-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathindex.html
More file actions
911 lines (864 loc) · 40.8 KB
/
Copy pathindex.html
File metadata and controls
911 lines (864 loc) · 40.8 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
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>PRISM — Divergent Paths, Unified Solutions</title>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&family=Sora:wght@400;600;700;800&family=JetBrains+Mono:wght@400;500&display=swap" rel="stylesheet">
<style>
:root {
/* ── 从图片提取的配色 ── */
--gold: #E8A020; /* 金黄 — 标题/Global Exploration */
--gold-lt: #FDF3E3; /* 金黄浅底 */
--gold-mid: #FAE0B0;
--green: #27AE60; /* 绿色 — 正确/Ground Truth */
--green-lt: #E8F8F0;
--green-mid: #A8E6C4;
--red: #E74C3C; /* 红色 — 错误/Limitation */
--red-lt: #FDECEA;
--red-mid: #F5B7B1;
--blue: #2980B9; /* 蓝色 — Synthesis/Judge */
--blue-lt: #EBF5FB;
--blue-mid: #AED6F1;
--purple: #8E44AD; /* 紫色 — Local Exploration */
--purple-lt: #F5EEF8;
--gray-dk: #2C3E50; /* 深色文字 */
--gray: #5D6D7E; /* 次要文字 */
--gray-lt: #F4F6F8; /* 浅灰背景 */
--border: #D5D8DC;
--white: #FFFFFF;
--bg: #F8F9FA; /* 页面背景 — 亮色学术风 */
--text: #1A252F;
--muted: #6C7A89;
}
*, *::before, *::after { margin:0; padding:0; box-sizing:border-box; }
html { scroll-behavior: smooth; }
body {
font-family: 'Inter', sans-serif;
background: var(--bg);
color: var(--text);
line-height: 1.7;
overflow-x: hidden;
}
/* ── Navbar ── */
nav {
position: fixed; top:0; left:0; right:0; z-index:200;
height: 58px;
display: flex; align-items: center; justify-content: space-between;
padding: 0 2.5rem;
background: rgba(255,255,255,0.95);
backdrop-filter: blur(16px);
border-bottom: 2px solid var(--gold);
box-shadow: 0 2px 12px rgba(232,160,32,.12);
}
.nav-brand {
display: flex; align-items: center; gap: .7rem;
font-family: 'Sora', sans-serif;
font-weight: 800; font-size: 1.2rem; letter-spacing: 3px;
color: var(--gold);
}
.nav-prism-icon { width: 26px; height: 26px; }
.nav-links { display: flex; gap: 1.6rem; }
.nav-links a {
color: var(--gray); text-decoration: none;
font-size: .82rem; font-weight: 600; letter-spacing: .4px;
transition: color .2s;
padding-bottom: 2px;
border-bottom: 2px solid transparent;
}
.nav-links a:hover { color: var(--gold); border-bottom-color: var(--gold); }
/* ── Hero ── */
.hero {
position: relative;
min-height: 100vh;
display: flex; align-items: center; justify-content: center;
text-align: center;
padding: 7rem 2rem 5rem;
background: linear-gradient(160deg, #FFFDF7 0%, #F8F9FA 50%, #EBF5FB 100%);
overflow: hidden;
}
/* 背景装饰 */
.hero::before {
content:''; position:absolute; top:-80px; right:-80px;
width:400px; height:400px; border-radius:50%;
background: radial-gradient(circle, rgba(232,160,32,.12) 0%, transparent 70%);
pointer-events:none;
}
.hero::after {
content:''; position:absolute; bottom:-60px; left:-60px;
width:320px; height:320px; border-radius:50%;
background: radial-gradient(circle, rgba(39,174,96,.1) 0%, transparent 70%);
pointer-events:none;
}
.hero-inner { max-width: 860px; position:relative; z-index:1; }
.hero-eyebrow {
display: inline-flex; align-items: center; gap: .55rem;
background: var(--gold-lt);
border: 1.5px solid var(--gold);
border-radius: 999px;
padding: .32rem 1.1rem;
font-size: .75rem; font-weight: 700;
color: var(--gold); letter-spacing: 1.2px; text-transform: uppercase;
margin-bottom: 1.2rem;
}
.pulse-dot {
width: 7px; height: 7px; border-radius: 50%;
background: var(--gold);
animation: pulse 2.2s ease-in-out infinite;
}
@keyframes pulse { 0%,100%{opacity:1;transform:scale(1);} 50%{opacity:.35;transform:scale(1.5);} }
.hero-venue {
display: inline-flex; align-items: center; gap: .4rem;
background: var(--blue-lt);
border: 1.5px solid var(--blue-mid);
border-radius: 999px;
padding: .28rem .9rem;
font-size: .72rem; font-weight: 600;
color: var(--blue); letter-spacing: .6px;
margin-bottom: 1.4rem;
}
.hero-title {
font-family: 'Sora', sans-serif;
font-size: clamp(2rem, 5.2vw, 3.4rem);
font-weight: 800; line-height: 1.14;
margin-bottom: .8rem;
color: var(--gray-dk);
}
.hero-title .accent { color: var(--gold); }
.hero-subtitle {
font-size: clamp(.9rem, 1.8vw, 1.05rem);
color: var(--muted);
max-width: 660px; margin: 0 auto 2.4rem;
}
.hero-subtitle em { color: var(--blue); font-style: normal; font-weight: 600; }
/* ── Stat chips ── */
.hero-stats {
display: flex; gap: 1rem; justify-content: center; flex-wrap: wrap;
margin-bottom: 2.4rem;
}
.stat-chip {
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 14px;
padding: .8rem 1.4rem;
text-align: center;
box-shadow: 0 2px 10px rgba(0,0,0,.06);
transition: transform .2s, box-shadow .2s;
}
.stat-chip:hover { transform: translateY(-3px); box-shadow: 0 6px 20px rgba(232,160,32,.15); border-color: var(--gold); }
.stat-chip .val {
font-family: 'Sora', sans-serif;
font-size: 1.6rem; font-weight: 800;
color: var(--gold); line-height: 1.2;
}
.stat-chip .val.green { color: var(--green); }
.stat-chip .val.blue { color: var(--blue); }
.stat-chip .lbl { font-size: .72rem; color: var(--muted); margin-top: .15rem; letter-spacing: .3px; }
/* ── Buttons ── */
.btn-row { display: flex; gap: .9rem; justify-content: center; flex-wrap: wrap; }
.btn {
display: inline-flex; align-items: center; gap: .5rem;
padding: .65rem 1.6rem; border-radius: 999px;
font-size: .86rem; font-weight: 600;
text-decoration: none; border: none; cursor: pointer;
transition: transform .22s, box-shadow .22s;
}
.btn svg { width:15px; height:15px; flex-shrink:0; }
.btn-primary {
background: var(--gold);
color: #fff;
box-shadow: 0 4px 14px rgba(232,160,32,.35);
}
.btn-primary:hover { transform: translateY(-2px); box-shadow: 0 8px 22px rgba(232,160,32,.45); }
.btn-outline {
background: var(--white);
border: 1.5px solid var(--border);
color: var(--gray-dk);
}
.btn-outline:hover { transform: translateY(-2px); border-color: var(--gold); color: var(--gold); }
/* ── Section scaffold ── */
section { padding: 5.5rem 2rem; }
.container { max-width: 940px; margin: 0 auto; }
.sec-head { text-align: center; margin-bottom: 3rem; }
.sec-label {
display: inline-block;
font-size: .72rem; font-weight: 700; letter-spacing: 2px;
text-transform: uppercase; color: var(--gold);
margin-bottom: .5rem;
}
.sec-title {
font-family: 'Sora', sans-serif;
font-size: clamp(1.6rem, 3.5vw, 2.2rem);
font-weight: 700; color: var(--gray-dk);
margin-bottom: .5rem;
}
.sec-desc { font-size: .92rem; color: var(--muted); max-width: 600px; margin: 0 auto; }
.divider-line { width: 50px; height: 3px; border-radius: 2px; background: var(--gold); margin: 1rem auto 0; }
.sep {
height: 1px;
background: linear-gradient(90deg, transparent 0%, var(--border) 30%, var(--border) 70%, transparent 100%);
margin: 0 3rem;
}
/* ── Section bg alternation ── */
.sec-white { background: var(--white); }
.sec-gray { background: var(--bg); }
.sec-gold { background: var(--gold-lt); }
/* ── TL;DR ── */
.tldr-wrap {
background: var(--white);
border: 1.5px solid var(--gold-mid);
border-left: 5px solid var(--gold);
border-radius: 14px;
padding: 2.2rem 2.6rem;
box-shadow: 0 4px 20px rgba(232,160,32,.08);
position: relative;
}
.tldr-wrap::before {
content: '"';
position: absolute; top: -.5rem; left: 1.5rem;
font-size: 7rem; font-family: 'Sora', serif;
color: rgba(232,160,32,.12); line-height:1; pointer-events:none;
}
.tldr-wrap p {
font-size: 1rem; line-height: 1.9;
color: var(--gray); position: relative; z-index:1;
}
.tldr-wrap p strong { color: var(--gray-dk); }
.tldr-wrap p .hl-gold { color: var(--gold); font-weight:700; }
.tldr-wrap p .hl-green { color: var(--green); font-weight:700; }
.tldr-wrap p .hl-blue { color: var(--blue); font-weight:700; }
.tldr-wrap p .hl-red { color: var(--red); font-weight:700; }
/* ── Motivation cards ── */
.motiv-grid {
display: grid; grid-template-columns: 1fr 1fr; gap: 1.4rem;
margin-bottom: 2.5rem;
}
.motiv-card {
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 16px; padding: 1.6rem 1.8rem;
box-shadow: 0 2px 12px rgba(0,0,0,.05);
transition: transform .22s, box-shadow .22s;
}
.motiv-card.lim1 { border-top: 4px solid var(--red); }
.motiv-card.lim2 { border-top: 4px solid var(--gold); }
.motiv-card:hover { transform: translateY(-4px); box-shadow: 0 10px 28px rgba(0,0,0,.1); }
.motiv-icon {
width: 42px; height: 42px; border-radius: 10px;
display: flex; align-items:center; justify-content:center;
margin-bottom: 1rem; font-size: 1.3rem;
}
.lim1 .motiv-icon { background: var(--red-lt); }
.lim2 .motiv-icon { background: var(--gold-lt); }
.motiv-card h4 { font-size: .97rem; font-weight: 700; margin-bottom: .5rem; }
.lim1 h4 { color: var(--red); }
.lim2 h4 { color: var(--gold); }
.motiv-card p { font-size: .86rem; color: var(--muted); line-height: 1.75; }
.motiv-card p strong { color: var(--gray-dk); }
/* ── Figure wrap ── */
.fig-wrap {
border-radius: 16px;
border: 1.5px solid var(--border);
background: var(--white);
overflow: hidden;
box-shadow: 0 4px 20px rgba(0,0,0,.07);
}
.fig-wrap img { width:100%; display:block; }
.fig-placeholder {
width: 100%; aspect-ratio: 16/7;
display: flex; flex-direction: column;
align-items: center; justify-content: center;
gap: .7rem; color: var(--muted); font-size: .85rem;
background: var(--gray-lt);
}
.fig-placeholder svg { width:36px; height:36px; opacity:.4; }
.fig-caption {
text-align: center; padding: .9rem 1.2rem;
font-size: .8rem; color: var(--muted);
border-top: 1px solid var(--border);
background: var(--gray-lt);
}
.fig-caption code {
background: var(--gold-lt); color: var(--gold);
padding: .1rem .35rem; border-radius: 4px;
font-family: 'JetBrains Mono', monospace; font-size:.78rem;
}
/* ── Pipeline ── */
.pipeline { display: flex; flex-direction: column; gap: 0; position: relative; }
.pipeline::before {
content: ''; position: absolute; left: 27px; top: 36px; bottom: 36px;
width: 3px;
background: linear-gradient(180deg, var(--gold) 0%, var(--green) 35%, var(--blue) 65%, var(--purple) 100%);
border-radius: 2px;
}
.pipe-step {
display: flex; gap: 1.4rem; align-items: flex-start;
padding: 1rem 0 1rem 0;
position: relative; transition: transform .2s;
}
.pipe-step:hover { transform: translateX(5px); }
.pipe-num {
min-width: 54px; height: 54px; border-radius: 50%;
display: flex; align-items: center; justify-content: center;
font-family: 'Sora', sans-serif;
font-weight: 800; font-size: .9rem; color: #fff;
position: relative; z-index:2; flex-shrink:0;
box-shadow: 0 4px 12px rgba(0,0,0,.15);
}
.pipe-num.c-gold { background: var(--gold); }
.pipe-num.c-green { background: var(--green); }
.pipe-num.c-blue { background: var(--blue); }
.pipe-num.c-purple { background: var(--purple); }
.pipe-body {
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 14px; padding: 1.3rem 1.6rem;
flex: 1;
box-shadow: 0 2px 10px rgba(0,0,0,.05);
transition: border-color .2s, box-shadow .2s;
}
.pipe-step:hover .pipe-body { box-shadow: 0 6px 20px rgba(0,0,0,.1); }
.pipe-step.s-gold:hover .pipe-body { border-color: var(--gold); }
.pipe-step.s-green:hover .pipe-body { border-color: var(--green); }
.pipe-step.s-blue:hover .pipe-body { border-color: var(--blue); }
.pipe-step.s-purple:hover .pipe-body { border-color: var(--purple); }
.pipe-body h4 {
font-size: .97rem; font-weight: 700;
color: var(--gray-dk); margin-bottom: .5rem;
display: flex; align-items: center; gap: .5rem; flex-wrap: wrap;
}
.tag {
font-size: .67rem; font-weight: 700; letter-spacing:.7px;
padding: .18rem .6rem; border-radius: 5px;
text-transform: uppercase;
}
.tag-gold { background: var(--gold-lt); color: var(--gold); border:1px solid var(--gold-mid); }
.tag-green { background: var(--green-lt); color: var(--green); border:1px solid var(--green-mid); }
.tag-blue { background: var(--blue-lt); color: var(--blue); border:1px solid var(--blue-mid); }
.tag-purple { background: var(--purple-lt); color: var(--purple); border:1px solid #D7BDE2; }
.pipe-body p { font-size: .87rem; color: var(--muted); line-height: 1.78; }
.pipe-body p strong { color: var(--gray-dk); }
.pipe-body p em { color: var(--blue); font-style:normal; font-weight:600; }
/* ── Formula block ── */
.formula-block {
margin-top: .9rem;
background: var(--gray-lt);
border: 1px solid var(--border);
border-left: 4px solid var(--gold);
border-radius: 8px; padding: .9rem 1.2rem;
font-family: 'JetBrains Mono', monospace;
font-size: .82rem; color: var(--gray-dk);
line-height: 1.9;
}
.formula-block .comment { color: var(--muted); font-style:italic; }
.formula-block .hl { color: var(--gold); font-weight:600; }
/* ── Prompt templates ── */
.prompt-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 1.2rem; margin-top: 2rem; }
.prompt-card {
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 14px; padding: 1.3rem 1.5rem;
font-size: .82rem;
box-shadow: 0 2px 8px rgba(0,0,0,.04);
}
.prompt-card h5 {
font-size: .78rem; font-weight: 700; letter-spacing: .8px;
text-transform: uppercase; color: var(--blue); margin-bottom: .8rem;
display: flex; align-items:center; gap:.4rem;
padding-bottom: .5rem;
border-bottom: 1.5px solid var(--blue-mid);
}
.prompt-field { margin-bottom: .6rem; }
.prompt-field .lbl { color: var(--gold); font-weight:700; font-size:.78rem; margin-bottom:.2rem; }
.prompt-field .val { color: var(--muted); font-size:.81rem; line-height:1.65; }
.prompt-field .val strong { color: var(--gray-dk); }
/* ── Results table ── */
.table-scroll { overflow-x: auto; border-radius: 14px; border: 1.5px solid var(--border); box-shadow: 0 2px 12px rgba(0,0,0,.06); }
table { width:100%; border-collapse:collapse; font-size:.875rem; }
thead tr { background: var(--gold-lt); }
th {
padding: .95rem 1.2rem; text-align:center;
font-weight:700; font-size:.76rem; letter-spacing:.6px;
text-transform:uppercase; color: var(--gold);
border-bottom: 2px solid var(--gold-mid);
white-space: nowrap;
}
th:first-child { text-align:left; }
td {
padding: .85rem 1.2rem; text-align:center;
color: var(--muted);
border-bottom: 1px solid var(--gray-lt);
transition: background .15s;
font-size: .85rem;
}
td:first-child { text-align:left; color:var(--gray-dk); font-weight:500; }
tr:last-child td { border-bottom:none; }
tbody tr:hover td { background: var(--gold-lt); }
.best { color: var(--green); font-weight:800; }
.ours td { background: rgba(39,174,96,.06); }
.ours td:first-child { color: var(--green); font-weight:700; }
.ours td:first-child::after { content: ' ★'; font-size:.72rem; color:var(--gold); }
.model-tag {
display: inline-block;
font-size: .68rem; padding: .1rem .45rem; border-radius: 4px;
background: var(--blue-lt); color: var(--blue);
font-family: 'JetBrains Mono', monospace;
margin-left: .3rem; vertical-align: middle;
border: 1px solid var(--blue-mid);
}
.tbd { color: var(--muted); font-style: italic; font-size: .8rem; }
.table-note { font-size:.78rem; color:var(--muted); margin-top:.9rem; text-align:center; }
/* ── Analysis RQs ── */
.rq-list { display: flex; flex-direction: column; gap: 1.6rem; }
.rq-item {
display: grid; grid-template-columns: auto 1fr;
gap: 1.4rem; align-items: start;
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 16px; padding: 1.6rem 1.8rem;
box-shadow: 0 2px 10px rgba(0,0,0,.05);
transition: border-color .22s, transform .22s, box-shadow .22s;
}
.rq-item:nth-child(1):hover { border-color: var(--gold); box-shadow: 0 8px 24px rgba(232,160,32,.12); }
.rq-item:nth-child(2):hover { border-color: var(--green); box-shadow: 0 8px 24px rgba(39,174,96,.12); }
.rq-item:nth-child(3):hover { border-color: var(--blue); box-shadow: 0 8px 24px rgba(41,128,185,.12); }
.rq-item:hover { transform: translateX(4px); }
.rq-badge {
min-width: 52px; height: 52px; border-radius: 12px;
display: flex; align-items:center; justify-content:center;
font-family: 'Sora', sans-serif;
font-weight: 800; font-size: .8rem; color:#fff;
letter-spacing:.5px; flex-shrink:0;
}
.rq-badge.b-gold { background: var(--gold); }
.rq-badge.b-green { background: var(--green); }
.rq-badge.b-blue { background: var(--blue); }
.rq-content h4 { font-size:.97rem; font-weight:700; margin-bottom:.4rem; color:var(--gray-dk); }
.rq-content p { font-size:.87rem; color:var(--muted); line-height:1.78; }
.rq-content p strong { color: var(--gray-dk); }
.rq-content p em { color: var(--blue); font-style:normal; font-weight:600; }
.rq-img-ph {
margin-top: 1rem; width:100%; aspect-ratio:16/5;
border-radius: 10px;
border: 1.5px dashed var(--border);
background: var(--gray-lt);
display:flex; align-items:center; justify-content:center;
font-size:.75rem; color:var(--muted);
}
/* ── Contribution cards ── */
.contrib-grid { display: grid; grid-template-columns: repeat(3, 1fr); gap: 1.4rem; }
.contrib-card {
background: var(--white);
border: 1.5px solid var(--border);
border-radius: 16px; padding: 1.6rem;
box-shadow: 0 2px 10px rgba(0,0,0,.05);
transition: border-color .22s, transform .22s, box-shadow .22s;
border-top: 4px solid var(--gold);
}
.contrib-card:nth-child(2) { border-top-color: var(--green); }
.contrib-card:nth-child(3) { border-top-color: var(--blue); }
.contrib-card:hover { transform: translateY(-4px); box-shadow: 0 10px 28px rgba(0,0,0,.1); }
.contrib-num {
font-family: 'Sora', sans-serif; font-size: 2rem; font-weight: 800;
color: var(--gold); line-height:1; margin-bottom: .7rem;
}
.contrib-card:nth-child(2) .contrib-num { color: var(--green); }
.contrib-card:nth-child(3) .contrib-num { color: var(--blue); }
.contrib-card h4 { font-size: .92rem; font-weight: 700; margin-bottom: .4rem; color: var(--gray-dk); }
.contrib-card p { font-size: .83rem; color: var(--muted); line-height: 1.68; }
/* ── Footer ── */
footer {
border-top: 2px solid var(--gold-mid);
padding: 2.2rem 2rem;
text-align:center;
font-size:.8rem; color:var(--muted);
background: var(--gold-lt);
}
footer a { color:var(--gold); text-decoration:none; font-weight:600; }
footer a:hover { text-decoration:underline; }
/* ── Scroll reveal ── */
.reveal { opacity:0; transform:translateY(22px); transition: opacity .65s ease, transform .65s ease; }
.reveal.visible { opacity:1; transform:none; }
/* ── Responsive ── */
@media(max-width:700px){
nav { padding:0 1.2rem; }
.nav-links { gap:1rem; }
.motiv-grid { grid-template-columns:1fr; }
.prompt-grid { grid-template-columns:1fr; }
.contrib-grid { grid-template-columns:1fr; }
.pipeline::before { left:24px; }
.pipe-num { min-width:48px; height:48px; }
.rq-item { grid-template-columns:1fr; }
.tldr-wrap { padding:1.6rem 1.4rem; }
.hero-stats { gap:.7rem; }
}
.btn-doi {
background: var(--blue);
color: #fff;
border-color: var(--blue);
box-shadow: 0 8px 22px rgba(41,128,185,.35);
}
</style>
</head>
<body>
<!-- ── Navbar ── -->
<nav>
<div class="nav-brand">
<svg class="nav-prism-icon" viewBox="0 0 28 28" fill="none">
<polygon points="14,2 26,24 2,24" fill="#E8A020" opacity=".9"/>
<polygon points="14,2 26,24 14,18" fill="#27AE60" opacity=".7"/>
<polygon points="14,18 26,24 2,24" fill="#2980B9" opacity=".5"/>
</svg>
PRISM
</div>
<div class="nav-links">
<a href="#overview">Overview</a>
<a href="#approach">Approach</a>
<a href="#results">Results</a>
<a href="#analysis">Analysis</a>
</div>
</nav>
<!-- ── Hero ── -->
<section class="hero">
<div class="hero-inner">
<!-- <div class="hero-eyebrow">
<span class="pulse-dot"></span>
Under Review · 2026
</div> -->
<!-- <div class="hero-venue">🏛 ASE 2026 · Munich, Germany · Oct 12–16</div> -->
<h1 class="hero-title">
<span class="accent">Prism</span>
</h1>
<p class="hero-subtitle">
A Multi-Solution Reasoning and Synthesis Framework for<br>
<em>Repository-Level Issue Resolution</em>
</p>
<div class="hero-stats">
<div class="stat-chip">
<div class="val">82.0%</div>
<div class="lbl">SWE-bench Verified<br>(Claude 4.6 Sonnet)</div>
</div>
<div class="stat-chip">
<div class="val">80.0%</div>
<div class="lbl">SWE-bench Verified<br>(DeepSeek-V3.2-Reasoner)</div>
</div>
<!-- <div class="stat-chip">
<div class="val green">+10%</div>
<div class="lbl">Over Mini-SWE-Agent<br>(same backbone)</div>
</div>
<div class="stat-chip">
<div class="val blue">15</div>
<div class="lbl">Issues only resolved<br>by synthesis stage</div>
</div> -->
</div>
<div class="btn-row">
<a href="https://github.com/prism-agent-code/prism-agent" class="btn btn-primary">
<svg viewBox="0 0 24 24" fill="currentColor">
<path d="M12 0C5.37 0 0 5.37 0 12c0 5.31 3.435 9.795 8.205 11.385.6.105.825-.255.825-.57 0-.285-.015-1.23-.015-2.235-3.015.555-3.795-.735-4.035-1.41-.135-.345-.72-1.41-1.23-1.695-.42-.225-1.02-.78-.015-.795.945-.015 1.62.87 1.845 1.23 1.08 1.815 2.805 1.305 3.495.99.105-.78.42-1.305.765-1.605-2.67-.3-5.46-1.335-5.46-5.925 0-1.305.465-2.385 1.23-3.225-.12-.3-.54-1.53.12-3.18 0 0 1.005-.315 3.3 1.23.96-.27 1.98-.405 3-.405s2.04.135 3 .405c2.295-1.56 3.3-1.23 3.3-1.23.66 1.65.24 2.88.12 3.18.765.84 1.23 1.905 1.23 3.225 0 4.605-2.805 5.625-5.475 5.925.435.375.81 1.095.81 2.22 0 1.605-.015 2.895-.015 3.3 0 .315.225.69.825.57A12.02 12.02 0 0 0 24 12c0-6.63-5.37-12-12-12z"/>
</svg>
Prism Code
</a>
<a href="https://doi.org/10.5281/zenodo.19249832" class="btn btn-doi" target="_blank">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2.5" stroke-linecap="round" stroke-linejoin="round">
<path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/>
<path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/>
</svg>
DOI
</a>
<!-- <a href="#overview" class="btn btn-outline">Read Paper →</a> -->
</div>
</div>
</section>
<div class="sep"></div>
<!-- ── Overview ── -->
<section id="overview" class="sec-white">
<div class="container reveal">
<div class="sec-head">
<div class="sec-label">Overview</div>
<h2 class="sec-title">TL;DR</h2>
<div class="divider-line"></div>
</div>
<div class="tldr-wrap">
<p>
We present <strong>Prism</strong>, a multi-solution reasoning and synthesis framework for
repository-level issue resolution. Existing agents face two critical bottlenecks:
<span class="hl-red">limited solution diversity</span> caused by LLM mode collapse and misleading self-reflection,
and a <span class="hl-red">lack of synthesis capabilities</span> that prevents leveraging complementary strengths
across candidate patches. Prism adopts a <span class="hl-gold">coarse-to-fine paradigm</span> across three stages —
<strong>Global Exploration</strong>, <strong>Local Exploration</strong>, and <strong>Solution Synthesis</strong> —
to systematically generate, refine, and integrate diverse repair solutions.
With <span class="hl-blue">Claude 4.6 Sonnet</span>, Prism achieves <strong>82.0% Pass@1</strong> on SWE-bench Verified,
setting a new state-of-the-art. Further analyses confirm that the synthesis stage
uniquely resolves <span class="hl-green">15 issues</span> that no isolated candidate could address alone.
</p>
</div>
<!-- Figure 1 -->
<div class="fig-wrap" style="margin-top:2.4rem;">
<img src="img/example.png" alt="Figure 1 - Motivation Example"
onerror="this.style.display='none'; this.nextElementSibling.style.display='flex';">
<div class="fig-placeholder" style="display:none;">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.3">
<rect x="3" y="3" width="18" height="18" rx="2"/><path d="M3 9h18M9 21V9"/>
</svg>
Figure 1 — Motivation Example (Django #33413)
</div>
<div class="fig-caption">
<strong>Figure 1</strong> · Illustration of existing agent limitations using Django Issue #33413 (<code>ForeignKey db_collation</code> inheritance failure). Live-SWE-Agent correctly fixes <code>related.py</code>; LingXi correctly fixes <code>schema.py</code>; neither alone produces a complete patch — only synthesis resolves the issue.
</div>
</div>
<!-- Limitations -->
<div style="margin-top:3.5rem;">
<div class="sec-head" style="margin-bottom:1.8rem;">
<div class="sec-label">Motivation</div>
<h2 class="sec-title" style="font-size:1.6rem;">Limitations of Existing Work</h2>
</div>
<div class="motiv-grid">
<div class="motiv-card lim1">
<div class="motiv-icon">⚠️</div>
<h4>Limitation 1 — Lack of Solution Diversity & Misleading Self-Reflection</h4>
<p>RLHF-fine-tuned models exhibit <strong>mode collapse</strong>, producing semantically homogeneous solutions even under high-temperature sampling. Simultaneously, erroneous self-reflection feedback causes agents (e.g., Trae, JoyCode) to abandon correct reasoning paths and converge prematurely on incorrect, repetitive patches.</p>
</div>
<div class="motiv-card lim2">
<div class="motiv-icon">🔗</div>
<h4>Limitation 2 — Lack of Solution Synthesis Capabilities</h4>
<p>Existing frameworks evaluate candidates in isolation via ranking or voting, discarding complementary partial solutions. Complex issues (e.g., Django #33413) require coordinated multi-file modifications — no single agent covers all critical dimensions, yet no mechanism exists to synthesize their complementary strengths.</p>
</div>
</div>
</div>
</div>
</section>
<div class="sep"></div>
<!-- ── Approach ── -->
<section id="approach" class="sec-gray">
<div class="container reveal">
<div class="sec-head">
<div class="sec-label">Methodology</div>
<h2 class="sec-title">Main Approach</h2>
<p class="sec-desc">A coarse-to-fine pipeline: Global Exploration → Local Exploration → Solution Synthesis → Patch Generation.</p>
<div class="divider-line"></div>
</div>
<!-- Figure 2 -->
<div class="fig-wrap" style="margin-bottom:3rem;">
<img src="./img/approach.png" alt="Figure 2 - Pipeline Architecture"
onerror="this.style.display='none'; this.nextElementSibling.style.display='flex';">
<div class="fig-placeholder" style="display:none; aspect-ratio:16/8;">
<svg viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1.3">
<circle cx="12" cy="12" r="10"/><path d="M8 12h8M12 8v8"/>
</svg>
Figure 2 — Prism Full Pipeline Architecture
</div>
<div class="fig-caption">
<strong>Figure 2</strong> · The workflow of Prism: Global Exploration generates N<sub>G</sub> semantically diverse solutions via contrastive constraints; Local Exploration uses A*-inspired branching to produce N<sub>L</sub> sub-solutions per global solution; Solution Synthesis employs Reviewer + Judge agents for cross-review and semantic synthesis; Patch Generation selects the final patch via majority voting.
</div>
</div>
<div class="pipeline">
<!-- Stage 1 -->
<div class="pipe-step s-gold">
<div class="pipe-num c-gold">01</div>
<div class="pipe-body">
<h4>Global Exploration <span class="tag tag-gold">Contrastive Constraints</span></h4>
<p>
Led by a <strong>plan agent</strong>, this stage iteratively generates <em>N<sub>G</sub></em> semantically diverse fixing solutions.
Each solution is decomposed into a four-tuple
<strong>G = ⟨R, E, M, C⟩</strong>: Root Cause, Expected Behavior, Modification Strategy, and Relevant Context.
When generating the <em>k</em>-th solution, all prior tuples are injected as
<strong>contrastive semantic constraints</strong>, explicitly requiring divergence across all four dimensions.
</p>
<div class="formula-block">
<span class="hl">P<sub>cons</sub></span> ← enforce (R<sub>k</sub> ≠ R<sub>i</sub>) ∧ (E<sub>k</sub> ≠ E<sub>i</sub>) ∧ (M<sub>k</sub> ≠ M<sub>i</sub>) ∧ (C<sub>k</sub> ≠ C<sub>i</sub>)
<span class="comment"> // for all G<sub>i</sub> ∈ H (historical solution set)</span>
</div>
</div>
</div>
<!-- Stage 2 -->
<div class="pipe-step s-green">
<div class="pipe-num c-green">02</div>
<div class="pipe-body">
<h4>Local Exploration <span class="tag tag-green">A* Branching</span></h4>
<p>
Reusing accumulated context, this stage diversifies the <strong>modification strategy dimension (M)</strong>.
The execution trajectory <strong>T = {s<sub>0</sub>, …, s<sub>L</sub>}</strong> is analyzed to find the
optimal branching point — a critical state with sufficient context but no committed strategy.
</p>
<div class="formula-block">
<span class="hl">f(s<sub>i</sub>)</span> = g(s<sub>i</sub>) + h(s<sub>i</sub>)<br>
g(s<sub>i</sub>) = 1 − i/L <span class="comment">// Context Deficiency Cost</span><br>
h(s<sub>i</sub>) = Rank(s<sub>i</sub>)/L <span class="comment">// Solution Convergence Cost (LLM-ranked)</span><br>
s* = argmin f(s<sub>i</sub>) <span class="comment">// Optimal branching point</span>
</div>
</div>
</div>
<!-- Stage 3 -->
<div class="pipe-step s-blue">
<div class="pipe-num c-blue">03</div>
<div class="pipe-body">
<h4>Solution Synthesis <span class="tag tag-blue">Multi-Agent</span></h4>
<p>
<strong>Reviewer agents</strong> perform cross-review along two dimensions:
<em>Completeness</em> (missing fix locations?) and <em>Regression Risk</em> (unintended side effects?).
The <strong>Judge agent</strong> then performs (1) Defect Verification, (2) Complementarity Analysis
across ⟨R, E, M, C⟩, and (3) Semantic Synthesis — rewriting retained strengths into a unified repair plan.
</p>
</div>
</div>
<!-- Stage 4 -->
<div class="pipe-step s-purple">
<div class="pipe-num c-purple">04</div>
<div class="pipe-body">
<h4>Patch Generation <span class="tag tag-purple">Majority Voting</span></h4>
<p>
All candidates from the three stages are consolidated into a <strong>unified candidate pool</strong>.
Each undergoes static and dynamic verification. A <strong>majority voting mechanism</strong>
selects the single final patch to be submitted.
</p>
</div>
</div>
</div>
</div>
</section>
<div class="sep"></div>
<!-- ── Results ── -->
<section id="results" class="sec-white">
<div class="container reveal">
<div class="sec-head">
<div class="sec-label">Experiments</div>
<h2 class="sec-title">Main Results</h2>
<p class="sec-desc">Evaluated on SWE-bench Verified and SWE-bench Live Lite. Metric: Pass@1 (% Resolved).</p>
<div class="divider-line"></div>
</div>
<div style="margin-bottom:.7rem; font-size:.82rem; font-weight:700; color:var(--gold); letter-spacing:.8px; text-transform:uppercase;">SWE-bench Verified</div>
<div class="table-scroll">
<table>
<thead>
<tr><th>Method</th><th>Backbone Model</th><th>Pass@1 (%)</th></tr>
</thead>
<tbody>
<tr><td>SWE-Agent</td><td><span class="model-tag">GPT-4o</span></td><td>38.0</td></tr>
<tr><td>OpenHands</td><td><span class="model-tag">Claude-3.5-Sonnet</span></td><td>53.0</td></tr>
<tr><td>Trae</td><td><span class="model-tag">Claude-3.7-Sonnet</span></td><td>68.0</td></tr>
<tr><td>JoyCode</td><td><span class="model-tag">Claude-3.7-Sonnet</span></td><td>70.5</td></tr>
<tr><td>Mini-SWE-Agent</td><td><span class="model-tag">DeepSeek-V3.2-Reasoner</span></td><td>70.0</td></tr>
<tr class="ours"><td>Prism</td><td><span class="model-tag">DeepSeek-V3.2-Reasoner</span></td><td class="best">80.0</td></tr>
<tr class="ours"><td>Prism</td><td><span class="model-tag">Claude 4.6 Sonnet</span></td><td class="best">82.0</td></tr>
</tbody>
</table>
</div>
<div style="margin-top:2.2rem; margin-bottom:.7rem; font-size:.82rem; font-weight:700; color:var(--gold); letter-spacing:.8px; text-transform:uppercase;">SWE-bench Live Lite</div>
<div class="table-scroll">
<table>
<thead>
<tr><th>Method</th><th>Backbone Model</th><th>Pass@1 (%)</th></tr>
</thead>
<tbody>
<tr><td>SWE-Agent (SOTA baseline)</td><td><span class="model-tag">Claude-4.5-Sonnet</span></td><td>36.0</td></tr>
<tr class="ours"><td>Prism</td><td><span class="model-tag">DeepSeek-V3.2-Reasoner</span></td><td class="tbd">42.3%</td></tr>
<tr class="ours"><td>Prism</td><td><span class="model-tag">Claude 4.6 Sonnet</span></td><td class="tbd">45.6%</td></tr>
</tbody>
</table>
</div>
<div class="table-note">★ Prism outperforms all baselines on SWE-bench Verified. Live Lite results pending.</div>
</div>
</section>
<div class="sep"></div>
<!-- ── Analysis ── -->
<section id="analysis" class="sec-gray">
<div class="container reveal">
<div class="sec-head">
<div class="sec-label">Ablation & Analysis</div>
<h2 class="sec-title">Analysis</h2>
<p class="sec-desc">Stage-wise ablation and component analysis using DeepSeek-V3.2-Reasoner on SWE-bench Verified.</p>
<div class="divider-line"></div>
</div>
<div class="rq-list">
<div class="rq-item">
<div class="rq-badge b-gold">RQ1</div>
<div class="rq-content">
<h4>Overall Performance:How effective is Prism in repositorylevel issue resolution compared to SOTA baseline agents</h4>
<p> On the SWE-Bench Verified benchmark, Prism establishes new SOTA results, achieving 82.0% Pass@1 when powered by Claude 4.6 Sonnet and 80.0% with the open-source DeepSeek-V3.2-Reasoner. Critically, on the dynamically updated SWE-Bench Live Lite dataset specifically constructed to reduce data leakage, Prism (utilizing DeepSeek-V3.2-Reasoner) outperforms the leading closed-source SOTA combination (SWE-Agent + Claude-4.5-Sonnet, 36.0%), attaining xx.x%. These findings confirm that Prism’s enhanced performance in repository-level issue resolution originates from its agentic architecture, rather than from memorization or overfitting to benchmark data.</p>
</div>
</div>
<div class="rq-item">
<div class="rq-badge b-green">RQ2</div>
<div class="rq-content">
<h4>Effectiveness of Exploration Mechanism:How do the Global and Local Exploration mechanisms contribute to Prism’s issue resolution rate?</h4>
<p>The progressive exploration mechanism of Prism demonstrates a significant advantage over random sampling. With the same
candidate solution pool size (N=12), Prism achieves a Pass@1 of 80.0%,
compared to only 71.8% achieved by the high-temperature sampling baseline. Evaluating the resolution rates of the global and local exploration
stages individually reveals that Prism effectively navigates the solution
space to generate high-quality patches, rather than merely relying on an
increased number of candidate solutions.</p>
<div class="rq-img-ph">
<img src="img/rq2.png"
alt="Figure 2 - Prism Pipeline"
style="width:100%; display:block; border-radius:18px;">
</div>
</div>
</div>
<div class="rq-item">
<div class="rq-badge b-blue">RQ3</div>
<div class="rq-content">
<h4>Effectiveness of Synthesis Mechanism: Can Solution
Synthesis mechanism effectively combine the advantages of different candidate solutions to improve Prism’s issue resolution rate?</h4>
<p>The solution synthesis stage is critical to enhancing
Prism’s resolution rate. With only three integrated candidate solutions,
Prism achieves a Pass@1 of 79.2%, surpassing both the global exploration
stage (74.8% with 3 solutions) and the local exploration phase (75.0%
with 6 solutions). The UpSet plot further validates the unique contribution
of the solution synthesis stage, revealing 15 issues that only the integrated
solutions could resolve. Furthermore, retaining all candidate solutions
across all three phases (12 in total) enables Prism to maximize its theoretical upper bound (Pass@N=84.6%) and, through majority voting, attain a
SOTA Pass@1 performance of 80.0%.</p>
<div class="rq-img-ph">
<img src="img/rq3.png"
alt="Figure 2 - Prism Pipeline"
style="width:100%; display:block; border-radius:18px;">
</div>
</div>
</div>
</div>
</div>
</section>
<div class="sep"></div>
<!-- ── Contributions ── -->
<section class="sec-white" style="padding-top:4.5rem; padding-bottom:5.5rem;">
<div class="container reveal">
<div class="sec-head">
<div class="sec-label">Summary</div>
<h2 class="sec-title">Contributions</h2>
<div class="divider-line"></div>
</div>
<div class="contrib-grid">
<div class="contrib-card">
<div class="contrib-num">01</div>
<h4>Effective Issue Resolution Framework</h4>
<p>Prism introduces a multi-path reasoning and synthesis framework with three stages — global exploration, local exploration, and solution synthesis — that systematically expands the solution space to produce high-quality patches.</p>
</div>
<div class="contrib-card">
<div class="contrib-num">02</div>
<h4>Thorough Evaluation</h4>
<p>Evaluated on SWE-bench Verified and SWE-bench Live Lite with both DeepSeek-V3.2-Reasoner and Claude 4.6 Sonnet, achieving SOTA performance against all existing issue resolution agents.</p>
</div>
<div class="contrib-card">
<div class="contrib-num">03</div>
<h4>Data Availability</h4>
<p>A full reproduction package — including ground-truth datasets, tools, and raw experimental data — is publicly available to support further research and reproducibility.</p>
</div>
</div>
</div>
</section>
<!-- ── Footer ── -->
<footer>
<p>
<!-- Anonymous submission · Under review at ASE 2026 · -->
Reproduction package: <a href="https://prism-agent-code.github.io/" target="_blank">prism-agent-code.github.io</a> ·
Licensed under <a href="http://creativecommons.org/licenses/by-sa/4.0/" target="_blank">CC BY-SA 4.0</a>
</p>
</footer>
<script>
const io = new IntersectionObserver(entries => {
entries.forEach(e => { if (e.isIntersecting) e.target.classList.add('visible'); });
}, { threshold: 0.06 });
document.querySelectorAll('.reveal').forEach(el => io.observe(el));
</script>
</body>
</html>