<p>
Building high-quality, multimodal web environments cannot be fully
automated, currently. We observe that coding agents frequently
<strong>hack</strong> their way through: using placeholder images,
skipping complex layouts, producing pages that pass automated checks
but fail under real interaction. Human review catches these issues
and ensures every mirror matches the look, feel, and depth of the
real site.
</p>
</div>
<!-- why evolving -->
<div class="thesis" style="margin-top:20px;">
<span class="thesis-kicker">Why evolving</span>
<p>
A live website like Amazon has millions of products, thousands of routes,
and decades of accumulated functionality. Cloning it entirely is neither
tractable nor necessary. But we can ensure every task has
full environment support. When agents master those tasks, we expand:
harder tasks → more features, richer data, deeper flows. The
environment <strong>evolves</strong> with the agent.
</p>
</div>
</div>
</section>
<!-- =================== VISION =================== -->
<div class="container hinge-inner">
<p class="hinge-q">Our vision</p>
<p class="hinge-a">
Leverage code to <span class="accent">mirror the world</span>.
Cultivate the agent.
Prepare it to interact with and learn from <span class="accent">the real world itself</span>.
</p>
</div>
<!-- =================== VALIDATION =================== -->
<section id="validation" class="section bordered">
<div class="container">
<div class="section-head">
<span class="tag">04 · Validation</span>
<h2>WebHarbor-WebVoyager preserves the model ranking observed on live-web benchmarks.</h2>
<p class="section-lede">
Our first release docks the 15
<a href="https://github.com/MinorJerry/WebVoyager" target="_blank" rel="noopener">WebVoyager</a>
sites. To verify the simulated environment behaves as a faithful proxy for the
real web, we evaluated three vision-language web agents on
<strong>WebHarbor-WebVoyager</strong> and on three independent live-web benchmarks.
The relative ordering
<a href="https://huggingface.co/Qwen/Qwen3-VL-4B-Thinking" target="_blank" rel="noopener">Qwen3-VL-4B-Thinking</a> < <a href="https://huggingface.co/Qwen/Qwen3-VL-235B-A22B-Thinking" target="_blank" rel="noopener">Qwen3-VL-235B-A22B-Thinking</a> < <a href="https://arxiv.org/pdf/2605.15040" target="_blank" rel="noopener">Orchard-GUI-4B</a>
is preserved on every benchmark — confirming <strong>WebHarbor ranks models
consistently with prior evaluations.</strong>
</p>
</div>
<div class="validation-card">
<div id="validation-chart"></div>
</div>
<p class="validation-caption">
Higher score is better. The
<span class="validation-pill">WebHarbor-WebVoyager</span>
group is highlighted; the remaining three columns are independent live-web
benchmarks (<a href="https://github.com/MinorJerry/WebVoyager" target="_blank" rel="noopener">WebVoyager</a>,
<a href="https://huggingface.co/datasets/osunlp/Online-Mind2Web" target="_blank" rel="noopener">Online-Mind2Web</a>,
and <a href="https://huggingface.co/datasets/DeepShop/DeepShop" target="_blank" rel="noopener">DeepShop</a>). WebHarbor-WebVoyager is a stable, authentication-free subset
curated from the original WebVoyager suite.
</p>
</div>
</section>
<!-- =================== QUICKSTART =================== -->
<section id="quickstart" class="section bordered quickstart">
<div class="container">
<div class="section-head">
<span class="tag">05 · Quickstart</span>
<h2>One command, 15 sites.</h2>
<p class="section-lede">
All mirrors ship in a single Docker image. Each site binds to a
fixed port offset from 40000; point your agent at
<code>http://localhost:40000–40014</code>.
</p>
</div>
<div class="code-block">
<div class="code-head">
<span class="code-lbl">shell</span>
<button class="copy-btn" data-copy="docker run -p 40000-40014:40000-40014 battalion7244/webharbor:latest">Copy</button>
</div>
<pre><code>docker run -p 40000-40014:40000-40014 battalion7244/webharbor:latest</code></pre>
</div>
<p style="color:var(--text-soft); line-height:1.8; margin:0;">
Sites are mapped to ports <code>:40000</code> through <code>:40014</code> in order:
Allrecipes, Amazon, Apple, ArXiv, BBC News, Booking, GitHub,
Google Flights, Google Maps, Google Search, Hugging Face,
Wolfram Alpha, Cambridge Dictionary, Coursera, and ESPN.
</p>
</div>
</section>
<!-- =================== CONTRIBUTION =================== -->
<section id="contribute" class="section bordered">
<div class="container">
<div class="section-head">
<span class="tag">06 · Contribution</span>
<h2>Scale to 100+ sites with community power.</h2>
<p class="section-lede">
We have built 15 high-quality website environments that fully support
the <a href="https://github.com/MinorJerry/WebVoyager" target="_blank" rel="noopener">WebVoyager</a>
benchmark. Our next goal is to scale to <strong>100+ websites</strong>,
covering all environments in
<a href="https://huggingface.co/datasets/osunlp/Online-Mind2Web" target="_blank" rel="noopener">Online-Mind2Web</a>
(147 sites).
We are inviting the community to build this together.
</p>
</div>
<!-- two tracks -->
<h3 class="compare-title">Two tracks to join the author list</h3>
<div class="impact-grid" style="margin-bottom: 40px;">
<div class="impact-card">
<h3>Track A: Contribute a new website</h3>
<p>
Using coding agent to build a new mirror site (frontend + backend + database + tasks).
Contributing <strong>one website</strong> qualifies you for consideration for the
final paper's author list, subject to review and quality standards.
</p>
<ol style="margin:16px 0 0; padding-left:20px; color:var(--text-soft); font-size:0.92rem; line-height:1.7;">
<li>Browse the <a href="https://docs.google.com/spreadsheets/d/1vZsrQjy9nJKze58fx4kbQtFi85NjVXIWCFyu3ShD7gk/edit?gid=0#gid=0" target="_blank" rel="noopener">website tracking sheet</a> and find an unclaimed site you're interested in.</li>
<li>Submit the <a href="https://forms.gle/ngcD1rzAfUEphNmRA" target="_blank" rel="noopener">contribution form</a> to claim it. We will lock it to prevent duplicate work.</li>
<li>Once confirmed, follow our <a href="guide-create.html">Website Contribution Guide</a> to build and submit a PR.</li>
</ol>
</div>
<div class="impact-card">
<h3>Track B: Review environments</h3>
<p>
Review submitted mirror sites by checking visual fidelity, functional
correctness, and task grounding. <strong>Reviewing 5 environments</strong>
earns a spot on the author list.
</p>
<ol style="margin:16px 0 0; padding-left:20px; color:var(--text-soft); font-size:0.92rem; line-height:1.7;">
<li>Browse open <a href="https://github.com/aiming-lab/WebHarbor/pulls" target="_blank" rel="noopener">Pull Requests</a> on GitHub.</li>
<li>Review the submitted environment: does it support all proposed tasks? Are the tasks meaningful and challenging?</li>
<li>Follow our recommended <a href="guide-review.html">Review Pipeline</a> for systematic verification.</li>
</ol>
</div>
</div>
<!-- acknowledgement track -->
<div class="thesis">
<span class="thesis-kicker">Acknowledgement</span>
<p>
Any improvement to existing environments (bug fixes, UI polish, data
enrichment, task suggestions, or even feedback) qualifies for the
paper's acknowledgement section. Every contribution
matters.
</p>
</div>
<!-- CTA buttons -->
<div style="display:flex; gap:14px; flex-wrap:wrap; margin-top:36px;">
<a class="btn btn-primary" href="https://docs.google.com/spreadsheets/d/1vZsrQjy9nJKze58fx4kbQtFi85NjVXIWCFyu3ShD7gk/edit?gid=0#gid=0" target="_blank" rel="noopener">
View website tracking sheet
</a>
<a class="btn btn-ghost" href="https://forms.gle/ngcD1rzAfUEphNmRA" target="_blank" rel="noopener">
Submit contribution form
</a>
<a class="btn btn-ghost" href="https://github.com/aiming-lab/WebHarbor/pulls" target="_blank" rel="noopener">
Review open PRs
</a>
</div>
</div>
</section>
<!-- =================== CITATION =================== -->
<section id="citation" class="section bordered citation-section">
<div class="container">
<div class="section-head">
<span class="tag">Citation</span>
<h2>BibTeX</h2>
</div>
<div class="code-block">
<div class="code-head">
<span class="code-lbl">bibtex</span>
<button class="copy-btn" data-copy="@misc{webharbor2026, title = {WebHarbor: Docking Real Websites for Evolving GUI Agent Environments}, author = {{WebHarbor Team and Contributors}}, year = {2026}, url = {https://aiming-lab.github.io/webharbor.github.io}, note = {Project website.} }">Copy</button>
</div>
<pre><code>@misc{webharbor2026,
title = {WebHarbor: Docking Real Websites for Evolving GUI Agent Environments},
author = {{WebHarbor Team and Contributors}},
year = {2026},
url = {https://aiming-lab.github.io/webharbor.github.io},
note = {Project website.}
}</code></pre>
</div>
</div>
</section>
</main>
<!-- =================== LIGHTBOX =================== -->
<div class="lightbox" id="lightbox" aria-hidden="true" role="dialog" aria-label="Demo preview">
<button class="lb-close" id="lb-close" aria-label="Close">×</button>
<button class="lb-nav lb-prev" id="lb-prev" aria-label="Previous">‹</button>
<button class="lb-nav lb-next" id="lb-next" aria-label="Next">›</button>
<div class="lb-stage">
<img class="lb-img" id="lb-img" alt="" />
<div class="lb-caption" id="lb-caption">
<div class="lb-title" id="lb-title"></div>
<div class="lb-sub" id="lb-sub"></div>
</div>
</div>
</div>
<script src="assets/main.js"></script>
<script>
// --- Validation chart: model ranking across benchmarks ----------------
(function renderValidationChart() {
function init() {
var el = document.getElementById("validation-chart");
if (!el || typeof Plotly === "undefined") return;
var benchmarks = ["WebHarbor-WebVoyager", "WebVoyager", "Online-Mind2Web", "Deepshop"];
var models = [
{ name: "Qwen3-VL-4B-Thinking", scores: [62.8, 52.6, 32.0, 33.3], color: "#9CB7DC" },
{ name: "Qwen3-VL-235B-A22B-Thinking", scores: [76.4, 66.4, 63.7, 56.7], color: "#4D72AE" },
{ name: "Orchard-GUI-4B", scores: [78.5, 74.1, 67.0, 64.0], color: "#1B4B8F" }
];
var traces = models.map(function (m) {
return {
type: "bar",
name: m.name,
x: benchmarks,
y: m.scores,
text: m.scores.map(function (v) { return v.toFixed(1); }),
textposition: "outside",
textfont: { family: "Instrument Sans, system-ui, sans-serif", size: 12, color: "#0C1222" },
marker: { color: m.color, line: { color: "#0F2E5A", width: 0.4 } },
hovertemplate: "<b>%{x}</b><br>" + m.name + ": %{y}<extra></extra>"
};
});
var layout = {
barmode: "group",
bargap: 0.28,
bargroupgap: 0.08,
margin: { l: 56, r: 24, t: 18, b: 70 },
font: { family: "Instrument Sans, system-ui, sans-serif", color: "#3D4559", size: 13 },
yaxis: {
title: { text: "Success Rate", font: { size: 13, color: "#3D4559" } },
range: [0, 92],
gridcolor: "#E8E5DC",
zerolinecolor: "#D6D3C8",
tickfont: { size: 12, color: "#6B7280" }
},
xaxis: {
tickfont: { size: 13, color: "#0C1222" },
tickangle: 0,
ticklen: 14,
tickcolor: "rgba(0,0,0,0)"
},
legend: {
orientation: "h",
y: -0.18,
x: 0.5,
xanchor: "center",
font: { size: 12, color: "#3D4559" }
},
shapes: [
{
type: "rect",
xref: "x",
yref: "paper",
x0: -0.5,
x1: 0.5,
y0: 0,
y1: 1,
fillcolor: "#FDF0E6",
opacity: 0.6,
line: { width: 0 },
layer: "below"
}
],
plot_bgcolor: "rgba(0,0,0,0)",
paper_bgcolor: "rgba(0,0,0,0)"
};
Plotly.newPlot(el, traces, layout, {
displayModeBar: false,
responsive: true
});
}
if (typeof Plotly !== "undefined") {
init();
} else {
// Plotly is loaded via `defer`, so wait for DOMContentLoaded.
if (document.readyState === "loading") {
document.addEventListener("DOMContentLoaded", init);
} else {
// DOM ready but Plotly script may still be loading — poll briefly.
var tries = 0;
var iv = setInterval(function () {
if (typeof Plotly !== "undefined") { clearInterval(iv); init(); }
else if (++tries > 50) { clearInterval(iv); }
}, 80);
}
}
})();
</script>
</body>
</html>