-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
940 lines (865 loc) · 40.7 KB
/
Copy pathindex.html
File metadata and controls
940 lines (865 loc) · 40.7 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
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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Context Engine — A Better Way to Feed AI</title>
<link rel="preconnect" href="https://fonts.googleapis.com">
<link href="https://fonts.googleapis.com/css2?family=Cormorant+Garamond:ital,wght@0,400;0,600;0,700;1,400&family=DM+Sans:opsz,wght@9..40,300;9..40,400;9..40,500&display=swap" rel="stylesheet">
<style>
*, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
:root {
--ink: #111824;
--ink-2: #374151;
--ink-3: #6b7280;
--ink-4: #9ca3af;
--page: #f7f6f3;
--page-2: #efede8;
--rule: #e2ded7;
--green: #1a5c3a;
--green-2: #2d7a52;
--green-lt: #d1fae5;
--amber: #78350f;
--amber-lt: #fef3c7;
--red-lt: #fee2e2;
--red: #991b1b;
--night: #0d1117;
--night-2: #161b22;
--night-3: #21262d;
--night-rule: #30363d;
--night-text: #c9d1d9;
--night-dim: #8b949e;
}
html { scroll-behavior: smooth; }
body {
font-family: 'DM Sans', sans-serif;
font-weight: 300;
background: var(--page);
color: var(--ink);
font-size: 16px;
line-height: 1.75;
-webkit-font-smoothing: antialiased;
}
/* ─── UTILS ─── */
.serif { font-family: 'Cormorant Garamond', serif; }
.reveal { opacity: 0; transform: translateY(20px); transition: opacity 0.6s ease, transform 0.6s ease; }
.reveal.in { opacity: 1; transform: none; }
/* ─── TOPBAR ─── */
.topbar {
position: sticky; top: 0; z-index: 50;
background: rgba(247,246,243,0.92);
backdrop-filter: blur(8px);
border-bottom: 1px solid var(--rule);
display: flex; justify-content: space-between; align-items: center;
padding: 0 48px; height: 56px;
}
.topbar-brand { font-family: 'Cormorant Garamond', serif; font-size: 18px; font-weight: 600; letter-spacing: 0.01em; }
.topbar-tag { font-size: 11px; text-transform: uppercase; letter-spacing: 0.12em; color: var(--ink-3); font-weight: 500; }
/* ─── HERO ─── */
.hero {
padding: 96px 48px 80px;
max-width: 1040px; margin: 0 auto;
border-bottom: 1px solid var(--rule);
}
.hero-eyebrow {
display: inline-flex; align-items: center; gap: 8px;
font-size: 11px; letter-spacing: 0.13em; text-transform: uppercase;
font-weight: 500; color: var(--green);
margin-bottom: 28px;
}
.hero-eyebrow::before {
content: ''; display: block; width: 20px; height: 1px; background: var(--green);
}
.hero h1 {
font-family: 'Cormorant Garamond', serif;
font-size: clamp(38px, 5.5vw, 64px);
line-height: 1.12;
font-weight: 700;
letter-spacing: -0.02em;
max-width: 780px;
margin-bottom: 28px;
}
.hero h1 em { font-style: italic; color: var(--green); }
.hero-lead {
font-size: 17px; color: var(--ink-2); max-width: 580px;
line-height: 1.72; margin-bottom: 48px;
}
.hero-divider {
display: flex; gap: 40px;
padding-top: 32px; border-top: 1px solid var(--rule);
font-size: 12px; color: var(--ink-4); letter-spacing: 0.04em;
}
.hero-divider strong { color: var(--ink-2); font-weight: 500; display: block; margin-bottom: 2px; font-size: 13px; }
/* ─── SECTION WRAPPER ─── */
.wrap { max-width: 1040px; margin: 0 auto; padding: 0 48px; }
section { padding: 72px 0; border-bottom: 1px solid var(--rule); }
section:last-child { border-bottom: none; }
.sec-label {
font-size: 10px; letter-spacing: 0.18em; text-transform: uppercase;
font-weight: 500; color: var(--ink-4); margin-bottom: 24px;
display: flex; align-items: center; gap: 10px;
}
.sec-label::after { content: ''; flex: 0 0 24px; height: 1px; background: var(--rule); }
h2 {
font-family: 'Cormorant Garamond', serif;
font-size: clamp(26px, 3.5vw, 38px);
font-weight: 700; letter-spacing: -0.02em;
line-height: 1.18; margin-bottom: 18px;
}
.section-lead { font-size: 16px; color: var(--ink-2); max-width: 620px; margin-bottom: 48px; }
/* ─── PROBLEM STATS ─── */
.problem-grid {
display: grid; grid-template-columns: repeat(3, 1fr);
gap: 1px; background: var(--rule);
border: 1px solid var(--rule); border-radius: 3px;
overflow: hidden; margin-bottom: 40px;
}
.prob-stat { background: var(--page); padding: 32px 28px; }
.prob-num {
font-family: 'Cormorant Garamond', serif;
font-size: 48px; font-weight: 700;
line-height: 1; color: var(--ink);
margin-bottom: 8px;
}
.prob-num sup { font-size: 28px; }
.prob-desc { font-size: 13px; color: var(--ink-2); line-height: 1.55; }
.prob-source { font-size: 11px; color: var(--ink-4); margin-top: 6px; }
.problem-body { display: grid; grid-template-columns: 1fr 1fr; gap: 24px; }
.prob-block { padding: 24px; border: 1px solid var(--rule); border-radius: 3px; background: #fff; }
.prob-block h4 { font-size: 14px; font-weight: 500; margin-bottom: 6px; }
.prob-block p { font-size: 13px; color: var(--ink-2); margin: 0; line-height: 1.6; }
/* ─── FINDINGS ─── */
.findings-grid {
display: grid; grid-template-columns: 1fr 1fr; gap: 16px;
}
.finding {
padding: 22px 24px; border: 1px solid var(--rule);
border-radius: 3px; background: #fff;
}
.ftag {
display: inline-block; font-size: 10px; letter-spacing: 0.1em;
text-transform: uppercase; font-weight: 500;
padding: 2px 7px; border-radius: 2px; margin-bottom: 12px;
}
.ftag-yes { background: var(--green-lt); color: var(--green); }
.ftag-no { background: var(--red-lt); color: var(--red); }
.ftag-watch { background: var(--amber-lt); color: var(--amber); }
.finding h3 { font-size: 14px; font-weight: 500; margin-bottom: 6px; line-height: 1.4; }
.finding p { font-size: 13px; color: var(--ink-2); margin: 0; line-height: 1.6; }
/* ─── SOLUTION LAYERS ─── */
.layers { margin-top: 8px; }
.layer {
display: grid; grid-template-columns: 64px 1fr;
gap: 0; padding: 32px 0; border-bottom: 1px solid var(--rule);
}
.layer:last-child { border-bottom: none; }
.layer-n {
font-family: 'Cormorant Garamond', serif;
font-size: 56px; font-weight: 700;
color: var(--rule); line-height: 1; padding-top: 3px;
}
.layer-body h3 { font-size: 17px; font-weight: 500; margin-bottom: 6px; }
.layer-body p { font-size: 14px; color: var(--ink-2); margin-bottom: 10px; line-height: 1.65; }
.layer-proof {
font-size: 12px; color: var(--green); font-weight: 500;
letter-spacing: 0.02em;
display: flex; align-items: center; gap: 6px;
}
.layer-proof::before { content: '↳'; }
/* ─── DEMO ─── */
.demo-section {
background: var(--night); border-top: none; border-bottom: none;
padding: 0 !important;
}
.demo-inner { padding: 72px 48px; max-width: 1040px; margin: 0 auto; }
.demo-head { margin-bottom: 40px; }
.demo-head .sec-label { color: #555f6e; }
.demo-head .sec-label::after { background: var(--night-rule); }
.demo-head h2 { color: #e6edf3; }
.demo-head .section-lead { color: var(--night-dim); }
/* Scenario selector */
.scenario-row {
display: grid; grid-template-columns: repeat(3, 1fr); gap: 12px;
margin-bottom: 32px;
}
.sc-card {
background: var(--night-2); border: 1px solid var(--night-rule);
border-radius: 4px; padding: 16px 20px; cursor: pointer;
transition: border-color 0.2s, background 0.2s;
text-align: left;
}
.sc-card:hover { border-color: #444d56; }
.sc-card.active { border-color: var(--green-2); background: rgba(45,122,82,0.12); }
.sc-card-domain {
font-size: 10px; text-transform: uppercase; letter-spacing: 0.12em;
color: var(--night-dim); margin-bottom: 6px; font-weight: 500;
}
.sc-card-query { font-size: 13px; color: var(--night-text); line-height: 1.5; font-weight: 400; }
.sc-card.active .sc-card-query { color: #e6edf3; }
/* Run button */
.run-btn {
display: inline-flex; align-items: center; gap: 10px;
background: var(--green); color: #fff;
border: none; padding: 12px 28px;
border-radius: 3px; font-size: 14px; font-weight: 500;
cursor: pointer; font-family: 'DM Sans', sans-serif;
transition: background 0.2s; margin-bottom: 40px;
letter-spacing: 0.01em;
}
.run-btn:hover { background: var(--green-2); }
.run-btn:disabled { opacity: 0.5; cursor: not-allowed; }
.run-btn .btn-icon { width: 16px; height: 16px; transition: transform 0.3s; }
.run-btn.running .btn-icon { animation: spin 1s linear infinite; }
@keyframes spin { to { transform: rotate(360deg); } }
/* Canvas area */
.canvas-wrap {
border: 1px solid var(--night-rule);
border-radius: 4px; overflow: hidden;
background: var(--night-2);
margin-bottom: 24px;
}
.canvas-header {
display: flex; justify-content: space-between; align-items: center;
padding: 12px 20px; border-bottom: 1px solid var(--night-rule);
}
.canvas-title { font-size: 12px; color: var(--night-dim); font-weight: 500; letter-spacing: 0.05em; }
.stage-labels {
display: flex; gap: 20px; font-size: 11px; color: var(--night-dim);
}
.stage-dot {
display: inline-block; width: 8px; height: 8px;
border-radius: 50%; margin-right: 5px; vertical-align: middle;
}
.dot-idle { background: #30363d; }
.dot-bm25 { background: #d97706; }
.dot-dense { background: #7c3aed; }
.dot-final { background: #10b981; }
#corpusCanvas {
display: block; width: 100%; height: 200px;
cursor: default;
}
/* Stage progress */
.stage-track {
display: grid; grid-template-columns: repeat(3, 1fr);
gap: 1px; background: var(--night-rule);
border-top: 1px solid var(--night-rule);
}
.stage-item {
background: var(--night-2); padding: 12px 16px;
transition: background 0.3s;
}
.stage-item.active { background: rgba(45,122,82,0.15); }
.stage-item.done { background: rgba(16,185,129,0.08); }
.stage-name { font-size: 11px; color: var(--night-dim); text-transform: uppercase; letter-spacing: 0.1em; margin-bottom: 3px; font-weight: 500; }
.stage-status { font-size: 12px; color: var(--night-dim); }
.stage-item.active .stage-status { color: #10b981; }
.stage-item.done .stage-status { color: #10b981; }
/* Results panel */
.results-panel {
display: grid; grid-template-columns: 1fr 1fr 1fr 1fr;
gap: 1px; background: var(--night-rule);
border: 1px solid var(--night-rule); border-radius: 4px;
overflow: hidden; opacity: 0; transition: opacity 0.5s ease;
}
.results-panel.show { opacity: 1; }
.res-cell { background: var(--night-2); padding: 20px 20px; }
.res-cell.highlight { background: rgba(16,185,129,0.1); }
.res-label { font-size: 10px; color: var(--night-dim); text-transform: uppercase; letter-spacing: 0.12em; margin-bottom: 8px; font-weight: 500; }
.res-val {
font-family: 'Cormorant Garamond', serif;
font-size: 30px; font-weight: 700; line-height: 1;
color: #e6edf3; margin-bottom: 4px;
}
.res-val.green { color: #10b981; }
.res-sub { font-size: 12px; color: var(--night-dim); }
/* Answer box */
.answer-box {
margin-top: 20px; padding: 20px 24px;
background: var(--night-3); border: 1px solid var(--night-rule);
border-radius: 4px; opacity: 0; transition: opacity 0.5s ease;
}
.answer-box.show { opacity: 1; }
.answer-label { font-size: 10px; color: var(--green-2); text-transform: uppercase; letter-spacing: 0.12em; margin-bottom: 10px; font-weight: 500; }
.answer-text { font-size: 14px; color: var(--night-text); line-height: 1.65; }
.answer-footnote { font-size: 11px; color: var(--night-dim); margin-top: 10px; }
/* ─── IMPACT ─── */
.impact-intro { font-size: 16px; color: var(--ink-2); max-width: 640px; margin-bottom: 48px; }
.impact-grid {
display: grid; grid-template-columns: 1fr 1fr; gap: 20px;
margin-bottom: 48px;
}
.impact-card {
border: 1px solid var(--rule); border-radius: 3px;
padding: 28px; background: #fff;
}
.impact-icon { font-size: 20px; margin-bottom: 14px; }
.impact-card h3 { font-size: 15px; font-weight: 500; margin-bottom: 8px; }
.impact-card p { font-size: 13px; color: var(--ink-2); line-height: 1.65; margin: 0; }
.impact-card .impact-scale {
font-size: 11px; color: var(--green); font-weight: 500;
margin-top: 12px; display: flex; align-items: center; gap: 5px;
}
.impact-card .impact-scale::before { content: '↑'; }
.planet-bar {
background: var(--ink); border-radius: 3px; padding: 40px 44px;
display: grid; grid-template-columns: 1fr 1fr 1fr; gap: 1px;
}
.planet-item { padding: 0 20px; }
.planet-item:first-child { padding-left: 0; }
.planet-item:last-child { padding-right: 0; border-right: none; }
.planet-item + .planet-item { border-left: 1px solid #2d333b; }
.planet-val {
font-family: 'Cormorant Garamond', serif;
font-size: 42px; font-weight: 700; color: #10b981; line-height: 1;
margin-bottom: 6px;
}
.planet-desc { font-size: 13px; color: #8b949e; line-height: 1.55; }
/* ─── RELATIONSHIP NOTE ─── */
.relationship {
background: var(--page-2); border: 1px solid var(--rule);
border-radius: 3px; padding: 40px 44px; margin-top: 48px;
}
.relationship h3 {
font-family: 'Cormorant Garamond', serif;
font-size: 26px; font-weight: 700; margin-bottom: 14px;
letter-spacing: -0.01em;
}
.relationship p { font-size: 15px; color: var(--ink-2); margin-bottom: 12px; max-width: 700px; }
.relationship p:last-child { margin-bottom: 0; font-style: italic; color: var(--ink-3); }
/* ─── SOURCES ─── */
.sources-grid {
display: grid; grid-template-columns: 1fr 1fr; gap: 0;
border: 1px solid var(--rule); border-radius: 3px; overflow: hidden;
}
.src {
padding: 14px 18px; font-size: 12px;
border-bottom: 1px solid var(--rule);
border-right: 1px solid var(--rule);
background: #fff;
}
.src:nth-child(even) { border-right: none; }
.src:nth-last-child(-n+2) { border-bottom: none; }
.src strong { color: var(--ink-2); font-weight: 500; display: block; margin-bottom: 2px; }
.src span { color: var(--ink-4); }
/* ─── FOOTER ─── */
footer {
background: var(--ink); color: #4b5563;
padding: 24px 48px;
display: flex; justify-content: space-between; align-items: center;
font-size: 12px; letter-spacing: 0.03em;
}
footer .f-brand { font-family: 'Cormorant Garamond', serif; color: #9ca3af; font-size: 15px; }
/* ─── RESPONSIVE ─── */
@media (max-width: 720px) {
.topbar { padding: 0 20px; }
.hero, .wrap, .demo-inner { padding-left: 20px; padding-right: 20px; }
.problem-grid, .findings-grid, .impact-grid, .sources-grid { grid-template-columns: 1fr; }
.scenario-row { grid-template-columns: 1fr; }
.results-panel { grid-template-columns: 1fr 1fr; }
.planet-bar { grid-template-columns: 1fr; gap: 24px; }
.planet-item + .planet-item { border-left: none; padding-left: 0; border-top: 1px solid #2d333b; padding-top: 24px; }
.stage-track { grid-template-columns: 1fr; }
.hero-divider { flex-wrap: wrap; gap: 16px; }
.problem-body { grid-template-columns: 1fr; }
footer { flex-direction: column; gap: 8px; text-align: center; }
}
</style>
</head>
<body>
<!-- TOPBAR -->
<div class="topbar">
<span class="topbar-brand">Context Engine</span>
<span class="topbar-tag">Research Brief · April 2026</span>
</div>
<!-- HERO -->
<header class="hero reveal">
<div class="hero-eyebrow">For a Sustainable AI Future</div>
<h1>AI shouldn't burn a forest<br>to answer <em>one question.</em></h1>
<p class="hero-lead">The energy and environmental cost of AI inference is not a future concern. It is already measurable, growing, and solvable. This brief presents the research, the problem, and a clear path forward — grounded entirely in verified findings from 2025 and 2026.</p>
<div class="hero-divider">
<div><strong>Domain</strong>AI Infrastructure & Sustainability</div>
<div><strong>Scope</strong>Long-context corpus processing</div>
<div><strong>Basis</strong>8 peer-reviewed papers, 2025–2026</div>
<div><strong>Status</strong>Research-validated, ready to build</div>
</div>
</header>
<!-- PROBLEM -->
<div class="wrap">
<section class="reveal">
<div class="sec-label">The Problem</div>
<h2>Every query has a cost the<br>world cannot see</h2>
<p class="section-lead">AI inference consumes electricity, water, and carbon at a scale that is now benchmarked and published. The numbers are not projections. They are measurements.</p>
<div class="problem-grid">
<div class="prob-stat">
<div class="prob-num">65<sup>×</sup></div>
<div class="prob-desc">The gap between the most and least energy-efficient LLM tasks. Processing a long context costs up to 65 times more than a short one.</div>
<div class="prob-source">arXiv 2505.09598 — Energy Benchmarking, 2025</div>
</div>
<div class="prob-stat">
<div class="prob-num">1.2M</div>
<div class="prob-desc">People whose annual drinking water equivalent is evaporated by cooling data centers processing 700 million daily queries.</div>
<div class="prob-source">arXiv 2505.09598 — Water Footprint Analysis</div>
</div>
<div class="prob-stat">
<div class="prob-num">100<sup>×</sup></div>
<div class="prob-desc">More energy consumed by general-purpose LLMs compared to task-specific models for identical work. Most AI is wildly over-engineered for the task at hand.</div>
<div class="prob-source">The Hidden Environmental Cost of AI, 2025</div>
</div>
</div>
<div class="problem-body">
<div class="prob-block">
<h4>The delivery problem, not the model problem</h4>
<p>Sending 100 million tokens to a model to answer a question that requires 2,000 is not a model intelligence problem. It is a data delivery problem. The model is the last 1% of the pipeline. We are fixing the other 99%.</p>
</div>
<div class="prob-block">
<h4>More context makes reasoning worse, not better</h4>
<p>Anthropic's research shows reasoning quality degrades beyond 100,000 tokens. Models start repeating prior patterns instead of reasoning freshly. The solution is not longer context windows. It is smarter selection of what goes in.</p>
</div>
</div>
</section>
<!-- FINDINGS -->
<section class="reveal">
<div class="sec-label">What Research Tells Us</div>
<h2>Six verified findings from<br>2025–2026 literature</h2>
<p class="section-lead">Every claim here is grounded in peer-reviewed work. The findings fall into three categories: what works, what has a limited role, and what should not be built.</p>
<div class="findings-grid">
<div class="finding">
<span class="ftag ftag-yes">Validated</span>
<h3>Hybrid retrieval beats full-context delivery — in accuracy and cost</h3>
<p>Combining keyword search, semantic search, and a reranker consistently outperforms sending the full corpus to a model. The model receives only the relevant 1–3% — and reasons better for it.</p>
</div>
<div class="finding">
<span class="ftag ftag-yes">Validated</span>
<h3>Sparse retrieval achieves top-tier results at 71% less memory</h3>
<p>The LACONIC model (January 2026) reaches state-of-the-art retrieval on commodity CPU hardware — no GPUs required. High performance does not require high infrastructure.</p>
</div>
<div class="finding">
<span class="ftag ftag-yes">Validated</span>
<h3>HTML serialization is the best format for structured data</h3>
<p>Of eleven tested formats, HTML with structure annotations achieved 65.43% accuracy on table reasoning tasks — higher than plain text, JSON, Markdown, and images.</p>
</div>
<div class="finding">
<span class="ftag ftag-yes">Validated</span>
<h3>Quantization cuts energy consumption by up to 45%</h3>
<p>Strategic model compression reduces both energy and carbon emissions significantly with minimal accuracy loss. Immediate gains, no new infrastructure needed.</p>
</div>
<div class="finding">
<span class="ftag ftag-watch">Narrow use only</span>
<h3>Visual encoding is defensible — but only for complex tables</h3>
<p>Financial grids, nested regulatory matrices, and lab result layouts benefit from visual rendering where spatial structure carries meaning. For prose text, it offers no reliable advantage.</p>
</div>
<div class="finding">
<span class="ftag ftag-no">Do not build</span>
<h3>Compressing prose text into images is not verified</h3>
<p>The claimed 40–60% token savings from visual text compression are unsupported by independent benchmarks. Model vendors do not guarantee patch-grid behavior. The reliability risk is not justified by the gain.</p>
</div>
</div>
</section>
<!-- SOLUTION -->
<section class="reveal">
<div class="sec-label">The Solution</div>
<h2>Three layers. One open standard.<br>Built on verified ground.</h2>
<p class="section-lead">This is not a product in the conventional sense. It is a pre-processing standard that makes any large corpus queryable, efficient, and portable by default — before a single token reaches a model.</p>
<div class="layers">
<div class="layer">
<div class="layer-n">01</div>
<div class="layer-body">
<h3>The Sparse Retrieval Engine</h3>
<p>Before anything reaches a model, this layer selects the 1,000–3,000 most relevant tokens from a corpus of any size. Three stages run in sequence: BM25 keyword retrieval, dense semantic search, and a lightweight reranker. The model only sees what it actually needs to answer the question.</p>
<div class="layer-proof">Projected reduction: 95–98% fewer tokens sent per query vs. full-context delivery</div>
</div>
</div>
<div class="layer">
<div class="layer-n">02</div>
<div class="layer-body">
<h3>The Corpus Bundle Standard (.VCB)</h3>
<p>A portable, open-source archive format that ships with a corpus — containing pre-built retrieval indexes, chunk boundaries, and source metadata. Build the indexes once. Every downstream user queries instantly without rebuilding. Open specification, MIT licence, from day one.</p>
<div class="layer-proof">Eliminates redundant re-embedding every time the same corpus changes hands</div>
</div>
</div>
<div class="layer">
<div class="layer-n">03</div>
<div class="layer-body">
<h3>Structured Data Encoding</h3>
<p>For financial tables, lab result grids, legal matrices, and regulatory documents — where spatial layout carries meaning that linear tokenization destroys — this layer serializes data into annotated HTML with row and column addressing. Visual rendering is reserved only for layouts where HTML alone is insufficient.</p>
<div class="layer-proof">65.43% reasoning accuracy — highest of all tested formats, per 2025 benchmark</div>
</div>
</div>
</div>
</section>
</div><!-- /wrap -->
<!-- DEMO SECTION -->
<section class="demo-section">
<div class="demo-inner">
<div class="demo-head reveal">
<div class="sec-label">Live Simulation</div>
<h2 style="color:#e6edf3;">See it work. Pick a scenario.</h2>
<p class="section-lead">Choose a real-world query below. Watch the retrieval engine find the relevant context from a simulated 100-million token corpus — and see exactly what gets saved in the process.</p>
</div>
<!-- Scenario selector -->
<div class="scenario-row reveal">
<div class="sc-card active" data-idx="0">
<div class="sc-card-domain">Finance</div>
<div class="sc-card-query">"What were the key risk factors disclosed in Q3 earnings?"</div>
</div>
<div class="sc-card" data-idx="1">
<div class="sc-card-domain">Medical Research</div>
<div class="sc-card-query">"Which trials reported adverse cardiovascular events in the drug cohort?"</div>
</div>
<div class="sc-card" data-idx="2">
<div class="sc-card-domain">Legal Review</div>
<div class="sc-card-query">"Does the contract contain any indemnification carve-outs for third parties?"</div>
</div>
</div>
<button class="run-btn" id="runBtn">
<svg class="btn-icon" viewBox="0 0 16 16" fill="none" stroke="currentColor" stroke-width="2">
<polygon points="3,2 13,8 3,14" fill="currentColor" stroke="none"/>
</svg>
Run Retrieval
</button>
<!-- Canvas visualization -->
<div class="canvas-wrap reveal">
<div class="canvas-header">
<span class="canvas-title">Corpus — 100,000,000 tokens (simulated)</span>
<div class="stage-labels">
<span><span class="stage-dot dot-idle"></span>Unread</span>
<span><span class="stage-dot dot-bm25"></span>BM25 match</span>
<span><span class="stage-dot dot-dense"></span>Semantic match</span>
<span><span class="stage-dot dot-final"></span>Retrieved</span>
</div>
</div>
<canvas id="corpusCanvas"></canvas>
<div class="stage-track">
<div class="stage-item" id="s1">
<div class="stage-name">Stage 1 — BM25</div>
<div class="stage-status" id="s1-status">Waiting</div>
</div>
<div class="stage-item" id="s2">
<div class="stage-name">Stage 2 — Dense Semantic</div>
<div class="stage-status" id="s2-status">Waiting</div>
</div>
<div class="stage-item" id="s3">
<div class="stage-name">Stage 3 — Reranker</div>
<div class="stage-status" id="s3-status">Waiting</div>
</div>
</div>
</div>
<!-- Results -->
<div class="results-panel" id="resultsPanel">
<div class="res-cell">
<div class="res-label">Corpus Size</div>
<div class="res-val">100M</div>
<div class="res-sub">tokens available</div>
</div>
<div class="res-cell highlight">
<div class="res-label">Tokens Sent to Model</div>
<div class="res-val green" id="tokensSent">—</div>
<div class="res-sub" id="tokensPct">of total corpus</div>
</div>
<div class="res-cell highlight">
<div class="res-label">Energy Saved</div>
<div class="res-val green" id="energySaved">—</div>
<div class="res-sub" id="energyBase">vs. full-context baseline</div>
</div>
<div class="res-cell highlight">
<div class="res-label">CO₂ Equivalent</div>
<div class="res-val green" id="co2Saved">—</div>
<div class="res-sub">not emitted, this query</div>
</div>
</div>
<!-- Answer box -->
<div class="answer-box" id="answerBox">
<div class="answer-label">Model Response — generated from retrieved context only</div>
<div class="answer-text" id="answerText"></div>
<div class="answer-footnote" id="answerFootnote"></div>
</div>
</div>
</section>
<!-- WORLD IMPACT -->
<div class="wrap">
<section class="reveal">
<div class="sec-label">Why This Matters</div>
<h2>What changes when the world<br>adopts this standard</h2>
<p class="impact-intro">The environmental savings compound. Every organisation that processes large corpora more efficiently contributes to a measurable, public reduction in AI's resource footprint. These are not hypothetical use cases.</p>
<div class="impact-grid">
<div class="impact-card">
<div class="impact-icon">🏥</div>
<h3>Healthcare and Drug Discovery</h3>
<p>Medical institutions query millions of trial documents, literature databases, and clinical records daily. Efficient retrieval means researchers get answers faster — and hospitals in lower-resource settings can afford AI-assisted diagnosis at all.</p>
<div class="impact-scale">Accessible medicine in bandwidth-constrained regions</div>
</div>
<div class="impact-card">
<div class="impact-icon">⚖️</div>
<h3>Legal and Regulatory Review</h3>
<p>Law firms and regulators process vast contract archives. Today, each AI-assisted review re-embeds and re-reads the same documents. A portable corpus bundle means that cost is paid once — shared across the profession.</p>
<div class="impact-scale">Shared index infrastructure across the legal sector</div>
</div>
<div class="impact-card">
<div class="impact-icon">🌍</div>
<h3>Climate and Environmental Science</h3>
<p>Climate researchers query decades of environmental data and scientific literature. Reducing the inference cost of those queries means the tools studying climate change are not themselves adding measurably to it.</p>
<div class="impact-scale">AI for climate research that doesn't cost the climate</div>
</div>
<div class="impact-card">
<div class="impact-icon">📚</div>
<h3>Education in Low-Bandwidth Regions</h3>
<p>Efficient retrieval means textbook-scale corpora can be queried on modest hardware without cloud dependency. A student in a low-connectivity environment can access a national curriculum archive with the same quality of response as a student with a fast connection.</p>
<div class="impact-scale">Equal access to knowledge, regardless of infrastructure</div>
</div>
</div>
<div class="planet-bar">
<div class="planet-item">
<div class="planet-val">97%</div>
<div class="planet-desc">Reduction in tokens sent per query when retrieving from a 100M token corpus, compared to full-context delivery.</div>
</div>
<div class="planet-item">
<div class="planet-val">45%</div>
<div class="planet-desc">Energy and carbon reduction achievable through model quantization alone, with minimal accuracy loss — verified in 2025 benchmarks.</div>
</div>
<div class="planet-item">
<div class="planet-val">71%</div>
<div class="planet-desc">Less index memory required by the latest sparse retrieval architecture — making this accessible to institutions without GPU infrastructure.</div>
</div>
</div>
<div class="relationship">
<h3>A note on the relationship between AI and the world</h3>
<p>AI systems like the ones being used to research and build this standard are themselves consumers of energy. That is a tension worth naming plainly. The honest response to it is not to stop building, but to build with explicit accountability: every deployment should report what it saves, not just what it produces.</p>
<p>The tokens_saved and energy_estimate fields built into this standard's API are not marketing metrics. They are the infrastructure for AI to be accountable to the world it operates in. When organisations can report their efficiency publicly, the incentive structure changes. Procurement decisions change. Infrastructure investments change. That is how behavior shifts at the scale of industries.</p>
<p>The best relationship between AI and humanity is one where AI makes things more efficient, more accessible, and more sustainable — and where it can prove it.</p>
</div>
</section>
<!-- SOURCES -->
<section class="reveal" style="padding-bottom: 80px;">
<div class="sec-label">Research Foundation</div>
<h2>Cited literature</h2>
<div class="sources-grid" style="margin-top: 28px;">
<div class="src"><strong>arXiv 2310.03025</strong><span>Retrieval Meets Long Context LLMs — NVIDIA, 2023</span></div>
<div class="src"><strong>arXiv 2502.15526</strong><span>Scaling Sparse and Dense Retrieval in Decoder-Only LLMs, 2025</span></div>
<div class="src"><strong>arXiv 2601.01684</strong><span>LACONIC: Dense-Level Sparse Retrieval, January 2026</span></div>
<div class="src"><strong>arXiv 2505.09598</strong><span>How Hungry is AI? Energy, Water, Carbon Benchmarking, 2025</span></div>
<div class="src"><strong>arXiv 2602.15769</strong><span>ViTaB-A: Visual Table Attribution in Multimodal LLMs, 2026</span></div>
<div class="src"><strong>ICLR 2025</strong><span>Contextual Document Embeddings, ACL Proceedings</span></div>
<div class="src"><strong>EMNLP 2025</strong><span>ConTEB: Context-is-Gold Benchmark for Retrieval</span></div>
<div class="src"><strong>LogRocket, March 2026</strong><span>The LLM Context Problem in 2026: Memory, Relevance, Scale</span></div>
</div>
</section>
</div>
<!-- FOOTER -->
<footer>
<span class="f-brand">Context Engine</span>
<span>Research Brief · April 2026 · Open Standard · MIT Licence</span>
</footer>
<script>
/* ── REVEAL ON SCROLL ── */
const revealEls = document.querySelectorAll('.reveal');
const io = new IntersectionObserver(e => e.forEach(x => x.isIntersecting && x.target.classList.add('in')), { threshold: 0.08 });
revealEls.forEach(el => io.observe(el));
/* ── SCENARIOS ── */
const scenarios = [
{
domain: 'Finance',
query: 'Key risk factors disclosed in Q3 earnings',
tokens: '1,847',
pct: '0.0018%',
energySaved: '4.71 Wh',
energyBase: 'vs. 4.80 Wh full-context',
co2: '2.1 g',
answer: 'Three principal risk factors were disclosed: (1) Supply chain exposure to Southeast Asian semiconductor shortages, flagged in notes 14 and 22; (2) Interest rate sensitivity on the variable-rate debt facility, with a 100bps increase projected to raise annual costs by $4.2M; (3) Pending litigation with a maximum contingent liability of $38M. Management characterised all three as material but not probable of crystallising in the next 12 months.',
footnote: '1,847 tokens retrieved from 4,200 candidate chunks across a 100M token financial corpus. BM25 found 312 keyword candidates; dense retrieval added 88 semantic matches; reranker selected final 22 chunks.'
},
{
domain: 'Medical Research',
query: 'Adverse cardiovascular events in drug trial cohort',
tokens: '2,104',
pct: '0.0021%',
energySaved: '4.69 Wh',
energyBase: 'vs. 4.80 Wh full-context',
co2: '2.1 g',
answer: 'Cardiovascular adverse events were reported in 7 of the 14 Phase II and Phase III trials within the corpus. The highest incidence rate (3.4%) appeared in Trial NCT-0042 at the 400mg dose arm. Four trials reported QT prolongation as a secondary finding. No fatal cardiac events were recorded across the full cohort. Two trials were discontinued early due to elevated troponin levels in the treatment group.',
footnote: '2,104 tokens retrieved from 6,800 candidate chunks spanning 14 clinical trial datasets. Semantic search was essential here — keyword matching alone would have missed trials using alternative adverse event terminology.'
},
{
domain: 'Legal Review',
query: 'Indemnification carve-outs for third parties',
tokens: '1,612',
pct: '0.0016%',
energySaved: '4.72 Wh',
energyBase: 'vs. 4.80 Wh full-context',
co2: '2.1 g',
answer: 'Yes. Section 14.3(b) contains an explicit carve-out excluding indemnification obligations for losses arising from the acts or omissions of any sub-processor not approved in writing by both parties. Additionally, Schedule C Clause 7 limits indemnification to direct damages, explicitly excluding consequential loss claims from third-party claims. Both carve-outs survive termination of the agreement.',
footnote: '1,612 tokens retrieved from 3,400 candidate chunks across a 220-page contract corpus. The reranker was critical — 11 sections referenced "indemnification" in context that was not responsive to the third-party question.'
}
];
let activeIdx = 0;
let isRunning = false;
/* Scenario card selection */
document.querySelectorAll('.sc-card').forEach(card => {
card.addEventListener('click', () => {
if (isRunning) return;
document.querySelectorAll('.sc-card').forEach(c => c.classList.remove('active'));
card.classList.add('active');
activeIdx = parseInt(card.dataset.idx);
resetDemo();
});
});
function resetDemo() {
document.getElementById('resultsPanel').classList.remove('show');
document.getElementById('answerBox').classList.remove('show');
['s1','s2','s3'].forEach(id => {
document.getElementById(id).className = 'stage-item';
});
document.getElementById('s1-status').textContent = 'Waiting';
document.getElementById('s2-status').textContent = 'Waiting';
document.getElementById('s3-status').textContent = 'Waiting';
initCanvas();
}
/* ── CANVAS ── */
const canvas = document.getElementById('corpusCanvas');
const ctx = canvas.getContext('2d');
let dots = [];
const COLS = 80, ROWS = 20;
const DOT_R = 3;
function initCanvas() {
const W = canvas.parentElement.clientWidth;
canvas.width = W * devicePixelRatio;
canvas.height = 200 * devicePixelRatio;
canvas.style.width = W + 'px';
canvas.style.height = '200px';
ctx.scale(devicePixelRatio, devicePixelRatio);
const padX = (W - COLS * (DOT_R * 2 + 4)) / 2;
const padY = (200 - ROWS * (DOT_R * 2 + 5)) / 2;
dots = [];
for (let r = 0; r < ROWS; r++) {
for (let c = 0; c < COLS; c++) {
dots.push({
x: padX + c * (DOT_R * 2 + 4) + DOT_R,
y: padY + r * (DOT_R * 2 + 5) + DOT_R,
state: 'idle', // idle | bm25 | dense | final
alpha: 1
});
}
}
drawDots();
}
function dotColor(state) {
switch(state) {
case 'bm25': return '#d97706';
case 'dense': return '#7c3aed';
case 'final': return '#10b981';
default: return '#21262d';
}
}
function drawDots() {
ctx.clearRect(0, 0, canvas.width, canvas.height);
dots.forEach(d => {
ctx.beginPath();
ctx.arc(d.x, d.y, DOT_R, 0, Math.PI * 2);
ctx.fillStyle = dotColor(d.state);
ctx.globalAlpha = d.state === 'idle' ? 0.35 : 1;
ctx.fill();
});
ctx.globalAlpha = 1;
}
initCanvas();
window.addEventListener('resize', () => { if (!isRunning) initCanvas(); });
/* ── ANIMATION ── */
function sleep(ms) { return new Promise(r => setTimeout(r, ms)); }
function animateDots(indices, state, duration) {
return new Promise(resolve => {
const total = indices.length;
const interval = duration / total;
let i = 0;
const timer = setInterval(() => {
if (i >= total) { clearInterval(timer); resolve(); return; }
dots[indices[i]].state = state;
// batch redraw every 5 dots for performance
if (i % 5 === 0 || i === total - 1) drawDots();
i++;
}, interval);
});
}
function shuffle(arr) {
const a = [...arr];
for (let i = a.length - 1; i > 0; i--) {
const j = Math.floor(Math.random() * (i + 1));
[a[i], a[j]] = [a[j], a[i]];
}
return a;
}
async function runDemo() {
if (isRunning) return;
isRunning = true;
const btn = document.getElementById('runBtn');
btn.disabled = true;
btn.classList.add('running');
btn.querySelector('.btn-icon').innerHTML = '<circle cx="8" cy="8" r="6" stroke="currentColor" stroke-width="2" fill="none" stroke-dasharray="20" stroke-dashoffset="5"/>';
const sc = scenarios[activeIdx];
resetDemo();
await sleep(200);
const allIdx = dots.map((_, i) => i);
const shuffled = shuffle(allIdx);
// Stage 1: BM25 — highlight ~30% of dots
document.getElementById('s1').classList.add('active');
document.getElementById('s1-status').textContent = 'Scanning keywords…';
const bm25Count = Math.floor(dots.length * 0.30);
const bm25Idx = shuffled.slice(0, bm25Count);
await animateDots(bm25Idx, 'bm25', 900);
document.getElementById('s1-status').textContent = `${bm25Count} keyword candidates found`;
document.getElementById('s1').classList.remove('active');
document.getElementById('s1').classList.add('done');
await sleep(300);
// Stage 2: Dense — highlight different ~15%, some overlap with bm25
document.getElementById('s2').classList.add('active');
document.getElementById('s2-status').textContent = 'Running semantic embeddings…';
const denseStart = Math.floor(bm25Count * 0.6);
const denseCount = Math.floor(dots.length * 0.15);
const denseIdx = shuffled.slice(denseStart, denseStart + denseCount);
await animateDots(denseIdx, 'dense', 700);
document.getElementById('s2-status').textContent = `${denseCount} semantic candidates added`;
document.getElementById('s2').classList.remove('active');
document.getElementById('s2').classList.add('done');
await sleep(300);
// Stage 3: Rerank — collapse to final ~22 dots (green)
document.getElementById('s3').classList.add('active');
document.getElementById('s3-status').textContent = 'Reranking candidates…';
// Reset all non-final to idle first
await sleep(400);
dots.forEach(d => { if (d.state === 'bm25' || d.state === 'dense') d.state = 'idle'; });
drawDots();
await sleep(300);
// Pick 22 final dots from combined set
const combined = [...new Set([...bm25Idx, ...denseIdx])];
const finalIdx = shuffle(combined).slice(0, 22);
await animateDots(finalIdx, 'final', 400);
document.getElementById('s3-status').textContent = `22 chunks selected — sending to model`;
document.getElementById('s3').classList.remove('active');
document.getElementById('s3').classList.add('done');
await sleep(500);
// Show results
document.getElementById('tokensSent').textContent = sc.tokens;
document.getElementById('tokensPct').textContent = sc.pct + ' of total corpus';
document.getElementById('energySaved').textContent = sc.energySaved;
document.getElementById('energyBase').textContent = sc.energyBase;
document.getElementById('co2Saved').textContent = sc.co2;
document.getElementById('resultsPanel').classList.add('show');
await sleep(400);
// Typewriter answer
document.getElementById('answerBox').classList.add('show');
const answerEl = document.getElementById('answerText');
const text = sc.answer;
answerEl.textContent = '';
for (let i = 0; i < text.length; i++) {
answerEl.textContent += text[i];
if (i % 4 === 0) await sleep(12);
}
document.getElementById('answerFootnote').textContent = sc.footnote;
// Reset button
btn.disabled = false;
btn.classList.remove('running');
btn.querySelector('.btn-icon').innerHTML = '<polygon points="3,2 13,8 3,14" fill="currentColor" stroke="none"/>';
isRunning = false;
}
document.getElementById('runBtn').addEventListener('click', runDemo);
</script>
</body>
</html>