-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathresults.html
More file actions
452 lines (431 loc) · 27.7 KB
/
results.html
File metadata and controls
452 lines (431 loc) · 27.7 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Results | MAFBench</title>
<script src="https://cdn.tailwindcss.com"></script>
<link rel="stylesheet" href="styles/main.css">
<link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/font-awesome/6.4.0/css/all.min.css">
</head>
<body class="bg-white">
<!-- Navigation -->
<nav id="navbar" class="fixed top-0 w-full bg-white/95 backdrop-blur-sm shadow-sm z-50">
<div class="max-w-7xl mx-auto px-4 sm:px-6 lg:px-8">
<div class="flex justify-between items-center h-16">
<div class="flex items-center">
<a href="index.html" class="text-2xl font-bold text-indigo-600">MAFBench</a>
</div>
<div class="hidden md:flex space-x-8 items-center">
<a href="index.html" class="text-gray-600 hover:text-indigo-600">Home</a>
<a href="design.html" class="text-gray-600 hover:text-indigo-600">Why Design Matters</a>
<div class="dropdown relative">
<a href="architecture/index.html" class="text-gray-600 hover:text-indigo-600 flex items-center">
Architecture Guide
<i class="fas fa-chevron-down ml-1 text-xs"></i>
</a>
<div class="dropdown-menu">
<a href="architecture/index.html">Overview</a>
<a href="architecture/single-agent.html">Single-Agent Characteristics</a>
<a href="architecture/multi-agent.html">Multi-Agent Characteristics</a>
<a href="architecture/environment.html">Environment</a>
</div>
</div>
<a href="results.html" class="text-gray-900 font-medium hover:text-indigo-600">Results</a>
<a href="mafbench.html" class="text-gray-600 hover:text-indigo-600">MAFBench</a>
<a href="paper.html" class="text-gray-600 hover:text-indigo-600">Paper & Code</a>
<a href="about.html" class="text-gray-600 hover:text-indigo-600">About</a>
</div>
<button id="mobile-menu-btn" class="md:hidden text-gray-600">
<i class="fas fa-bars text-xl"></i>
</button>
</div>
</div>
<div id="mobile-menu" class="hidden md:hidden bg-white border-t">
<div class="px-4 py-4 space-y-3">
<a href="index.html" class="block text-gray-600">Home</a>
<a href="design.html" class="block text-gray-600">Why Design Matters</a>
<div class="pl-2 border-l-2 border-gray-200">
<a href="architecture/index.html" class="block text-gray-900 font-medium mb-2">Architecture Guide</a>
<div class="pl-4 space-y-2">
<a href="architecture/index.html" class="block text-gray-600 text-sm">Overview</a>
<a href="architecture/single-agent.html" class="block text-gray-600 text-sm">Single-Agent Characteristics</a>
<a href="architecture/multi-agent.html" class="block text-gray-600 text-sm">Multi-Agent Characteristics</a>
<a href="architecture/environment.html" class="block text-gray-600 text-sm">Environment</a>
</div>
</div>
<a href="results.html" class="block text-gray-900 font-medium">Results</a>
<a href="mafbench.html" class="block text-gray-600">MAFBench</a>
<a href="paper.html" class="block text-gray-600">Paper & Code</a>
<a href="about.html" class="block text-gray-600">About</a>
</div>
</div>
</nav>
<!-- Hero Section -->
<section class="pt-32 pb-16 px-4 sm:px-6 lg:px-8 bg-white border-b border-gray-200">
<div class="max-w-4xl mx-auto text-center">
<div class="bg-blue-50 border-l-4 border-blue-400 p-4 mb-8 rounded max-w-3xl mx-auto">
<p class="text-sm text-gray-700">
<i class="fas fa-info-circle text-blue-600 mr-2"></i>
<strong>Note:</strong> All numerical values and tables on this page should be verified against the paper PDF for the most accurate and up-to-date results.
</p>
</div>
<h1 class="text-5xl md:text-6xl font-bold text-gray-900 mb-6">
Results
</h1>
<p class="text-xl text-gray-700 leading-relaxed">
Architecture, not models, determines performance. Same LLM, same task, wildly different outcomes.
</p>
</div>
</section>
<!-- Orchestration Overhead -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-white border-b border-gray-200">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-gray-900 mb-6">Orchestration Overhead</h2>
<p class="text-lg text-gray-700 mb-8 leading-relaxed">
Framework orchestration alone can create 100× latency differences, even for trivial tasks. We fix the LLM, query, and prompts, varying only the orchestration layer.
</p>
<div class="bg-gray-50 border border-gray-300 rounded-lg overflow-hidden mb-8">
<table class="w-full text-sm">
<thead class="bg-gray-100">
<tr>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Architecture Type</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Latency (p50)</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Throughput (req/s)</th>
</tr>
</thead>
<tbody class="divide-y divide-gray-200">
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Direct LLM</td>
<td class="px-6 py-4 text-gray-700">0.38s</td>
<td class="px-6 py-4 text-gray-700">8.88</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Graph-based (LangGraph)</td>
<td class="px-6 py-4 text-gray-700">0.52s (1.4×)</td>
<td class="px-6 py-4 text-gray-700">6.38</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Role-based (AutoGen, OpenAgents)</td>
<td class="px-6 py-4 text-gray-700">0.50s (1.3×)</td>
<td class="px-6 py-4 text-gray-700">6.83-7.06</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Role-based (CrewAI, Agno, OpenAI SDK)</td>
<td class="px-6 py-4 text-gray-700">0.61-1.17s (1.6-3.1×)</td>
<td class="px-6 py-4 text-gray-700">2.86-4.16</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">GABM (Concordia)</td>
<td class="px-6 py-4 text-gray-700">44.47s (117×)</td>
<td class="px-6 py-4 text-gray-700">0.089</td>
</tr>
</tbody>
</table>
</div>
<div class="bg-white border border-gray-300 rounded-lg p-6">
<h3 class="text-xl font-semibold text-gray-900 mb-3">Key Takeaway</h3>
<p class="text-gray-700 leading-relaxed">
Orchestration architecture alone governs baseline scalability. Graph- and role-based designs introduce modest overhead (1.3-3×), while GABM execution incurs orders-of-magnitude higher runtime and output even for trivial tasks, driven by execution semantics rather than task complexity.
</p>
</div>
</div>
</section>
<!-- Memory Architecture -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-gray-50 border-b border-gray-200">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-gray-900 mb-6">Memory Architecture Effects</h2>
<p class="text-lg text-gray-700 mb-8 leading-relaxed">
Memory structure matters more than context size. Retrieval enables stable recall, accumulation enables learning but scales poorly, and hybrid designs work best under bounded context.
</p>
<div class="bg-white border border-gray-300 rounded-lg overflow-hidden mb-8">
<table class="w-full text-sm">
<thead class="bg-gray-100">
<tr>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Memory Type</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">AR Score</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">TTL Score</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">LRU Score</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Overall</th>
</tr>
</thead>
<tbody class="divide-y divide-gray-200">
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Retrieval-only (LangGraph)</td>
<td class="px-6 py-4 text-gray-700">33.2</td>
<td class="px-6 py-4 text-gray-700">11.4</td>
<td class="px-6 py-4 text-gray-700">30.4</td>
<td class="px-6 py-4 text-gray-700">23.8</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Hybrid (LangGraph W=512)</td>
<td class="px-6 py-4 text-gray-700">44.9</td>
<td class="px-6 py-4 text-gray-700">24.2</td>
<td class="px-6 py-4 text-gray-700">17.6</td>
<td class="px-6 py-4 text-gray-700">21.7</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Accumulation (OpenAI SDK W=8192)</td>
<td class="px-6 py-4 text-gray-700">33.9</td>
<td class="px-6 py-4 text-gray-700">20.7</td>
<td class="px-6 py-4 text-gray-700">27.5</td>
<td class="px-6 py-4 text-gray-700">20.6</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Accumulation (OpenAI SDK W=50)</td>
<td class="px-6 py-4 text-gray-700">8.4</td>
<td class="px-6 py-4 text-gray-700">1.2</td>
<td class="px-6 py-4 text-gray-700">0.0</td>
<td class="px-6 py-4 text-gray-700">6.1</td>
</tr>
</tbody>
</table>
</div>
<div class="bg-white border border-gray-300 rounded-lg p-6">
<h3 class="text-xl font-semibold text-gray-900 mb-3">Key Takeaways</h3>
<ul class="space-y-2 text-gray-700">
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Retrieval enables stable recall:</strong> Retrieval-only designs dominate on factual recall (AR) and long-range understanding (LRU), achieving 30.4 on LRU vs 27.5 for large accumulation.</span>
</li>
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Hybrid works best for learning:</strong> Hybrid retrieval-accumulation achieves highest TTL (28.9) by anchoring learning to retrieved signals rather than raw prompt growth.</span>
</li>
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Accumulation scales poorly:</strong> Runtime grows rapidly with context window as full histories are replayed for each query. Retrieval avoids repeated processing.</span>
</li>
</ul>
</div>
</div>
</section>
<!-- Planning Effects -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-white border-b border-gray-200">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-gray-900 mb-6">Planning Interface Effects</h2>
<p class="text-lg text-gray-700 mb-8 leading-relaxed">
Schema-constrained planning reduces accuracy and introduces high failure rates. Free-form planning preserves or improves accuracy with minimal overhead.
</p>
<div class="bg-gray-50 border border-gray-300 rounded-lg overflow-hidden mb-8">
<table class="w-full text-sm">
<thead class="bg-gray-100">
<tr>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Planning Interface</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Accuracy Impact</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Formatting Failures</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Runtime Overhead</th>
</tr>
</thead>
<tbody class="divide-y divide-gray-200">
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">No Plan (Direct)</td>
<td class="px-6 py-4 text-gray-700">Baseline</td>
<td class="px-6 py-4 text-gray-700">0%</td>
<td class="px-6 py-4 text-gray-700">1×</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Schema-constrained (Crew-Plan)</td>
<td class="px-6 py-4 text-gray-700">-30% to -50%</td>
<td class="px-6 py-4 text-gray-700">Up to 84.7%</td>
<td class="px-6 py-4 text-gray-700">7.4× to 31.3×</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Free-form (Direct-LLM-Plan)</td>
<td class="px-6 py-4 text-gray-700">+15% to -5%</td>
<td class="px-6 py-4 text-gray-700">0%</td>
<td class="px-6 py-4 text-gray-700">1.2× to 6.6×</td>
</tr>
</tbody>
</table>
</div>
<div class="bg-white border border-gray-300 rounded-lg p-6">
<h3 class="text-xl font-semibold text-gray-900 mb-3">Key Takeaway</h3>
<p class="text-gray-700 leading-relaxed">
Planning outcomes are driven primarily by interface design, not LLM planning ability. Schema-constrained planning introduces large formatting failure rates (up to 84.7%) and high orchestration overhead (up to 31×). Free-form planning preserves accuracy with minimal overhead. Planning should be implemented as a permissive stage that tolerates variability in plan text.
</p>
</div>
</div>
</section>
<!-- Specialization -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-gray-50 border-b border-gray-200">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-gray-900 mb-6">Agent Specialization</h2>
<p class="text-lg text-gray-700 mb-8 leading-relaxed">
Specialization is governed by how frameworks inject task-specific reasoning structure, not role identity alone. Expert-guided conditioning improves F1 scores by 58 points.
</p>
<div class="bg-white border border-gray-300 rounded-lg overflow-hidden mb-8">
<table class="w-full text-sm">
<thead class="bg-gray-100">
<tr>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Conditioning Strategy</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Multiclass F1</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Impact</th>
</tr>
</thead>
<tbody class="divide-y divide-gray-200">
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">No Role</td>
<td class="px-6 py-4 text-gray-700">41.9</td>
<td class="px-6 py-4 text-gray-700">Baseline</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Role-based prompting</td>
<td class="px-6 py-4 text-gray-700">41.9-42.1</td>
<td class="px-6 py-4 text-gray-700">No improvement</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Planning-based conditioning</td>
<td class="px-6 py-4 text-gray-700">42.0</td>
<td class="px-6 py-4 text-gray-700">No improvement</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Expert-guided conditioning</td>
<td class="px-6 py-4 text-gray-700">96.0-100.0</td>
<td class="px-6 py-4 text-gray-700">+54 to +58 points</td>
</tr>
</tbody>
</table>
</div>
<div class="bg-white border border-gray-300 rounded-lg p-6">
<h3 class="text-xl font-semibold text-gray-900 mb-3">Key Takeaway</h3>
<p class="text-gray-700 leading-relaxed">
Role labels and generic planning interfaces fail to activate domain knowledge. Explicit procedural instructions impose structured solution workflows that reliably improve performance. Specialization should be implemented through reasoning procedures embedded in the framework, not through role naming or lightweight prompt modifications.
</p>
</div>
</div>
</section>
<!-- Coordination -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-white border-b border-gray-200">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-gray-900 mb-6">Coordination and Scaling</h2>
<p class="text-lg text-gray-700 mb-8 leading-relaxed">
Coordination performance is governed by the match between task structure and communication geometry. Topology choice becomes critical at scale.
</p>
<div class="bg-gray-50 border border-gray-300 rounded-lg overflow-hidden mb-8">
<table class="w-full text-sm">
<thead class="bg-gray-100">
<tr>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Topology</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Coloring (n=100)</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Consensus (n=100)</th>
<th class="px-6 py-3 text-left font-semibold text-gray-900 border-b border-gray-300">Rounds</th>
</tr>
</thead>
<tbody class="divide-y divide-gray-200">
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Scale-Free</td>
<td class="px-6 py-4 text-gray-700">97%</td>
<td class="px-6 py-4 text-gray-700">Failed</td>
<td class="px-6 py-4 text-gray-700">11</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Small-World</td>
<td class="px-6 py-4 text-gray-700">98%</td>
<td class="px-6 py-4 text-gray-700">Failed</td>
<td class="px-6 py-4 text-gray-700">15</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Fully Connected</td>
<td class="px-6 py-4 text-gray-700">98%</td>
<td class="px-6 py-4 text-gray-700">100%</td>
<td class="px-6 py-4 text-gray-700">3</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Delaunay</td>
<td class="px-6 py-4 text-gray-700">81%</td>
<td class="px-6 py-4 text-gray-700">Failed</td>
<td class="px-6 py-4 text-gray-700">19</td>
</tr>
<tr>
<td class="px-6 py-4 text-gray-900 font-medium">Sequential</td>
<td class="px-6 py-4 text-gray-700">Failed (>40 rounds)</td>
<td class="px-6 py-4 text-gray-700">Failed</td>
<td class="px-6 py-4 text-gray-700">>40</td>
</tr>
</tbody>
</table>
</div>
<div class="bg-white border border-gray-300 rounded-lg p-6">
<h3 class="text-xl font-semibold text-gray-900 mb-3">Key Takeaways</h3>
<ul class="space-y-2 text-gray-700">
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Local coordination:</strong> Sparse topologies (scale-free, small-world) achieve high success (97-98%) on local tasks like Coloring, converging in 11-15 rounds.</span>
</li>
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Global agreement:</strong> Only fully connected topologies succeed on Consensus (100%), converging in 3 rounds independent of network size. All sparse topologies fail despite higher runtime and token expenditure.</span>
</li>
<li class="flex items-start">
<span class="text-indigo-600 mr-2">•</span>
<span><strong>Sequential pipelines fail at scale:</strong> Sequential topologies exceed 40 rounds and fail entirely on large networks, showing that increasing interaction budgets alone does not improve outcomes.</span>
</li>
</ul>
</div>
</div>
</section>
<!-- Summary -->
<section class="py-20 px-4 sm:px-6 lg:px-8 bg-gray-50">
<div class="max-w-6xl mx-auto">
<h2 class="text-4xl font-bold text-center text-gray-900 mb-12">Summary</h2>
<div class="bg-white border border-gray-300 rounded-lg p-8">
<p class="text-lg text-gray-700 mb-6 leading-relaxed">
These results demonstrate that performance in multi-agent LLM systems is governed by framework architecture, not model quality alone. Architectural choices create order-of-magnitude differences in latency, accuracy, and coordination success.
</p>
<div class="grid md:grid-cols-2 gap-6 mt-8">
<div>
<h3 class="text-xl font-semibold text-gray-900 mb-4">Architecture > Model</h3>
<p class="text-gray-700 leading-relaxed">
Framework-level design choices can create 100× latency differences and 30% accuracy drops, even when using identical LLM models.
</p>
</div>
<div>
<h3 class="text-xl font-semibold text-gray-900 mb-4">Design Dimensions Matter</h3>
<p class="text-gray-700 leading-relaxed">
Orchestration, memory, planning, specialization, and coordination each independently drive performance. No single dimension can compensate for poor choices in others.
</p>
</div>
</div>
</div>
</div>
</section>
<!-- Footer -->
<footer class="bg-gray-900 text-gray-300 py-12 px-4 sm:px-6 lg:px-8">
<div class="max-w-7xl mx-auto">
<div class="grid md:grid-cols-4 gap-8 mb-8">
<div>
<h3 class="text-white font-bold text-lg mb-4">MAFBench</h3>
<p class="text-sm">A unified benchmark for evaluating multi-agent LLM frameworks.</p>
</div>
<div>
<h4 class="text-white font-semibold mb-4">Resources</h4>
<ul class="space-y-2 text-sm">
<li><a href="paper.html" class="hover:text-white">Paper</a></li>
<li><a href="mafbench.html" class="hover:text-white">Benchmark</a></li>
<li><a href="https://github.com/CoDS-GCS/MAFBench" target="_blank" class="hover:text-white">GitHub</a></li>
</ul>
</div>
<div>
<h4 class="text-white font-semibold mb-4">Learn</h4>
<ul class="space-y-2 text-sm">
<li><a href="architecture/index.html" class="hover:text-white">Architecture Guide</a></li>
<li><a href="results.html" class="hover:text-white">Results</a></li>
</ul>
</div>
<div>
<h4 class="text-white font-semibold mb-4">Connect</h4>
<ul class="space-y-2 text-sm">
<li><a href="about.html" class="hover:text-white">About</a></li>
<li><a href="https://github.com/CoDS-GCS/MAFBench" target="_blank" class="hover:text-white">GitHub</a></li>
</ul>
</div>
</div>
<div class="border-t border-gray-800 pt-8 text-center text-sm">
<p>© 2026 MAFBench. Research by Concordia University.</p>
</div>
</div>
</footer>
<script src="scripts/main.js"></script>
</body>
</html>