Repository navigation
Expand file tree
/
Copy pathshowcase.html
More file actions
1849 lines (1720 loc) · 115 KB
/
Copy pathshowcase.html
File metadata and controls
1849 lines (1720 loc) · 115 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
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en" class="scroll-smooth">
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>ARC — Autonomous Agent Architecture Showcase</title>
<script src="https://cdn.tailwindcss.com"></script>
<script>
tailwind.config = {
darkMode: 'class',
theme: {
extend: {
colors: {
void: '#0A0A0F',
surface: '#13131F',
card: '#13131F',
panel: '#1A1A2E',
line: 'rgba(148,163,184,0.12)',
primary: '#2563EB',
primaryGlow: '#60A5FA',
secondary: '#FF6B6B',
secondaryGlow: '#FF8E8E',
accent: '#2563EB',
muted: '#8B94A7',
},
fontFamily: {
sans: ['Inter', 'system-ui', '-apple-system', 'sans-serif'],
mono: ['JetBrains Mono', 'ui-monospace', 'monospace'],
}
}
}
};
</script>
<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/chart.umd.min.js"></script>
<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=Inter:wght@300;400;500;600;700;800;900&family=JetBrains+Mono:wght@400;500;600;700&display=swap" rel="stylesheet">
<style>
:root {
--primary: #2563EB;
--primary-glow: #60A5FA;
--secondary: #FF6B6B;
--bg: #0A0A0F;
}
body {
background: #0A0A0F;
color: #E6EAF2;
font-family: 'Inter', system-ui, sans-serif;
overflow-x: hidden;
}
::selection {
background: rgba(37, 99, 235, 0.35);
color: #FFFFFF;
}
.bg-grid {
background-image: linear-gradient(rgba(148, 163, 184, 0.05) 1px, transparent 1px),
linear-gradient(90deg, rgba(148, 163, 184, 0.05) 1px, transparent 1px);
background-size: 44px 44px;
mask-image: radial-gradient(ellipse 90% 70% at 50% 0%, black 40%, transparent 100%);
}
.glass-card {
background: rgba(13, 20, 36, 0.75);
backdrop-filter: blur(16px);
-webkit-backdrop-filter: blur(16px);
border: 1px solid rgba(148, 163, 184, 0.12);
}
.glass-card-primary {
background: rgba(13, 20, 36, 0.85);
backdrop-filter: blur(16px);
-webkit-backdrop-filter: blur(16px);
border: 1px solid rgba(37, 99, 235, 0.3);
box-shadow: 0 0 30px rgba(37, 99, 235, 0.12);
}
.glass-card-secondary {
background: rgba(13, 20, 36, 0.85);
backdrop-filter: blur(16px);
-webkit-backdrop-filter: blur(16px);
border: 1px solid rgba(255, 107, 107, 0.3);
box-shadow: 0 0 30px rgba(255, 107, 107, 0.12);
}
.glow-primary {
box-shadow: 0 0 0 1px rgba(37, 99, 235, 0.25), 0 0 35px rgba(37, 99, 235, 0.15);
}
input[type=range] {
-webkit-appearance: none;
width: 100%;
height: 8px;
border-radius: 9999px;
background: #1E293B;
outline: none;
}
input[type=range]::-webkit-slider-thumb {
-webkit-appearance: none;
width: 26px;
height: 26px;
border-radius: 50%;
background: #0A0A0F;
border: 3px solid #2563EB;
box-shadow: 0 0 0 6px rgba(37, 99, 235, 0.2), 0 0 20px rgba(37, 99, 235, 0.6);
cursor: grab;
transition: transform 0.1s ease;
}
input[type=range]::-webkit-slider-thumb:hover {
transform: scale(1.15);
}
input[type=range]::-moz-range-thumb {
width: 24px;
height: 24px;
border-radius: 50%;
background: #0A0A0F;
border: 3px solid #2563EB;
box-shadow: 0 0 20px rgba(37, 99, 235, 0.6);
cursor: grab;
}
@keyframes pulse-slow {
0%, 100% { opacity: 0.8; transform: scale(1); }
50% { opacity: 0.3; transform: scale(1.08); }
}
.animate-pulse-slow {
animation: pulse-slow 3s ease-in-out infinite;
}
@keyframes flow-line {
to { stroke-dashoffset: -32; }
}
.flow-active {
stroke-dasharray: 6 6;
animation: flow-line 1.2s linear infinite;
}
</style>
</head>
<body class="antialiased selection:bg-blue-500/30 selection:text-white">
<!-- Ambient background glow layers -->
<div class="fixed inset-0 pointer-events-none z-0">
<div class="absolute inset-0 bg-grid"></div>
<div class="absolute -top-40 left-1/2 -translate-x-1/2 w-[1000px] h-[550px] rounded-full opacity-20 blur-[140px] bg-blue-500"></div>
<div class="absolute top-[40%] -left-64 w-[600px] h-[600px] rounded-full opacity-10 blur-[130px] bg-red-500"></div>
<div class="absolute bottom-10 -right-64 w-[600px] h-[600px] rounded-full opacity-10 blur-[140px] bg-blue-400"></div>
</div>
<!-- Global Navigation Header -->
<nav class="relative z-30 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 pt-6 pb-4 flex items-center justify-between">
<a href="index.html" class="flex items-center gap-3 group">
<div class="relative w-11 h-11 shrink-0">
<svg viewBox="0 0 44 44" class="w-11 h-11" role="img" aria-label="ARC logo">
<defs>
<linearGradient id="arc-logo-primary" x1="0%" y1="0%" x2="100%" y2="100%">
<stop offset="0%" stop-color="#60A5FA"/>
<stop offset="100%" stop-color="#2563EB"/>
</linearGradient>
<linearGradient id="arc-logo-secondary" x1="0%" y1="100%" x2="100%" y2="0%">
<stop offset="0%" stop-color="#60A5FA"/>
<stop offset="100%" stop-color="#2563EB"/>
</linearGradient>
</defs>
<rect x="1" y="1" width="42" height="42" rx="11" fill="#0A0A0F" stroke="rgba(37,99,235,.45)" stroke-width="1.5"/>
<path d="M22 8 L32 17 L26 17 L30 27 L22 27 L22 35 L12 20 L19 20 Z" fill="url(#arc-logo-primary)"/>
<circle cx="31.5" cy="30.5" r="5.2" fill="#0A0A0F" stroke="url(#arc-logo-secondary)" stroke-width="1.8"/>
<circle cx="31.5" cy="30.5" r="1.7" fill="#60A5FA"/>
</svg>
<span class="absolute -bottom-0.5 -right-0.5 w-3 h-3 rounded-full bg-blue-400 border-2 border-void"></span>
</div>
<div>
<div class="font-extrabold text-lg tracking-tight leading-none flex items-center gap-1.5">
ARC <span class="text-[10px] font-mono font-bold px-1.5 py-0.5 rounded bg-blue-500/20 text-blue-300 border border-blue-500/30 tracking-normal">ON SOLARI</span>
</div>
<div class="text-[10px] font-mono text-muted tracking-[.22em] uppercase mt-1">Harness + ARC Index</div>
</div>
</a>
<div class="hidden md:flex items-center gap-5 text-sm text-slate-300 font-medium">
<a href="#hero" class="hover:text-blue-400 transition-colors">Overview</a>
<a href="#measured" class="hover:text-blue-400 transition-colors">Measured</a>
<a href="#mcp" class="hover:text-blue-400 transition-colors">Claude Code</a>
<a href="#architecture" class="hover:text-blue-400 transition-colors">Architecture</a>
<a href="#simulator" class="hover:text-blue-400 transition-colors">Simulator</a>
<a href="#benchmarks" class="hover:text-blue-400 transition-colors">Benchmarks</a>
<a href="#monitors" class="hover:text-blue-400 transition-colors">Monitors</a>
<a href="#timeline" class="hover:text-blue-400 transition-colors">Phases</a>
<a href="#roadmap" class="hover:text-blue-400 transition-colors">Roadmap</a>
</div>
<div class="flex items-center gap-3">
<span class="hidden sm:inline-flex items-center gap-1.5 px-3 py-1 rounded-full text-xs font-mono bg-blue-500/10 text-blue-400 border border-blue-500/30">
<span class="w-2 h-2 rounded-full bg-blue-400 animate-ping"></span>
267 TESTS
</span>
<a href="explain.html" class="px-3 py-1.5 rounded-lg border border-blue-500/30 bg-blue-500/10 hover:bg-blue-500/20 text-blue-300 font-mono text-xs font-bold transition flex items-center gap-1.5">
<svg class="w-3.5 h-3.5 text-blue-400" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5"><polygon points="13 2 3 14 12 14 11 22 21 10 12 10 13 2"/></svg>
<span>Motion Explainer</span>
</a>
<a href="#simulator" class="hidden sm:block px-3.5 py-1.5 rounded-lg bg-blue-500 hover:bg-blue-400 text-black font-bold text-xs uppercase tracking-wider transition-all shadow-[0_0_20px_rgba(37,99,235,0.3)]">
Simulate
</a>
</div>
</nav>
<!-- ========================================================================= -->
<!-- 1. HERO SECTION -->
<!-- ========================================================================= -->
<section id="hero" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 pt-12 md:pt-20 pb-16">
<div class="grid lg:grid-cols-12 gap-12 items-center">
<!-- Left Hero Column -->
<div class="lg:col-span-7">
<div class="inline-flex items-center gap-2 px-3.5 py-1.5 rounded-full glass-card text-xs font-mono text-slate-300 mb-6 border border-blue-500/20">
<span class="w-2 h-2 rounded-full bg-blue-400"></span>
<span>MEASURED ON SOLARI · 2026-09-29</span>
<span class="text-slate-500">|</span>
<span class="text-blue-400 font-semibold">NOW AN MCP SERVER FOR CLAUDE CODE</span>
</div>
<h1 class="text-4xl sm:text-5xl lg:text-6xl font-black tracking-tight leading-[1.05] text-white">
ARC on Solari
<span class="block mt-2 text-transparent bg-clip-text bg-gradient-to-r from-blue-300 via-blue-400 to-blue-600">
Verified actions, measured
</span>
</h1>
<p class="mt-6 text-lg text-slate-300 leading-relaxed max-w-2xl">
<strong>ARC CUA</strong> is a browser harness for agents: a token-budgeted accessibility tree with <code>[#N]</code> actions, verified by state-change checks and batched fills, on Solari cloud browsers.
<strong>ARC Index</strong> builds on it to turn documents into verified actions. Every field needs an evidence quote from the cited row, and a value that can't be proven is left blank instead of typed into the form.
</p>
<!-- CTA Buttons -->
<div class="mt-8 flex flex-wrap gap-4">
<a href="#measured" class="px-6 py-3.5 rounded-xl bg-blue-500 hover:bg-blue-400 text-black font-black text-sm tracking-wide transition-all shadow-[0_0_30px_rgba(37,99,235,0.35)] flex items-center gap-2">
<span>See Measured Results</span>
<svg class="w-4 h-4" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2.5" d="M19 14l-7 7m0 0l-7-7m7 7V3"></path></svg>
</a>
<a href="#architecture" class="px-6 py-3.5 rounded-xl glass-card hover:border-blue-500/40 text-slate-200 font-bold text-sm tracking-wide transition-all flex items-center gap-2">
<span>View Cascading Architecture</span>
<svg class="w-4 h-4 text-blue-400" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9 5l7 7-7 7"></path></svg>
</a>
</div>
<!-- Three Stat Cards (measured, see #measured) -->
<div class="mt-12 grid grid-cols-1 sm:grid-cols-3 gap-4">
<div class="p-5 rounded-2xl glass-card glow-primary relative overflow-hidden">
<div class="text-3xl sm:text-4xl font-black text-blue-400 tracking-tight">8.4×</div>
<div class="text-xs uppercase font-mono tracking-wider text-slate-300 font-bold mt-1">Cheaper per appeal</div>
<div class="text-[11px] text-muted mt-1">$0.00305 vs $0.02575 · ARC Index vs traditional agent</div>
</div>
<div class="p-5 rounded-2xl glass-card relative overflow-hidden border border-teal-500/30">
<div class="text-3xl sm:text-4xl font-black text-blue-300 tracking-tight">24 / 24</div>
<div class="text-xs uppercase font-mono tracking-wider text-slate-300 font-bold mt-1">vs 21/24 on Solari MCP</div>
<div class="text-[11px] text-muted mt-1">Same model · 3.3× cheaper per task</div>
</div>
<div class="p-5 rounded-2xl glass-card relative overflow-hidden border border-slate-700">
<div class="text-3xl sm:text-4xl font-black text-white tracking-tight">16 → 0</div>
<div class="text-xs uppercase font-mono tracking-wider text-slate-300 font-bold mt-1">Wrong values let through</div>
<div class="text-[11px] text-blue-400 font-mono mt-1">Forced pick, two denied lines</div>
</div>
</div>
</div>
<!-- Right Hero Column: Interactive Terminal Preview -->
<div class="lg:col-span-5">
<div class="rounded-2xl glass-card-primary p-5 font-mono text-xs shadow-2xl relative">
<!-- Terminal Header -->
<div class="flex items-center justify-between pb-3 mb-3 border-b border-white/10 text-slate-400">
<div class="flex items-center gap-2">
<span class="w-3 h-3 rounded-full bg-red-500/80 inline-block"></span>
<span class="w-3 h-3 rounded-full bg-blue-500/80 inline-block"></span>
<span class="w-3 h-3 rounded-full bg-blue-500/80 inline-block"></span>
<span class="ml-2 text-slate-300 font-semibold">arc-trace (illustrative)</span>
</div>
<span class="text-amber-300 font-bold">● MOCK TRACE</span>
</div>
<!-- Telemetry Stream Content -->
<div class="space-y-2 text-slate-300 leading-relaxed overflow-x-auto">
<div class="text-slate-500">// Illustrative trace from the mock design model (not timed on Solari)</div>
<div><span class="text-blue-400">[0.018ms]</span> <span class="text-blue-300">AT-SPI2 D-Bus:</span> Accessible desktop tree serialized</div>
<div><span class="text-blue-400">[0.024ms]</span> <span class="text-blue-300">CDP AXTree:</span> Sanitized DOM accessibility tree (41 nodes)</div>
<div><span class="text-blue-400">[1.050ms]</span> <span class="text-slate-200">ReflexRunner:</span> Public Playwright click <span class="text-blue-300">#checkout-btn</span></div>
<div><span class="text-blue-400">[1.053ms]</span> <span class="text-slate-200">StateVerifier:</span> 64-bit SimHash Hamming dist = <span class="text-blue-400">14 (STATE_CHANGED)</span></div>
<div><span class="text-blue-400">[1.095ms]</span> <span class="text-purple-300">StuckMonitor:</span> 7-pattern check = <span class="text-blue-400">HEALTHY (p=0.012)</span></div>
<div><span class="text-blue-400">[1.110ms]</span> <span class="text-purple-300">MilestoneMonitor:</span> Goal progress delta = <span class="text-blue-400">+0.85</span></div>
<div class="p-2.5 rounded-lg bg-blue-950/40 border border-blue-500/30 text-blue-300 text-xs">
<div class="font-bold font-mono text-[11px] uppercase tracking-wider text-blue-400 mb-1">Step Routing Verdict:</div>
<div class="text-xs text-slate-200 mb-1">Local Reflex handled step without cloud reasoning.</div>
<div class="flex flex-wrap items-center justify-between gap-1 text-[11px] font-mono pt-1 border-t border-blue-500/20">
<span>Marginal Cost: <strong class="text-blue-400">$0.000000</strong></span>
<span>Escalate: <strong class="text-blue-400">NO (98% Local)</strong></span>
</div>
</div>
<div class="pt-2 text-slate-500">// Measured on Solari, 2026-09-29</div>
<div><span class="text-blue-400">→ fill_many:</span> 7 fields in one call, 386 ms</div>
<div><span class="text-blue-400">→ appeal filed:</span> 11.4 s (traditional agent: 89 s)</div>
<div class="text-blue-400 font-bold">[PASS] wrong values let through: 0 · confirmation matched</div>
</div>
</div>
</div>
</div>
</section>
<!-- ========================================================================= -->
<!-- 1a. MEASURED RESULTS (live runs on Solari, 2026-09-27/29) -->
<!-- ========================================================================= -->
<section id="measured" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 py-16 border-t border-white/5">
<div class="text-center max-w-3xl mx-auto mb-12">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase mb-2">Live runs on Solari cloud browsers</div>
<h2 class="text-3xl sm:text-4xl font-black text-white tracking-tight">Measured results</h2>
<p class="mt-4 text-slate-300">Two products, both measured against the usual way of doing the same job, with the same model on the same
Solari browsers. Result files: <a class="underline" href="https://github.com/itw-code/arc-cua/tree/feat/gemini-harness-reel/artifacts/benchmarks">artifacts/benchmarks</a>.</p>
</div>
<div class="grid lg:grid-cols-2 gap-6">
<div class="glass-card rounded-2xl p-6">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase">ARC CUA · browser harness</div>
<h3 class="text-xl font-bold text-white mt-1">vs Solari's own MCP, same model</h3>
<p class="text-xs text-muted mt-1">Gemini 3.8 Flash as a tool-calling agent over <code>@solarisdk/mcp</code>, or through ARC · 8 tasks × 3 runs</p>
<table class="w-full mt-4 text-sm">
<thead><tr class="text-left text-xs font-mono text-muted uppercase"><th class="py-2"></th><th class="py-2 text-right">Solari MCP</th><th class="py-2 text-right">ARC</th></tr></thead>
<tbody class="divide-y divide-white/5">
<tr><td class="py-2 text-slate-300">Success</td><td class="py-2 text-right">21/24</td><td class="py-2 text-right font-bold text-blue-300">24/24</td></tr>
<tr><td class="py-2 text-slate-300">Input tokens, 24 runs</td><td class="py-2 text-right">458,546</td><td class="py-2 text-right font-bold text-blue-300">122,034 <span class="text-muted font-normal">(3.8×)</span></td></tr>
<tr><td class="py-2 text-slate-300">Model + browser, per 1,000 tasks</td><td class="py-2 text-right">$16.01</td><td class="py-2 text-right font-bold text-blue-300">$4.84 <span class="text-muted font-normal">(3.3×)</span></td></tr>
<tr><td class="py-2 text-slate-300">Sum of medians, 7 tasks both pass</td><td class="py-2 text-right">133.7 s</td><td class="py-2 text-right font-bold text-blue-300">110.6 s</td></tr>
</tbody>
</table>
<p class="text-xs text-muted mt-3">Google Flights: 0/3 for the tool-calling agent, 3/3 for ARC. <a class="underline" href="https://github.com/itw-code/arc-cua/blob/feat/gemini-harness-reel/docs/BENCHMARK_VS_SOLARI_MCP.md">Part E</a></p>
</div>
<div class="glass-card rounded-2xl p-6">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase">ARC Index · documents → verified actions</div>
<h3 class="text-xl font-bold text-white mt-1">vs a traditional browser agent, same denial appeal</h3>
<p class="text-xs text-muted mt-1">Same letters, same portal, Solari browsers on both sides · Gemini 3.8 Flash, n=10 appeals each</p>
<table class="w-full mt-4 text-sm">
<thead><tr class="text-left text-xs font-mono text-muted uppercase"><th class="py-2">Per appeal</th><th class="py-2 text-right">Traditional</th><th class="py-2 text-right">ARC Index</th></tr></thead>
<tbody class="divide-y divide-white/5">
<tr><td class="py-2 text-slate-300">Input tokens</td><td class="py-2 text-right">26,955</td><td class="py-2 text-right font-bold text-blue-300">1,209 <span class="text-muted font-normal">(22×)</span></td></tr>
<tr><td class="py-2 text-slate-300">Model + browser cost</td><td class="py-2 text-right">$0.02575</td><td class="py-2 text-right font-bold text-blue-300">$0.00305 <span class="text-muted font-normal">(8.4×)</span></td></tr>
<tr><td class="py-2 text-slate-300">Wall time, mean</td><td class="py-2 text-right">89 s</td><td class="py-2 text-right font-bold text-blue-300">11.4 s <span class="text-muted font-normal">(7.8×)</span></td></tr>
<tr><td class="py-2 text-slate-300">Right / wrong, of 60 fields</td><td class="py-2 text-right">49 / 0</td><td class="py-2 text-right font-bold text-blue-300">49 / 0</td></tr>
</tbody>
</table>
<p class="text-xs text-muted mt-3">Forced to pick between two denied lines, wrong values let through: 16 (old check) → <strong class="text-blue-300">0</strong> (uniqueness rule), Claude Sonnet 5.5, n=8.
<a class="underline" href="https://github.com/itw-code/arc-cua/blob/feat/gemini-harness-reel/docs/ARC_INDEX_PLAN.md">Plan §10–11</a></p>
</div>
</div>
<div class="grid sm:grid-cols-2 gap-6 mt-8">
<figure class="glass-card rounded-2xl overflow-hidden">
<video class="w-full block" src="artifacts/arc-index-economics-showreel.mp4" poster="artifacts/arc-index-economics-showreel-poster.jpg" controls preload="none" playsinline></video>
<figcaption class="p-4"><div class="font-bold text-white text-sm">Economics: traditional agent vs ARC Index</div>
<div class="text-xs text-muted mt-1">8.4× cheaper and 7.8× faster per appeal, same accuracy</div></figcaption>
</figure>
<figure class="glass-card rounded-2xl overflow-hidden">
<video class="w-full block" src="artifacts/arc-index-benchmark-showreel.mp4" poster="artifacts/arc-index-benchmark-showreel-poster.jpg" controls preload="none" playsinline></video>
<figcaption class="p-4"><div class="font-bold text-white text-sm">Wrong values let through</div>
<div class="text-xs text-muted mt-1">A confident pick between two denied lines, and the rule that rejects it</div></figcaption>
</figure>
<figure class="glass-card rounded-2xl overflow-hidden">
<video class="w-full block" src="artifacts/arc-index-showreel.mp4" poster="artifacts/arc-index-showreel-poster.jpg" controls preload="none" playsinline></video>
<figcaption class="p-4"><div class="font-bold text-white text-sm">ARC Index: one value, letter to receipt</div>
<div class="text-xs text-muted mt-1">Cited row, batched fill on Solari, SimHash check, receipt</div></figcaption>
</figure>
<figure class="glass-card rounded-2xl overflow-hidden">
<video class="w-full block" src="artifacts/gemini-harness-showreel.mp4" poster="artifacts/gemini-harness-showreel-poster.jpg" controls preload="none" playsinline></video>
<figcaption class="p-4"><div class="font-bold text-white text-sm">Same model, two harnesses</div>
<div class="text-xs text-muted mt-1">Gemini 3.8 Flash on Solari's MCP vs on ARC</div></figcaption>
</figure>
</div>
</section>
<!-- ========================================================================= -->
<!-- 1b. USE WITH CLAUDE CODE (MCP) -->
<!-- ========================================================================= -->
<section id="mcp" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 py-16 border-t border-white/5">
<div class="text-center max-w-3xl mx-auto mb-12">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase mb-2">Phases 9–13 · Agent Integration</div>
<h2 class="text-3xl sm:text-4xl font-black text-white tracking-tight">Use ARC from Claude Code</h2>
<p class="mt-3 text-slate-400 text-sm sm:text-base">
<code class="text-blue-300 font-mono">arc-cua-mcp</code> gives any MCP host five browser tools. It complements Solari's own MCP server: ARC hands the agent a hard-budgeted accessibility tree where every actionable element carries a <code class="text-blue-300 font-mono">[#N]</code> handle, verifies each action, and flags stalls. It drives a local Chromium, a Solari cloud browser, or any CDP endpoint.
</p>
</div>
<div class="grid lg:grid-cols-12 gap-6 mb-10">
<!-- Install -->
<div class="lg:col-span-5 min-w-0 p-6 rounded-2xl glass-card-primary flex flex-col">
<div class="text-xs font-mono font-bold text-blue-400 uppercase tracking-wider mb-3">Install</div>
<div class="p-4 rounded-xl bg-black/60 border border-white/10 font-mono text-xs text-blue-300 space-y-1 overflow-x-auto">
<div><span class="text-slate-500"># Package with the MCP extra</span></div>
<div class="whitespace-nowrap">pip install -e ".[mcp]"</div>
<div class="whitespace-nowrap">playwright install chromium</div>
<div><span class="text-slate-500"># Register with Claude Code</span></div>
<div class="whitespace-nowrap">claude mcp add --scope user arc -- arc-cua-mcp</div>
</div>
<ul class="mt-4 space-y-2 text-xs text-slate-300 leading-relaxed">
<li>For <code class="text-blue-300 font-mono">backend="solari"</code>, set <code class="text-blue-300 font-mono">SOLARI_API_KEY</code> in the environment Claude Code starts from. The server inherits it, so the key never appears in MCP config.</li>
<li>Solari browsers bill hourly until <code class="text-blue-300 font-mono">arc_close</code>; the server also releases them on shutdown.</li>
<li>Other MCP hosts: same command over stdio, <code class="text-blue-300 font-mono">{"command": "arc-cua-mcp"}</code>.</li>
</ul>
</div>
<!-- Tools -->
<div class="lg:col-span-7 grid grid-cols-1 sm:grid-cols-2 gap-4">
<div class="p-5 rounded-2xl glass-card border border-white/10">
<div class="font-mono text-sm font-bold text-blue-300">arc_open</div>
<p class="text-xs text-slate-400 mt-1">Open a URL in a local Chromium, a Solari cloud browser, or an existing CDP endpoint.</p>
</div>
<div class="p-5 rounded-2xl glass-card border border-white/10">
<div class="font-mono text-sm font-bold text-blue-300">arc_inspect</div>
<p class="text-xs text-slate-400 mt-1">Accessibility tree capped at ~1,200 tokens. Plain text is evicted before affordances; <code class="font-mono text-blue-300">query</code> reaches evicted elements by name.</p>
</div>
<div class="p-5 rounded-2xl glass-card border border-white/10">
<div class="font-mono text-sm font-bold text-blue-300">arc_act</div>
<p class="text-xs text-slate-400 mt-1">Click, type, select, press keys, scroll or navigate by <code class="font-mono text-blue-300">[#N]</code> index. Returns the page after the action with fresh indices, so no re-inspect is needed; repeated no-ops are flagged as a stall.</p>
</div>
<div class="p-5 rounded-2xl glass-card border border-white/10">
<div class="font-mono text-sm font-bold text-blue-300">arc_screenshot</div>
<p class="text-xs text-slate-400 mt-1">Screenshot with <code class="font-mono text-blue-300">[#N]</code> marks drawn on elements, for canvas and visual checks.</p>
</div>
<div class="p-5 rounded-2xl glass-card border border-white/10 sm:col-span-2">
<div class="font-mono text-sm font-bold text-blue-300">arc_close</div>
<p class="text-xs text-slate-400 mt-1">Close the browser and release the Solari session.</p>
</div>
</div>
</div>
<!-- Head-to-head -->
<div class="rounded-2xl glass-card border border-white/10 p-6 overflow-x-auto">
<div class="flex flex-wrap items-baseline justify-between gap-2 mb-4">
<h3 class="text-xl font-bold text-white">Head-to-head vs Solari's MCP (<code class="font-mono text-base">@solarisdk/mcp</code> v0.4.6)</h3>
<span class="text-[11px] font-mono text-slate-400">Both on Solari cloud browsers · 800×600 · 2026-09-27</span>
</div>
<div class="grid grid-cols-2 lg:grid-cols-4 gap-4 mb-6">
<div class="p-4 rounded-xl bg-black/40 border border-blue-500/30">
<div class="text-2xl font-black text-blue-400">7 / 7 <span class="text-slate-500 text-base font-bold">vs 6 / 7</span></div>
<div class="text-xs font-mono text-slate-300 uppercase mt-1">Grounding tasks succeeded</div>
</div>
<div class="p-4 rounded-xl bg-black/40 border border-blue-500/30">
<div class="text-2xl font-black text-blue-400">3.3× <span class="text-slate-500 text-base font-bold">fewer tokens</span></div>
<div class="text-xs font-mono text-slate-300 uppercase mt-1">8,438 vs 28,198 perception tokens</div>
</div>
<div class="p-4 rounded-xl bg-black/40 border border-blue-500/30">
<div class="text-2xl font-black text-blue-400">15× <span class="text-slate-500 text-base font-bold">fewer CDP</span></div>
<div class="text-xs font-mono text-slate-300 uppercase mt-1">255 vs 3,868 protocol commands</div>
</div>
<div class="p-4 rounded-xl bg-black/40 border border-blue-500/30">
<div class="text-2xl font-black text-blue-400">70 s <span class="text-slate-500 text-base font-bold">vs 86 s</span></div>
<div class="text-xs font-mono text-slate-300 uppercase mt-1">Time in tool calls</div>
</div>
</div>
<table class="w-full text-xs font-mono text-left min-w-[640px]">
<thead class="text-slate-400 border-b border-white/10">
<tr><th class="py-2 pr-4">Task</th><th class="py-2 pr-4 text-center">ARC</th><th class="py-2 pr-4 text-center">Solari</th><th class="py-2 pr-4 text-right">ARC calls / tokens / s / CDP</th><th class="py-2 text-right">Solari calls / tokens / s / CDP</th></tr>
</thead>
<tbody class="text-slate-300 divide-y divide-white/5">
<tr><td class="py-2 pr-4">HN: click "new"</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">3 / 1,237 / 6.9 / 28</td><td class="text-right">3 / 5,575 / 9.0 / 964</td></tr>
<tr><td class="py-2 pr-4">HN: 2nd "N comments" link</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">3 / 1,237 / 6.2 / 28</td><td class="text-right">3 / 5,575 / 10.3 / 964</td></tr>
<tr><td class="py-2 pr-4">Wikipedia: search</td><td class="text-center">✅</td><td class="text-center">❌</td><td class="text-right pr-4">5 / 2,568 / 21.6 / 51</td><td class="text-right">3 / 7,519 / 16.4 / 70</td></tr>
<tr><td class="py-2 pr-4">httpbin: fill form, pick radio, submit</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">5 / 468 / 12.3 / 65</td><td class="text-right">5 / 359 / 18.8 / 84</td></tr>
<tr><td class="py-2 pr-4">GitHub: Issues tab</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">3 / 1,223 / 9.0 / 29</td><td class="text-right">3 / 6,160 / 11.2 / 1,395</td></tr>
<tr><td class="py-2 pr-4">Python docs: "Coroutines and tasks"</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">3 / 1,239 / 7.9 / 28</td><td class="text-right">3 / 2,255 / 10.1 / 347</td></tr>
<tr><td class="py-2 pr-4">TodoMVC: add item</td><td class="text-center">✅</td><td class="text-center">✅</td><td class="text-right pr-4">4 / 466 / 6.3 / 26</td><td class="text-right">3 / 755 / 10.3 / 44</td></tr>
</tbody>
</table>
<p class="mt-4 text-xs text-slate-400 leading-relaxed">
<strong class="text-slate-200">Speed:</strong> each message to a Solari browser is a network round trip (0.2–0.5 s). ARC's first timed run was the slower one (99 s vs 78 s); reusing one CDP session per page and merging calls halved its extraction and action times. It is now faster on 6 of 7 tasks; Wikipedia still takes two extra calls.
</p>
<p class="mt-4 text-xs text-slate-400 leading-relaxed">
<strong class="text-slate-200">With a real model:</strong> a small, fast LLM (<code class="font-mono">laguna-xs-2-1</code>) driving ARC through the Oh My Pi agent, with ARC as its only tools, succeeded on all 13 task runs that reached the model, in 3–5 tool calls and about 25 s each.
</p>
<p class="mt-4 text-xs text-slate-400 leading-relaxed">
<strong class="text-slate-200">Why Solari missed Wikipedia:</strong> at 800 px the search box collapses behind a button. The HTML still contains the input, so an HTML-reading policy typed into a box that wasn't visible. The accessibility tree shows visibility, so ARC's policy clicked <em>Search</em> first.
</p>
<div class="mt-4 inline-flex items-start gap-2 px-3 py-2 rounded-lg bg-blue-500/5 border border-blue-500/25 text-blue-300 text-[11px] font-mono">
<span><strong>CAVEATS:</strong> scripted policies stand in for an agent (no LLM), one run each on public sites that change daily. Token counts are what each tool hands the agent; Solari's <code>text</code> read is cheaper on most pages but carries no handles. Full method in <a class="underline decoration-blue-500/50 hover:text-blue-200" href="https://github.com/itw-code/arc-cua/blob/master/docs/BENCHMARK_VS_SOLARI_MCP.md" target="_blank" rel="noopener">BENCHMARK_VS_SOLARI_MCP.md</a>.</span>
</div>
</div>
<!-- Showreel: Opus + Solari MCP vs Gemini Flash + ARC -->
<div class="mt-8 rounded-2xl glass-card border border-white/10 p-6">
<div class="flex flex-wrap items-baseline justify-between gap-2 mb-4">
<h3 class="text-xl font-bold text-white">Opus + Solari MCP vs Gemini 3.8 Flash + ARC, in 15 seconds</h3>
<span class="text-[11px] font-mono text-slate-400">Same Solari browser · 8 tasks · 2026-09-27</span>
</div>
<video class="w-full rounded-xl border border-white/10 bg-black" controls playsinline muted loop preload="metadata"
poster="artifacts/solari-race-showreel-poster.jpg">
<source src="artifacts/solari-race-showreel.mp4" type="video/mp4">
<a href="artifacts/solari-race-showreel.mp4">Download the video (MP4, 5 MB)</a>
</video>
<div class="mt-4 grid grid-cols-2 lg:grid-cols-4 gap-3 text-center">
<div class="p-3 rounded-xl bg-black/40 border border-blue-500/30"><div class="text-xl font-black text-blue-400">24/24 <span class="text-slate-500 text-sm font-bold">vs 7/8</span></div><div class="text-[11px] font-mono text-slate-300 uppercase mt-1">Tasks passed</div></div>
<div class="p-3 rounded-xl bg-black/40 border border-blue-500/30"><div class="text-xl font-black text-blue-400">1.9×</div><div class="text-[11px] font-mono text-slate-300 uppercase mt-1">Faster · 172 s vs 323 s</div></div>
<div class="p-3 rounded-xl bg-black/40 border border-blue-500/30"><div class="text-xl font-black text-blue-400">39×</div><div class="text-[11px] font-mono text-slate-300 uppercase mt-1">Fewer input tokens</div></div>
<div class="p-3 rounded-xl bg-black/40 border border-blue-500/30"><div class="text-xl font-black text-blue-400">33×</div><div class="text-[11px] font-mono text-slate-300 uppercase mt-1">Cheaper · $4.83 vs $160 / 1k tasks</div></div>
</div>
<p class="mt-4 text-xs text-slate-400 leading-relaxed">
Lane A is Claude Opus 5.5 in Claude Code using Solari's own MCP tools. Lane B is ARC's reflex policy: Gemini 3.8 Flash picks one action per call from ARC's numbered tree. A is a general-purpose agent and ran once per task; B is a narrow browsing loop and ran three times. A's Google Flights run opened a search URL that lands on the home page, not results. Method and raw data in <a class="underline decoration-blue-500/50 hover:text-blue-200" href="https://github.com/itw-code/arc-cua/blob/master/docs/BENCHMARK_VS_SOLARI_MCP.md" target="_blank" rel="noopener">BENCHMARK_VS_SOLARI_MCP.md</a>.
</p>
</div>
<div class="mt-8 flex flex-wrap justify-center gap-4">
<a href="https://github.com/itw-code/arc-cua/blob/master/skills/solari-hybrid-cua/SKILL.md" target="_blank" rel="noopener" class="px-6 py-3 rounded-xl glass-card text-white font-bold text-xs uppercase tracking-wider hover:border-blue-500/40 transition-all">Agent Skill</a>
<a href="https://github.com/itw-code/arc-cua/blob/master/docs/SOLARI_API.md" target="_blank" rel="noopener" class="px-6 py-3 rounded-xl glass-card text-white font-bold text-xs uppercase tracking-wider hover:border-blue-500/40 transition-all">Solari API Contract</a>
<a href="https://github.com/itw-code/arc-cua" target="_blank" rel="noopener" class="px-6 py-3 rounded-xl bg-blue-500 hover:bg-blue-400 text-black font-bold text-xs uppercase tracking-wider transition-all shadow-[0_0_20px_rgba(37,99,235,0.3)]">View Source</a>
</div>
</section>
<!-- ========================================================================= -->
<!-- 2. ARCHITECTURE DIAGRAM SECTION -->
<!-- ========================================================================= -->
<section id="architecture" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 py-16 border-t border-white/5">
<div class="mb-8 p-4 rounded-xl border border-amber-500/40 bg-amber-500/10 text-amber-200 text-sm leading-relaxed">
<strong class="font-mono text-xs uppercase tracking-wider text-amber-300">Design model, not live results.</strong>
The figures in this section (routing percentages, per-step costs and latencies) come from the repo's mock-execution scorecard
(<code>PROJECTED_BASED_ON_MOCK_EXECUTION</code>). The in-VM runner they assume is not built yet.
For live-run numbers, see <a href="#measured" class="underline text-amber-100">Measured results</a>.
</div>
<div class="text-center max-w-3xl mx-auto mb-12">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase mb-2">Hierarchical Execution</div>
<h2 class="text-3xl sm:text-4xl font-black text-white tracking-tight">Reflex + Cortex Cascading Architecture</h2>
<p class="mt-3 text-slate-400 text-sm sm:text-base">
Traditional agents query cloud LLMs on every single step. Arc confines routine UI execution to sub-10ms local reflex cycles, reserving frontier reasoning for genuine edge-case recovery.
</p>
</div>
<!-- Interactive Architecture Canvas -->
<div class="glass-card rounded-3xl p-6 sm:p-10 border border-white/10 relative overflow-hidden">
<!-- Tier 1 vs Tier 2 Badges -->
<div class="flex flex-wrap items-center justify-between gap-4 pb-6 mb-8 border-b border-white/10">
<div class="flex items-center gap-3">
<span class="px-3 py-1 rounded-lg bg-blue-500/20 text-blue-300 border border-blue-500/40 text-xs font-mono font-bold uppercase tracking-wider">
Tier 1: Local Fast Path (Reflex)
</span>
<span class="text-xs text-slate-400 hidden sm:inline">Sub-10ms UI Actuation · $0.0000 Cost · 98% Traffic</span>
</div>
<div class="flex items-center gap-3">
<span class="px-3 py-1 rounded-lg bg-red-500/20 text-red-300 border border-red-500/40 text-xs font-mono font-bold uppercase tracking-wider">
Tier 2: Cloud Cortex (Frontier LLM)
</span>
<span class="text-xs text-slate-400 hidden sm:inline">Targeted Recovery · 2% Anomaly Traffic</span>
</div>
</div>
<!-- Responsive Cascading Architecture Flow -->
<div class="relative max-w-4xl mx-auto py-2">
<!-- Step 1: Input Goal Card -->
<div class="max-w-md mx-auto p-4 rounded-xl bg-panel border border-slate-700 text-center shadow-lg">
<div class="text-[11px] font-mono text-slate-400 uppercase tracking-widest mb-1">User Intent & Benchmark Task</div>
<div class="text-sm font-bold text-white flex items-center justify-center gap-2">
<span class="w-2 h-2 rounded-full bg-blue-400"></span>
Goal & Instruction Input
</div>
</div>
<!-- Connector Down 1 -->
<div class="flex justify-center my-3">
<div class="w-0.5 h-7 bg-gradient-to-b from-blue-500 to-blue-400"></div>
</div>
<!-- Step 2: Tier 1 Reflex Engine Card -->
<div class="p-6 rounded-2xl glass-card-primary glow-primary relative overflow-hidden">
<div class="flex flex-wrap items-center justify-between gap-3 mb-3">
<span class="px-2.5 py-1 rounded bg-blue-500/20 text-blue-300 border border-blue-500/40 text-xs font-mono font-bold">
TIER 1: REFLEX ENGINE (LOCAL FAST PATH)
</span>
<div class="flex items-center gap-2 text-xs font-mono">
<span class="text-blue-400 font-bold">2.31 ms Avg Latency</span>
<span class="text-slate-600">•</span>
<span class="text-blue-400 font-bold">$0.000000 Cost</span>
</div>
</div>
<h3 class="text-lg sm:text-xl font-bold text-white mb-2">Deterministic UI Perception & Actuation</h3>
<p class="text-xs sm:text-sm text-slate-300 leading-relaxed mb-4">
Runs entirely inside the local Arc MicroVM without cloud WAN latency or multimodal token fees.
</p>
<div class="grid grid-cols-2 sm:grid-cols-4 gap-3 text-xs font-mono">
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 text-slate-300">
<div class="text-muted text-[10px] uppercase">CDP AXTree</div>
<div class="text-blue-300 font-bold mt-0.5">p50: 0.024ms</div>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 text-slate-300">
<div class="text-muted text-[10px] uppercase">AT-SPI2 D-Bus</div>
<div class="text-blue-300 font-bold mt-0.5">p50: 0.018ms</div>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 text-slate-300">
<div class="text-muted text-[10px] uppercase">State Verifier</div>
<div class="text-blue-300 font-bold mt-0.5">64-bit SimHash</div>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 text-slate-300">
<div class="text-muted text-[10px] uppercase">Playwright API</div>
<div class="text-blue-300 font-bold mt-0.5">0 AST Leaks</div>
</div>
</div>
</div>
<!-- Connector Down 2 -->
<div class="flex justify-center my-3">
<div class="w-0.5 h-7 bg-gradient-to-b from-blue-500 to-blue-400"></div>
</div>
<!-- Step 3: Cascading Monitor Hub -->
<div class="p-4 rounded-xl bg-panel border border-sky-500/30 text-center shadow-lg">
<div class="text-xs font-mono text-sky-400 font-bold uppercase tracking-wider mb-2">
Cascading Perception & Health Gatekeepers
</div>
<div class="flex flex-wrap justify-center items-center gap-2 sm:gap-4 text-xs font-mono text-slate-300">
<span class="px-3 py-1 rounded-md bg-black/40 border border-blue-500/30 text-blue-300 font-bold">
Stuck Monitor (94.2%)
</span>
<span class="px-3 py-1 rounded-md bg-black/40 border border-blue-500/30 text-blue-300 font-bold">
Milestone Monitor (98.1%)
</span>
<span class="px-3 py-1 rounded-md bg-black/40 border border-blue-500/30 text-blue-300 font-bold">
Session Guard (100%)
</span>
</div>
</div>
<!-- Connector Down 3 (Split Branch) -->
<div class="grid grid-cols-1 md:grid-cols-2 gap-6 mt-4">
<!-- Left Branch: Healthy (98%) -->
<div class="flex flex-col">
<div class="flex items-center gap-2 mb-2 text-xs font-mono font-bold text-blue-400">
<span class="w-2 h-2 rounded-full bg-blue-400 animate-pulse"></span>
HEALTHY STATE (98.0% OF ALL STEPS)
</div>
<div class="flex-1 p-5 rounded-2xl glass-card-primary border border-blue-500/30 flex flex-col justify-between">
<div>
<h4 class="text-base font-bold text-blue-300 mb-1">Advance Next Action Locally</h4>
<p class="text-xs text-slate-300 leading-relaxed mb-3">
Action executes deterministically in MicroVM. Next action queued with zero network hops.
</p>
</div>
<div class="p-2.5 rounded bg-black/40 border border-blue-500/20 font-mono text-xs flex justify-between text-slate-300">
<span>Execution Cost: <strong class="text-blue-400">$0.000000</strong></span>
<span>Latency: <strong class="text-blue-400">2.31 ms</strong></span>
</div>
</div>
</div>
<!-- Right Branch: Anomaly (2%) -->
<div class="flex flex-col">
<div class="flex items-center gap-2 mb-2 text-xs font-mono font-bold text-red-400">
<span class="w-2 h-2 rounded-full bg-red-400"></span>
ANOMALY DETECTED (2.0% OF STEPS)
</div>
<div class="flex-1 p-5 rounded-2xl glass-card-secondary border border-red-500/30 flex flex-col justify-between">
<div>
<div class="flex items-center justify-between mb-1">
<h4 class="text-base font-bold text-red-300">Tier 2: Cloud Cortex Escalation</h4>
<span class="text-[10px] font-mono px-2 py-0.5 rounded bg-red-500/20 text-red-300">Frontier LLM</span>
</div>
<p class="text-xs text-slate-300 leading-relaxed mb-3">
TypeSafe Jev, GPT-4o, or Claude Sonnet synthesizes targeted recovery plan; Recovery Compiler validates syntax before execution.
</p>
</div>
<div class="p-2.5 rounded bg-black/40 border border-red-500/20 font-mono text-xs flex justify-between text-slate-300">
<span>Recovery Cost: <strong class="text-red-400">~$0.001500</strong></span>
<span class="text-blue-300">Resume Reflex Queue →</span>
</div>
</div>
</div>
</div>
</div>
<!-- Explanatory Pillars Below Diagram -->
<div class="mt-8 grid grid-cols-1 md:grid-cols-2 gap-6 pt-6 border-t border-white/10 text-sm">
<div class="p-4 rounded-xl bg-black/40 border border-blue-500/20">
<div class="font-bold text-blue-400 font-mono uppercase text-xs mb-1">Tier 1: Deterministic Reflex (98% Traffic)</div>
<p class="text-slate-300 text-xs sm:text-sm leading-relaxed">
Perception operates via zero-copy Chrome DevTools Protocol (CDP) and Linux AT-SPI2 D-Bus accessibility streams rather than raw multimodal pixel grids. Actuation runs through public Playwright APIs with a 6-tier resilient selector chain and 64-bit SimHash state verification.
</p>
</div>
<div class="p-4 rounded-xl bg-black/40 border border-red-500/20">
<div class="font-bold text-red-400 font-mono uppercase text-xs mb-1">Tier 2: Escalation Cortex (2% Traffic)</div>
<p class="text-slate-300 text-xs sm:text-sm leading-relaxed">
Frontier reasoning models (Claude 3.5 Sonnet, GPT-4o, TypeSafe Jev) are called strictly when the local Stuck Monitor, Milestone Monitor, or Session Guard flags an anomaly. The Recovery Compiler normalizes and sanitizes the LLM plan before re-engaging local reflex execution.
</p>
</div>
</div>
<!-- Live Synchronized Observation-Action Loop Simulation (ColdStart Pattern) -->
<div class="mt-12 pt-8 border-t border-white/10">
<div class="flex flex-col sm:flex-row sm:items-center justify-between gap-4 mb-3">
<div>
<div class="font-mono text-xs tracking-widest text-blue-400 uppercase mb-1">02 — LIVE OBSERVATION–ACTION LOOP COMPARISON</div>
<h3 class="text-xl sm:text-2xl font-black text-white">Side-by-Side Execution Timing</h3>
<p class="text-xs text-slate-400 mt-1">Watch latency and cost bleed in real time: red crawls through WAN round-trips while green executes locally.</p>
</div>
<!-- View Toggles -->
<div class="inline-flex p-1 rounded-xl bg-black/50 border border-white/10 gap-1 font-mono text-xs">
<button id="view-compare-btn" onclick="setArchitectureView('compare')" class="px-3 py-1.5 rounded-lg bg-white/10 text-white font-bold transition">Compare</button>
<button id="view-trad-btn" onclick="setArchitectureView('trad')" class="px-3 py-1.5 rounded-lg text-slate-400 hover:text-white transition">Traditional</button>
<button id="view-arc-btn" onclick="setArchitectureView('arc')" class="px-3 py-1.5 rounded-lg text-slate-400 hover:text-white transition">Arc</button>
</div>
</div>
<!-- Epistemic status notice for the visualization -->
<div class="mb-5 inline-flex items-start gap-2 px-3 py-2 rounded-lg bg-blue-500/5 border border-blue-500/25 text-blue-300 text-[11px] font-mono">
<svg class="w-4 h-4 text-blue-400 shrink-0 mt-0.5" fill="none" stroke="currentColor" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M12 8v4m0 4h.01M21 12a9 9 0 11-18 0 9 9 0 0118 0z"/></svg>
<span>
<strong>VISUALIZED BENCHMARK — PROJECTED_BASED_ON_MOCK_EXECUTION.</strong>
Timing behavior is animated from the mock-scorecard parameters recorded in <code class="text-blue-200">artifacts/production/final_scorecard.json</code> (aggregate Reflex step: 2.31 ms; frontier baseline: >2,500 ms per step). Per-step rows are illustrative and normalized to those published aggregates; the browser does not measure these durations itself.
</span>
</div>
<div class="grid grid-cols-1 lg:grid-cols-2 gap-6 items-stretch">
<!-- Panel 1: Traditional CUA -->
<div id="arch-panel-trad" class="rounded-2xl glass-card-secondary p-5 sm:p-6 flex flex-col justify-between">
<div>
<div class="flex items-center justify-between pb-4 border-b border-red-500/20 mb-4">
<div>
<div class="font-black text-white text-base">Traditional Frontier CUA</div>
<div class="font-mono text-[10px] text-red-300 tracking-widest uppercase">SCREENSHOT → WAN UPLOAD → VLM INFERENCE</div>
</div>
<div class="text-right">
<div class="font-mono text-[10px] text-muted">STEP TIMER</div>
<div id="live-trad-timer" class="font-mono text-2xl font-black text-red-400">0 ms</div>
</div>
</div>
<!-- Steps List (illustrative rows normalized to published aggregates) -->
<div class="space-y-2 text-xs font-mono" id="live-trad-steps">
<div class="flex items-center justify-between px-2.5 pb-1 text-[10px] text-slate-500">
<span>PIPELINE STAGE</span>
<span>NORMALIZED SHARE · 2,850 ms CYCLE</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 flex items-center justify-between text-slate-300" data-trad-step="0">
<div class="flex items-center gap-2">
<span class="text-red-400 font-bold">01</span>
<span>Full display buffer raster capture</span>
</div>
<span class="text-muted">320 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 flex items-center justify-between text-slate-300" data-trad-step="1">
<div class="flex items-center gap-2">
<span class="text-red-400 font-bold">02</span>
<span>PNG upload over WAN to cloud provider</span>
</div>
<span class="text-muted">410 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 flex items-center justify-between text-slate-300" data-trad-step="2">
<div class="flex items-center gap-2">
<span class="text-red-400 font-bold">03</span>
<span>Multimodal 70B+ parameter VLM inference</span>
</div>
<span class="text-red-300 font-bold">1,840 ms · $0.048</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 flex items-center justify-between text-slate-300" data-trad-step="3">
<div class="flex items-center gap-2">
<span class="text-red-400 font-bold">04</span>
<span>Coordinate parse & actuation download</span>
</div>
<span class="text-muted">190 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-white/5 flex items-center justify-between text-slate-300" data-trad-step="4">
<div class="flex items-center gap-2">
<span class="text-red-400 font-bold">05</span>
<span>Blind mouse click dispatch (drift prone)</span>
</div>
<span class="text-muted">90 ms</span>
</div>
</div>
</div>
<div class="mt-4 pt-3 border-t border-red-500/20">
<div class="flex justify-between text-[11px] font-mono text-slate-400 mb-1">
<span>STEP CYCLE PROGRESS</span>
<span id="live-trad-pct" class="text-red-400 font-bold">0%</span>
</div>
<div class="w-full h-2 rounded-full bg-slate-800 overflow-hidden">
<div id="live-trad-bar" class="h-full rounded-full bg-red-500 transition-all duration-75" style="width: 0%"></div>
</div>
<div class="mt-2 text-[11px] font-mono text-red-300/80">
[LATENCY] Total step latency: ~2,850 ms · Cost per task: ~$0.4820
</div>
</div>
</div>
<!-- Panel 2: ARC -->
<div id="arch-panel-arc" class="rounded-2xl glass-card-primary p-5 sm:p-6 flex flex-col justify-between">
<div>
<div class="flex items-center justify-between pb-4 border-b border-blue-500/20 mb-4">
<div>
<div class="font-black text-white text-base">Arc Hybrid CUA</div>
<div class="font-mono text-[10px] text-blue-300 tracking-widest uppercase">LOCAL CDP/DOM → SIMHASH VERIFY → ZERO WAN</div>
</div>
<div class="text-right">
<div class="font-mono text-[10px] text-muted">COMPLETED LOOPS</div>
<div id="live-arc-loops" class="font-mono text-2xl font-black text-blue-400">0</div>
</div>
</div>
<!-- Steps List (illustrative rows normalized to published aggregates) -->
<div class="space-y-2 text-xs font-mono" id="live-arc-steps">
<div class="flex items-center justify-between px-2.5 pb-1 text-[10px] text-slate-500">
<span>PIPELINE STAGE</span>
<span>MEASURED AVG · 2.31 ms AGGREGATE</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-blue-500/20 flex items-center justify-between text-slate-300" data-arc-step="0">
<div class="flex items-center gap-2">
<span class="text-blue-400 font-bold">01</span>
<span>Zero-copy CDP AXTree / AT-SPI extraction</span>
</div>
<span class="text-blue-400 font-bold">0.024 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-blue-500/20 flex items-center justify-between text-slate-300" data-arc-step="1">
<div class="flex items-center gap-2">
<span class="text-blue-400 font-bold">02</span>
<span>6-tier resilient selector LRU cache lookup</span>
</div>
<span class="text-blue-400 font-bold">0.001 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-blue-500/20 flex items-center justify-between text-slate-300" data-arc-step="2">
<div class="flex items-center gap-2">
<span class="text-blue-400 font-bold">03</span>
<span>Public Playwright UI actuation</span>
</div>
<span class="text-blue-400 font-bold">1.050 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-blue-500/20 flex items-center justify-between text-slate-300" data-arc-step="3">
<div class="flex items-center gap-2">
<span class="text-blue-400 font-bold">04</span>
<span>64-bit SimHash Hamming state verification</span>
</div>
<span class="text-blue-400 font-bold">0.003 ms</span>
</div>
<div class="p-2.5 rounded-lg bg-black/40 border border-blue-500/20 flex items-center justify-between text-slate-300" data-arc-step="4">
<div class="flex items-center gap-2">
<span class="text-blue-400 font-bold">05</span>
<span>Stuck & Milestone monitor health gating</span>
</div>
<span class="text-blue-400 font-bold">0.047 ms</span>
</div>
</div>
</div>
<div class="mt-4 pt-3 border-t border-blue-500/20">
<div class="flex justify-between text-[11px] font-mono text-slate-400 mb-1">
<span>STEP EXECUTION LATENCY</span>
<span class="text-blue-400 font-bold">2.31 ms (AVG)</span>
</div>
<div class="w-full h-2 rounded-full bg-slate-800 overflow-hidden">
<div id="live-arc-bar" class="h-full rounded-full bg-blue-400 transition-all duration-75" style="width: 100%"></div>
</div>
<div class="mt-2 text-[11px] font-mono text-blue-300/90 flex justify-between">
<span>[CONFIRMED] Local step cost: $0.000000</span>
<span>98.0% steps handled off-API</span>
</div>
</div>
</div>
</div>
</div>
</div>
</section>
<!-- ========================================================================= -->
<!-- 3. INTERACTIVE COST & LATENCY SIMULATOR -->
<!-- ========================================================================= -->
<section id="simulator" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 py-16 border-t border-white/5">
<div class="mb-8 p-4 rounded-xl border border-amber-500/40 bg-amber-500/10 text-amber-200 text-sm leading-relaxed">
<strong class="font-mono text-xs uppercase tracking-wider text-amber-300">Design model, not live results.</strong>
The figures in this section ($0.0015 vs $0.4820 per task, 2.31 ms vs 2,850 ms per step) come from the repo's mock-execution scorecard
(<code>PROJECTED_BASED_ON_MOCK_EXECUTION</code>). The in-VM runner they assume is not built yet.
For live-run numbers, see <a href="#measured" class="underline text-amber-100">Measured results</a>.
</div>
<div class="text-center max-w-3xl mx-auto mb-12">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase mb-2">Live Economic Model</div>
<h2 class="text-3xl sm:text-4xl font-black text-white tracking-tight">Interactive Cost & Latency Simulator</h2>
<p class="mt-3 text-slate-400 text-sm sm:text-base">
Adjust the slider to simulate task volume and see the cost divergence the design model predicts between monolithic frontier LLMs and Arc Hybrid execution, parameterized with the mock-scorecard figures from <code class="text-blue-300 font-mono">FINAL_RESEARCH_REPORT.md</code>.
</p>
</div>
<!-- Simulator Workspace -->
<div class="glass-card rounded-3xl p-6 sm:p-10 border border-blue-500/20 glow-primary">
<!-- Slider Control Card -->
<div class="mb-10 p-6 rounded-2xl bg-black/50 border border-white/10">
<div class="flex flex-col sm:flex-row sm:items-center justify-between gap-4 mb-4">
<div>
<label for="step-slider" class="text-sm font-bold text-slate-200 uppercase tracking-wider font-mono">
Number of Autonomous UI Steps:
</label>
<div class="text-xs text-muted mt-0.5">Simulates agent interaction steps across production workflows</div>
</div>
<div class="flex items-center gap-3">
<span id="slider-value-badge" class="px-4 py-1.5 rounded-xl bg-blue-500/20 text-blue-300 border border-blue-500/40 text-2xl font-black font-mono">
1,000
</span>
<span class="text-xs font-mono text-slate-400">steps</span>
</div>
</div>
<!-- HTML Range Input -->
<input id="step-slider" type="range" min="10" max="10000" step="10" value="1000" class="w-full cursor-pointer" />
<!-- Quick Preset Buttons -->
<div class="mt-4 flex flex-wrap items-center justify-between gap-2 pt-3 border-t border-white/5 text-xs">
<span class="text-muted font-mono">Quick Presets:</span>
<div class="flex flex-wrap gap-2">
<button onclick="setSlider(50)" class="px-3 py-1 rounded-lg bg-surface hover:bg-slate-700 text-slate-300 font-mono transition">50 (Micro)</button>
<button onclick="setSlider(500)" class="px-3 py-1 rounded-lg bg-surface hover:bg-slate-700 text-slate-300 font-mono transition">500 (Small Batch)</button>
<button onclick="setSlider(1000)" class="px-3 py-1 rounded-lg bg-blue-500/20 text-blue-300 border border-blue-500/30 font-mono transition">1,000 (Typical)</button>
<button onclick="setSlider(5000)" class="px-3 py-1 rounded-lg bg-surface hover:bg-slate-700 text-slate-300 font-mono transition">5,000 (Enterprise)</button>
<button onclick="setSlider(10000)" class="px-3 py-1 rounded-lg bg-surface hover:bg-slate-700 text-slate-300 font-mono transition">10,000 (Full-Scale)</button>
</div>
</div>
</div>
<!-- Comparative Stat Cards -->
<div class="grid grid-cols-1 sm:grid-cols-2 lg:grid-cols-4 gap-4 mb-8">
<!-- Card 1: Frontier LLM Baseline Cost -->
<div class="p-5 rounded-2xl glass-card-secondary">
<div class="text-xs font-mono font-bold text-red-400 uppercase tracking-wider">Frontier LLM Baseline</div>
<div id="frontier-cost-display" class="text-3xl font-black text-red-400 mt-2 font-mono">$48.20</div>
<div class="text-xs text-slate-400 mt-1">Cost at ~$0.482 / 10 steps</div>
<div class="mt-3 text-[11px] font-mono text-slate-400 border-t border-red-500/20 pt-2 flex justify-between">
<span>Avg Latency:</span>
<span class="text-red-300 font-bold">2,850 ms / step</span>
</div>
</div>
<!-- Card 2: Arc Hybrid CUA Cost -->
<div class="p-5 rounded-2xl glass-card-primary">
<div class="text-xs font-mono font-bold text-blue-400 uppercase tracking-wider">Arc Hybrid CUA</div>
<div id="arc-cost-display" class="text-3xl font-black text-blue-400 mt-2 font-mono">$0.15</div>
<div class="text-xs text-slate-400 mt-1">Cost at ~$0.0015 / 10 steps</div>
<div class="mt-3 text-[11px] font-mono text-slate-400 border-t border-blue-500/20 pt-2 flex justify-between">
<span>Avg Latency:</span>
<span class="text-blue-300 font-bold">2.31 ms / step</span>
</div>
</div>
<!-- Card 3: Net Operational Savings -->
<div class="p-5 rounded-2xl glass-card border border-blue-500/40">
<div class="text-xs font-mono font-bold text-blue-300 uppercase tracking-wider">Net Capital Savings</div>
<div id="savings-display" class="text-3xl font-black text-blue-300 mt-2 font-mono">$48.05</div>
<div class="text-xs text-slate-400 mt-1">Operational budget preserved</div>
<div class="mt-3 text-[11px] font-mono text-slate-400 border-t border-white/10 pt-2 flex justify-between">
<span>Efficiency Delta:</span>
<span class="text-blue-400 font-bold">99.69% Margin</span>
</div>
</div>
<!-- Card 4: Cumulative Time Saved -->
<div class="p-5 rounded-2xl glass-card border border-slate-700">
<div class="text-xs font-mono font-bold text-slate-300 uppercase tracking-wider">Cumulative Time Saved</div>
<div id="time-saved-display" class="text-3xl font-black text-white mt-2 font-mono">47.4 min</div>
<div class="text-xs text-slate-400 mt-1">Zero WAN round-trip lag</div>
<div class="mt-3 text-[11px] font-mono text-slate-400 border-t border-white/10 pt-2 flex justify-between">
<span>Throughput Speedup:</span>
<span class="text-blue-300 font-bold">1,233× faster</span>
</div>
</div>
</div>
<!-- Chart.js Visualization Container -->
<div class="p-6 rounded-2xl bg-black/60 border border-white/10">
<div class="flex flex-col sm:flex-row sm:items-center justify-between gap-2 mb-4">
<div class="font-bold text-sm text-slate-200">Cumulative Execution Cost ($ USD) vs Step Volume</div>
<div class="text-xs font-mono text-slate-400 flex items-center gap-4">
<span class="flex items-center gap-1.5"><span class="w-3 h-3 rounded-full bg-red-500 inline-block"></span> Frontier Baseline</span>
<span class="flex items-center gap-1.5"><span class="w-3 h-3 rounded-full bg-blue-500 inline-block"></span> Arc Hybrid</span>
</div>
</div>
<div class="h-72 sm:h-80 w-full">
<canvas id="costSimulatorChart"></canvas>
</div>
</div>
</div>
</section>
<!-- ========================================================================= -->
<!-- 4. BENCHMARK RESULTS SECTION -->
<!-- ========================================================================= -->
<section id="benchmarks" class="relative z-10 max-w-7xl mx-auto px-4 sm:px-6 lg:px-8 py-16 border-t border-white/5">
<div class="mb-8 p-4 rounded-xl border border-amber-500/40 bg-amber-500/10 text-amber-200 text-sm leading-relaxed">
<strong class="font-mono text-xs uppercase tracking-wider text-amber-300">Design model, not live results.</strong>
The figures in this section (WebArena and OSWorld 100% success, per-task costs) come from the repo's mock-execution scorecard
(<code>PROJECTED_BASED_ON_MOCK_EXECUTION</code>). The in-VM runner they assume is not built yet.
For live-run numbers, see <a href="#measured" class="underline text-amber-100">Measured results</a>.
</div>
<div class="text-center max-w-3xl mx-auto mb-12">
<div class="text-xs font-mono font-bold tracking-widest text-blue-400 uppercase mb-2">Standardized Evaluation</div>
<h2 class="text-3xl sm:text-4xl font-black text-white tracking-tight">Autonomous Benchmark Results</h2>
<p class="mt-3 text-slate-400 text-sm sm:text-base">
Mock-execution evaluation comparing Arc Hybrid against leading autonomous agents on industry-standard WebArena (web automation) and OSWorld (desktop automation).
Task definitions follow Zhou et al. (WebArena) and Xie et al. (OSWorld) — see <a class="underline decoration-blue-500/50 text-blue-300 hover:text-blue-200" href="https://github.com/itw-code/arc-cua/blob/feat/gemini-harness-reel/docs/REFERENCES.md">docs/REFERENCES.md</a>.
Repository-measured figures (2.31 ms avg latency, $0.0015/task, 100% mock success) come from this repo's scorecard, not from those papers; external GPT-4o/Sonnet rows are this repo's recorded comparison values, not paper results.
</p>
<div class="mt-4 inline-flex items-center gap-2 px-3 py-1 rounded-md bg-blue-500/10 border border-blue-500/30 text-blue-300 text-xs font-mono">