-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathpresentation.html
More file actions
1444 lines (1376 loc) · 62.5 KB
/
presentation.html
File metadata and controls
1444 lines (1376 loc) · 62.5 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Evaluating AI Code Review Tools</title>
<link href="https://fonts.googleapis.com/css2?family=DM+Sans:opsz,wght@9..40,400;9..40,500;9..40,600;9..40,700&family=DM+Mono:wght@400;500&family=IBM+Plex+Sans:wght@300;400;500;600;700&family=IBM+Plex+Mono:wght@400;500;600&display=swap" rel="stylesheet">
<style>
:root {
--bg: #0a0a0a;
--bg-card: #111111;
--bg-card-alt: #0f0f0f;
--bg-elevated: #161616;
--bg-highlight: #1a1a1a;
--bg-info: #0d1117;
--border: #1e1e1e;
--border-strong: #222222;
--text: #e2e2e2;
--text-bright: #f5f5f5;
--text-dim: #999999;
--text-muted: #888888;
--green: #4ade80;
--green-bg: #052e16;
--green-border: #14532d;
--red: #f87171;
--red-bg: #2a1a1a;
--red-border: #7f1d1d;
--yellow: #facc15;
--yellow-bg: #1a1500;
--yellow-border: #713f12;
--blue: #60a5fa;
--blue-bg: #0c1929;
--blue-border: #1e3a5f;
--purple: #c084fc;
--purple-bg: #1a0a2e;
--purple-border: #4c1d95;
}
* { margin: 0; padding: 0; box-sizing: border-box; }
html, body {
width: 100%; height: 100%;
overflow: hidden;
background: var(--bg);
color: var(--text);
font-family: 'DM Sans', 'IBM Plex Sans', -apple-system, sans-serif;
}
/* ── Slide container ── */
.deck {
position: relative;
width: 100%; height: 100%;
}
.slide {
position: absolute;
inset: 0;
display: flex;
flex-direction: column;
justify-content: center;
align-items: center;
padding: 60px 80px;
opacity: 0;
transform: translateX(40px);
pointer-events: none;
transition: opacity 0.45s ease, transform 0.45s ease;
}
.slide.entering-back {
transform: translateX(-40px);
}
.slide.active {
opacity: 1;
transform: translateX(0);
pointer-events: auto;
}
.slide.exiting {
opacity: 0;
transform: translateX(-40px);
}
.slide.exiting-back {
opacity: 0;
transform: translateX(40px);
}
/* ── Progress bar ── */
.progress-track {
position: fixed;
top: 0; left: 0; right: 0;
height: 3px;
background: var(--border);
z-index: 100;
}
.progress-fill {
height: 100%;
background: linear-gradient(90deg, var(--green), var(--blue));
transition: width 0.4s ease;
}
/* ── Slide counter ── */
.slide-counter {
position: fixed;
bottom: 24px; right: 32px;
font-family: 'DM Mono', 'IBM Plex Mono', monospace;
font-size: 12px;
color: var(--text-muted);
z-index: 100;
letter-spacing: 1px;
}
/* ── Back link ── */
/* ── Typography ── */
.slide-title {
font-size: 42px;
font-weight: 700;
color: var(--text-bright);
letter-spacing: -0.5px;
margin-bottom: 32px;
line-height: 1.2;
}
.slide-subtitle {
font-size: 18px;
color: var(--text-dim);
font-weight: 400;
line-height: 1.6;
max-width: 800px;
}
.section-label {
font-size: 10px;
font-weight: 700;
text-transform: uppercase;
letter-spacing: 1.5px;
margin-bottom: 16px;
}
.label-green { color: var(--green); }
.label-blue { color: var(--blue); }
.label-yellow { color: var(--yellow); }
.label-red { color: var(--red); }
.label-purple { color: var(--purple); }
.label-dim { color: var(--text-muted); }
/* ── Content wrapper ── */
.content {
width: 100%;
max-width: 1100px;
}
.content-wide {
width: 100%;
max-width: 1200px;
}
/* ── Tables ── */
.data-table {
width: 100%;
border-collapse: collapse;
border: 1px solid var(--border);
border-radius: 8px;
overflow: hidden;
background: var(--bg-card);
}
.data-table thead tr {
background: var(--bg-elevated);
border-bottom: 2px solid var(--border-strong);
}
.data-table th {
padding: 12px 18px;
text-align: left;
font-size: 10px;
font-weight: 700;
color: var(--text-muted);
text-transform: uppercase;
letter-spacing: 1px;
white-space: nowrap;
}
.data-table td {
padding: 14px 18px;
border-bottom: 1px solid var(--border);
font-size: 13px;
line-height: 1.5;
vertical-align: middle;
}
.data-table tbody tr:nth-child(even) { background: var(--bg-card-alt); }
.data-table tbody tr:nth-child(odd) { background: var(--bg-card); }
.val-bright { color: var(--text-bright); font-weight: 600; }
.val-green { color: var(--green); font-weight: 700; }
.val-red { color: var(--red); font-weight: 600; }
.val-yellow { color: var(--yellow); font-weight: 600; }
.val-blue { color: var(--blue); font-weight: 600; }
.val-dim { color: var(--text-muted); }
.val-mono { font-family: 'DM Mono', 'IBM Plex Mono', monospace; font-size: 12px; }
/* ── Badges ── */
.badge {
display: inline-block;
padding: 3px 10px;
border-radius: 4px;
font-size: 11px;
font-weight: 600;
letter-spacing: 0.5px;
text-transform: uppercase;
}
.badge-green { background: var(--green-bg); color: var(--green); border: 1px solid var(--green-border); }
.badge-yellow { background: var(--yellow-bg); color: var(--yellow); border: 1px solid var(--yellow-border); }
.badge-red { background: var(--red-bg); color: var(--red); border: 1px solid var(--red-border); }
.badge-blue { background: var(--blue-bg); color: var(--blue); border: 1px solid var(--blue-border); }
.badge-purple { background: var(--purple-bg); color: var(--purple); border: 1px solid var(--purple-border); }
.badge-dim { background: #1a1a1a; color: #888; border: 1px solid #333; }
/* ── Cards ── */
.card {
background: var(--bg-card);
border: 1px solid var(--border);
border-radius: 8px;
padding: 24px;
}
.card-row {
display: grid;
gap: 16px;
}
.card-row-2 { grid-template-columns: 1fr 1fr; }
.card-row-3 { grid-template-columns: 1fr 1fr 1fr; }
.card-row-4 { grid-template-columns: 1fr 1fr 1fr 1fr; }
.card h3 {
font-size: 14px;
font-weight: 700;
color: var(--text-bright);
margin-bottom: 8px;
}
.card p {
font-size: 13px;
color: var(--text-dim);
line-height: 1.6;
}
/* ── Stat blocks ── */
.stat {
text-align: center;
padding: 20px;
}
.stat-value {
font-family: 'DM Mono', 'IBM Plex Mono', monospace;
font-size: 48px;
font-weight: 700;
line-height: 1;
margin-bottom: 8px;
}
.stat-label {
font-size: 11px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 1px;
color: var(--text-muted);
}
/* ── Bullet lists ── */
.bullets {
list-style: none;
padding: 0;
}
.bullets li {
position: relative;
padding: 10px 0 10px 24px;
font-size: 15px;
line-height: 1.6;
color: var(--text);
border-bottom: 1px solid var(--border);
}
.bullets li:last-child { border-bottom: none; }
.bullets li::before {
content: '';
position: absolute;
left: 0; top: 18px;
width: 8px; height: 8px;
border-radius: 2px;
background: var(--green);
}
.bullets.red li::before { background: var(--red); }
.bullets.blue li::before { background: var(--blue); }
.bullets.yellow li::before { background: var(--yellow); }
.bullets.purple li::before { background: var(--purple); }
.bullets li strong { color: var(--text-bright); }
/* ── Pipeline flow ── */
.pipeline {
display: flex;
align-items: center;
gap: 0;
flex-wrap: wrap;
justify-content: center;
}
.pipeline-step {
background: var(--bg-elevated);
border: 1px solid var(--border);
border-radius: 6px;
padding: 14px 22px;
text-align: center;
}
.pipeline-step .step-name {
font-size: 13px;
font-weight: 700;
color: var(--text-bright);
}
.pipeline-step .step-desc {
font-size: 11px;
color: var(--text-muted);
margin-top: 4px;
}
.pipeline-arrow {
font-size: 18px;
color: var(--text-muted);
padding: 0 8px;
}
/* ── Funnel ── */
.funnel {
display: flex;
flex-direction: column;
align-items: center;
gap: 0;
}
.funnel-step {
display: flex;
align-items: center;
gap: 16px;
padding: 14px 0;
}
.funnel-bar {
height: 36px;
border-radius: 4px;
display: flex;
align-items: center;
justify-content: flex-end;
padding-right: 14px;
font-family: 'DM Mono', 'IBM Plex Mono', monospace;
font-size: 13px;
font-weight: 600;
color: var(--text-bright);
transition: width 0.8s ease;
}
.funnel-label {
font-size: 13px;
color: var(--text-dim);
width: 200px;
text-align: right;
}
.funnel-arrow {
font-size: 14px;
color: var(--text-muted);
padding: 4px 0;
}
/* ── Comparison bars ── */
.bar-chart {
display: flex;
flex-direction: column;
gap: 12px;
width: 100%;
}
.bar-row {
display: grid;
grid-template-columns: 180px 1fr 60px;
align-items: center;
gap: 12px;
}
.bar-label {
font-size: 13px;
font-weight: 600;
color: var(--text-bright);
text-align: right;
}
.bar-track {
height: 32px;
background: var(--bg-elevated);
border-radius: 4px;
overflow: hidden;
position: relative;
}
.bar-fill {
height: 100%;
border-radius: 4px;
transition: width 1s ease;
position: relative;
}
.bar-value {
font-family: 'DM Mono', 'IBM Plex Mono', monospace;
font-size: 13px;
font-weight: 600;
}
/* ── Stagger animation ── */
.stagger > * {
opacity: 0;
transform: translateY(16px);
transition: opacity 0.4s ease, transform 0.4s ease;
}
.slide.active .stagger > * {
opacity: 1;
transform: translateY(0);
}
.slide.active .stagger > *:nth-child(1) { transition-delay: 0.1s; }
.slide.active .stagger > *:nth-child(2) { transition-delay: 0.2s; }
.slide.active .stagger > *:nth-child(3) { transition-delay: 0.3s; }
.slide.active .stagger > *:nth-child(4) { transition-delay: 0.4s; }
.slide.active .stagger > *:nth-child(5) { transition-delay: 0.5s; }
.slide.active .stagger > *:nth-child(6) { transition-delay: 0.6s; }
.slide.active .stagger > *:nth-child(7) { transition-delay: 0.7s; }
.slide.active .stagger > *:nth-child(8) { transition-delay: 0.8s; }
.slide.active .stagger > *:nth-child(9) { transition-delay: 0.9s; }
.slide.active .stagger > *:nth-child(10) { transition-delay: 1.0s; }
/* ── Callout ── */
.callout {
background: var(--bg-info);
border: 1px solid var(--border);
border-radius: 8px;
padding: 20px 24px;
margin-top: 24px;
}
.callout-title {
font-weight: 700;
color: var(--text-bright);
font-size: 13px;
margin-bottom: 8px;
}
.callout p {
font-size: 13px;
color: var(--text-dim);
line-height: 1.7;
}
/* ── Versus layout ── */
.vs-grid {
display: grid;
grid-template-columns: 1fr auto 1fr;
gap: 24px;
align-items: center;
}
.vs-divider {
font-size: 24px;
font-weight: 700;
color: var(--text-muted);
}
/* ── Strikethrough for corrected values ── */
.corrected {
text-decoration: line-through;
color: var(--text-muted);
font-weight: 400;
}
/* ── Caveat tag ── */
.caveat {
display: inline-block;
font-size: 10px;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.8px;
padding: 2px 7px;
border-radius: 3px;
margin-left: 6px;
vertical-align: middle;
}
.caveat-yellow { background: var(--yellow-bg); color: var(--yellow); border: 1px solid var(--yellow-border); }
.caveat-red { background: var(--red-bg); color: var(--red); border: 1px solid var(--red-border); }
/* ── Ensemble diagram ── */
.ensemble-row {
display: flex;
align-items: center;
gap: 12px;
justify-content: center;
flex-wrap: wrap;
}
.ensemble-tool {
background: var(--bg-elevated);
border: 1px solid var(--border);
border-radius: 8px;
padding: 16px 24px;
text-align: center;
min-width: 160px;
}
.ensemble-tool .tool-name {
font-size: 14px;
font-weight: 700;
color: var(--text-bright);
margin-bottom: 4px;
}
.ensemble-tool .tool-role {
font-size: 11px;
color: var(--text-muted);
}
.ensemble-plus {
font-size: 28px;
font-weight: 300;
color: var(--text-muted);
}
.ensemble-equals {
font-size: 28px;
font-weight: 300;
color: var(--green);
}
.ensemble-result {
background: var(--green-bg);
border: 1px solid var(--green-border);
border-radius: 8px;
padding: 16px 28px;
text-align: center;
}
.ensemble-result .tool-name {
font-size: 16px;
font-weight: 700;
color: var(--green);
margin-bottom: 4px;
}
.ensemble-result .tool-role {
font-size: 11px;
color: var(--green);
opacity: 0.7;
}
</style>
</head>
<body>
<div class="progress-track"><div class="progress-fill" id="progress"></div></div>
<div class="slide-counter" id="counter"></div>
<a href="/" style="position:fixed;bottom:24px;left:32px;font-size:11px;color:#888;text-decoration:none;z-index:100;">← Dashboard</a>
<div class="deck" id="deck">
<!-- ============================================ -->
<!-- SLIDE 1: Title -->
<!-- ============================================ -->
<section class="slide active">
<div class="content" style="text-align:center;">
<div class="stagger">
<div class="section-label label-green" style="margin-bottom:24px;">Provable Engineering</div>
<h1 style="font-size:52px; font-weight:700; color:var(--text-bright); letter-spacing:-1px; line-height:1.15; margin-bottom:20px;">
Evaluating AI<br>Code Review Tools
</h1>
<div style="margin-bottom:40px;"></div>
<div style="display:flex; gap:8px; justify-content:center; flex-wrap:wrap;">
<span class="badge badge-green">Pilot Results</span>
<span class="badge badge-blue">67 Bug Cases</span>
<span class="badge badge-purple">8 Tool Configs</span>
</div>
<p style="font-size:13px; color:var(--text-muted); margin-top:12px;">v3 evaluation · Opus-judged · Det≥2 metric</p>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 2: The Problem -->
<!-- ============================================ -->
<section class="slide">
<div class="content">
<div class="stagger">
<div class="section-label label-red">The Problem</div>
<h2 class="slide-title">General benchmarks don't test<br>what matters for Provable</h2>
<div class="card-row card-row-2" style="margin-top:8px;">
<div class="card">
<h3 style="color:var(--red);">What benchmarks test</h3>
<ul class="bullets red" style="margin-top:12px;">
<li>Generic web app bugs</li>
<li>Common language patterns</li>
<li>Well-documented frameworks</li>
<li>Typical CRUD logic errors</li>
</ul>
</div>
<div class="card">
<h3 style="color:var(--green);">What Provable needs</h3>
<ul class="bullets" style="margin-top:12px;">
<li><strong>ZK circuit constraint</strong> correctness</li>
<li><strong>Soundness bugs</strong> in proof systems</li>
<li>Leo language semantics</li>
<li>Custom VM and compiler internals</li>
</ul>
</div>
</div>
<div class="callout">
<p style="color:var(--text);">There are 10+ commercial AI review tools now. No established methodology to compare them on domain-specific code. <strong style="color:var(--text-bright);">So we built our own evaluation framework.</strong></p>
</div>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 3: Tool Landscape -->
<!-- ============================================ -->
<section class="slide">
<div class="content-wide">
<div class="stagger">
<div class="section-label label-blue">Landscape</div>
<h2 class="slide-title" style="font-size:34px;">AI Code Review Tools — March 2026</h2>
<div style="overflow-x:auto; border:1px solid var(--border); border-radius:8px;">
<table class="data-table">
<thead>
<tr>
<th>Tool</th>
<th>Price</th>
<th>Unlimited?</th>
<th>Platforms</th>
<th>Key Strength</th>
<th>Pilot</th>
</tr>
</thead>
<tbody>
<tr>
<td class="val-bright">CodeRabbit</td>
<td>$30/user</td>
<td class="val-green">Yes</td>
<td><span class="badge badge-dim" style="font-size:10px;">GH</span> <span class="badge badge-dim" style="font-size:10px;">GL</span> <span class="badge badge-dim" style="font-size:10px;">BB</span> <span class="badge badge-dim" style="font-size:10px;">Azure</span></td>
<td style="font-size:12px;">Most installed GitHub AI app (2M+ repos)</td>
<td><span class="badge badge-yellow">Tested</span></td>
</tr>
<tr>
<td class="val-bright">CodeAnt AI</td>
<td>$10–24/user</td>
<td class="val-green">Yes</td>
<td><span class="badge badge-dim" style="font-size:10px;">GH</span> <span class="badge badge-dim" style="font-size:10px;">GL</span> <span class="badge badge-dim" style="font-size:10px;">BB</span> <span class="badge badge-dim" style="font-size:10px;">Azure</span></td>
<td style="font-size:12px;">Bundles SAST, secrets, IaC, SCA, DORA</td>
<td><span class="badge badge-dim">Planned</span></td>
</tr>
<tr>
<td class="val-bright">GitHub Copilot</td>
<td>$10–39/user</td>
<td class="val-red">No</td>
<td><span class="badge badge-dim" style="font-size:10px;">GH</span></td>
<td style="font-size:12px;">Zero friction, bundled with existing tooling</td>
<td><span class="badge badge-green">Tested</span></td>
</tr>
<tr>
<td class="val-bright">Greptile</td>
<td>$30 + $1/review</td>
<td class="val-red">No</td>
<td><span class="badge badge-dim" style="font-size:10px;">GH</span> <span class="badge badge-dim" style="font-size:10px;">GL</span></td>
<td style="font-size:12px;">Full codebase graph indexing, 82% catch rate</td>
<td><span class="badge badge-yellow">Tested</span></td>
</tr>
<tr>
<td class="val-bright">Claude Code Review</td>
<td>$15–25/review</td>
<td class="val-red">No</td>
<td><span class="badge badge-dim" style="font-size:10px;">GH</span></td>
<td style="font-size:12px;">Deepest analysis, <1% false positives</td>
<td><span class="badge badge-dim">Planned</span></td>
</tr>
<tr>
<td class="val-bright">In-house Agent</td>
<td>~$0.65/review</td>
<td class="val-green">Yes</td>
<td><span class="badge badge-dim" style="font-size:10px;">Any</span></td>
<td style="font-size:12px;">Full control, domain prompting, customizable</td>
<td><span class="badge badge-green">Tested</span></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 4: Our Approach -->
<!-- ============================================ -->
<section class="slide">
<div class="content">
<div class="stagger">
<div class="section-label label-green">Approach</div>
<h2 class="slide-title">Full evaluation design</h2>
<div class="card" style="margin-bottom:16px; border-color:var(--green-border);">
<div class="section-label label-green">How It Works</div>
<p style="font-size:15px; color:var(--text); line-height:1.7;">For each known bug: we check out the code <strong style="color:var(--text-bright);">before the fix</strong>, show it to each tool as if it were a new PR, and measure whether the tool identifies the bug. We know the answer because the fix has already been merged — that's our ground truth.</p>
</div>
<div class="card-row card-row-2">
<div class="card">
<div class="section-label label-purple">The In-House Agent</div>
<p style="font-size:13px; line-height:1.7;">Uses the <strong style="color:var(--text-bright);">Claude Agent SDK</strong> to spawn a Claude Code session in a cloned repo. The agent gets <strong style="color:var(--text-bright);">Bash, Grep, Read, Glob, and WebSearch</strong> — the same tools a human reviewer would use. It explores the codebase, reads related files, and writes a structured review. All within a sandboxed workspace.</p>
</div>
<div class="card">
<div class="section-label label-yellow">Context Levels</div>
<p style="font-size:13px; line-height:1.7;"><strong style="color:var(--text-bright);">diff-only</strong> — agent sees only the patch. No repo, no shell tools. Tests pure reasoning from the diff alone.</p>
<p style="font-size:13px; line-height:1.7; margin-top:8px;"><strong style="color:var(--text-bright);">diff+repo</strong> — agent gets the full repo checkout at the base commit plus Bash and ripgrep. Tests whether codebase access improves catch rate.</p>
</div>
</div>
<div style="display:flex; gap:16px; margin-top:16px; flex-wrap:wrap; justify-content:center;">
<span style="font-size:12px; color:var(--text-dim);">Do custom agents beat commercial tools?</span>
<span style="color:var(--text-muted);">·</span>
<span style="font-size:12px; color:var(--text-dim);">Does repo context improve catch rates?</span>
<span style="color:var(--text-muted);">·</span>
<span style="font-size:12px; color:var(--text-dim);">What's the cost per bug?</span>
<span style="color:var(--text-muted);">·</span>
<span style="font-size:12px; color:var(--text-dim);">How noisy are the tools?</span>
</div>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 5: Dataset Mining -->
<!-- ============================================ -->
<section class="slide">
<div class="content">
<div class="stagger">
<div class="section-label label-yellow">Dataset Mining</div>
<h2 class="slide-title">500 PRs in, 67 cases out</h2>
<div style="max-width:700px; margin:0 auto 24px;">
<div style="display:flex; align-items:center; gap:12px; margin-bottom:10px;">
<span style="font-size:12px; color:var(--text-dim); width:130px; text-align:right; flex-shrink:0;">PRs fetched</span>
<div style="flex:1; height:28px; background:var(--bg-elevated); border-radius:4px;">
<div style="width:100%; height:100%; background:linear-gradient(90deg, var(--blue-bg), var(--blue)); border-radius:4px; display:flex; align-items:center; padding-left:12px;">
<span style="font-family:'DM Mono',monospace; font-size:12px; font-weight:600; color:var(--text-bright);">500</span>
</div>
</div>
</div>
<div style="margin-left:142px; padding:0 0 6px; font-size:11px; color:var(--red);">↓ −189 no fix commit found</div>
<div style="display:flex; align-items:center; gap:12px; margin-bottom:10px;">
<span style="font-size:12px; color:var(--text-dim); width:130px; text-align:right; flex-shrink:0;">Cases mined</span>
<div style="flex:1; height:28px; background:var(--bg-elevated); border-radius:4px;">
<div style="width:62%; height:100%; background:linear-gradient(90deg, var(--purple-bg), var(--purple)); border-radius:4px; display:flex; align-items:center; padding-left:12px;">
<span style="font-family:'DM Mono',monospace; font-size:12px; font-weight:600; color:var(--text-bright);">311</span>
</div>
</div>
</div>
<div style="margin-left:142px; padding:0 0 6px; font-size:11px; color:var(--red);">↓ −93 dep bumps, −45 duplicates, −27 no buggy lines</div>
<div style="display:flex; align-items:center; gap:12px; margin-bottom:10px;">
<span style="font-size:12px; color:var(--text-dim); width:130px; text-align:right; flex-shrink:0;">After rule filters</span>
<div style="flex:1; height:28px; background:var(--bg-elevated); border-radius:4px;">
<div style="width:29%; height:100%; background:linear-gradient(90deg, var(--yellow-bg), var(--yellow)); border-radius:4px; display:flex; align-items:center; padding-left:12px;">
<span style="font-family:'DM Mono',monospace; font-size:12px; font-weight:600; color:var(--text-bright);">146</span>
</div>
</div>
</div>
<div style="margin-left:142px; padding:0 0 6px; font-size:11px; color:var(--red);">↓ −79 LLM curation (features, CI fixes, doc fixes, test-only)</div>
<div style="display:flex; align-items:center; gap:12px;">
<span style="font-size:12px; color:var(--green); width:130px; text-align:right; flex-shrink:0; font-weight:600;">Active cases</span>
<div style="flex:1; height:28px; background:var(--bg-elevated); border-radius:4px;">
<div style="width:13%; height:100%; background:linear-gradient(90deg, var(--green-bg), var(--green)); border-radius:4px; display:flex; align-items:center; padding-left:12px;">
<span style="font-family:'DM Mono',monospace; font-size:12px; font-weight:600; color:var(--text-bright);">67</span>
</div>
</div>
</div>
</div>
<div class="card-row card-row-2">
<div class="card">
<div class="section-label label-red">Exclusions (244)</div>
<div style="display:grid; grid-template-columns:auto 1fr; gap:4px 12px; font-size:12px;">
<span class="val-mono val-dim">93</span><span>Dependency bumps</span>
<span class="val-mono val-dim">45</span><span>Merged sibling (duplicate fix)</span>
<span class="val-mono val-dim">27</span><span>No buggy lines found</span>
<span class="val-mono val-dim">19</span><span>Features mislabeled as fix</span>
<span class="val-mono val-dim">16</span><span>Test-expectation only</span>
<span class="val-mono val-dim">14</span><span>CI/release script fixes</span>
<span class="val-mono val-dim">12</span><span>Doc-only fixes</span>
<span class="val-mono val-dim">18</span><span>Other (self-ref, typo, lint, perf)</span>
</div>
</div>
<div class="card">
<div class="section-label label-green">Why 67?</div>
<ul class="bullets" style="font-size:13px;">
<li>857 merged PRs over 3 years → <strong>164 real bug fixes</strong> (19% bug rate)</li>
<li>67 have <strong>provable ground truth</strong> — verified introducing commit + precise buggy lines</li>
<li>90 more are real bugs with weak ground truth (recoverable with LLM-based labeling)</li>
<li>Not a sample — these are <strong>all the bugs we can rigorously verify</strong></li>
</ul>
</div>
</div>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 6: The Framework -->
<!-- ============================================ -->
<section class="slide">
<div class="content">
<div class="stagger">
<div class="section-label label-purple">Framework</div>
<h2 class="slide-title">End-to-end evaluation pipeline</h2>
<div class="pipeline" style="margin-bottom:16px;">
<div class="pipeline-step">
<div class="step-name">mine</div>
<div class="step-desc">Scrape fix PRs</div>
</div>
<div class="pipeline-arrow">→</div>
<div class="pipeline-step">
<div class="step-name">blame</div>
<div class="step-desc">Find introducing commits</div>
</div>
<div class="pipeline-arrow">→</div>
<div class="pipeline-step">
<div class="step-name">curate</div>
<div class="step-desc">Auto-exclude bad cases</div>
</div>
<div class="pipeline-arrow">→</div>
<div class="pipeline-step" style="border-color:var(--green-border);">
<div class="step-name" style="color:var(--green);">evaluate</div>
<div class="step-desc">Run tools on cases</div>
</div>
</div>
<div class="pipeline" style="margin-bottom:24px;">
<div class="pipeline-step">
<div class="step-name">normalize</div>
<div class="step-desc">Common format</div>
</div>
<div class="pipeline-arrow">→</div>
<div class="pipeline-step" style="border-color:var(--blue-border);">
<div class="step-name" style="color:var(--blue);">judge</div>
<div class="step-desc">LLM scores each review</div>
</div>
<div class="pipeline-arrow">→</div>
<div class="pipeline-step">
<div class="step-name">analyze</div>
<div class="step-desc">Reports & charts</div>
</div>
</div>
<div class="card-row card-row-3">
<div class="card" style="text-align:center;">
<div class="stat-value val-green" style="font-size:28px;">67</div>
<div class="stat-label">Bug cases scored</div>
</div>
<div class="card" style="text-align:center;">
<div class="stat-value val-blue" style="font-size:28px;">3</div>
<div class="stat-label">Eval modes (PR, API, Agent)</div>
</div>
<div class="card" style="text-align:center;">
<div class="stat-value val-purple" style="font-size:28px;">~$509</div>
<div class="stat-label">Total pilot cost</div>
</div>
</div>
<div class="callout">
<div class="callout-title">How we score</div>
<div style="display:grid; grid-template-columns:1fr 1fr; gap:12px 24px; margin-top:8px;">
<div>
<p style="font-size:12px; color:var(--text-bright); font-weight:600; margin-bottom:4px;">Detection (0–3): did the tool find the bug?</p>
<p style="font-size:12px; color:var(--text-dim); line-height:1.6;"><strong>0</strong> missed → <strong>1</strong> wrong area → <strong>2</strong> correct file+issue → <strong>3</strong> correct ID + actionable fix</p>
</div>
<div>
<p style="font-size:12px; color:var(--text-bright); font-weight:600; margin-bottom:4px;">Quality (0–4): how useful was the review?</p>
<p style="font-size:12px; color:var(--text-dim); line-height:1.6;"><strong>0</strong> useless → <strong>1</strong> shallow → <strong>2</strong> adequate → <strong>3</strong> strong → <strong>4</strong> exceptional</p>
</div>
</div>
<p style="font-size:12px; color:var(--text-muted); margin-top:8px;">Scored by Claude Opus as judge — tool names hidden to prevent bias.</p>
</div>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 7: Results Leaderboard -->
<!-- ============================================ -->
<section class="slide">
<div class="content-wide">
<div class="stagger">
<div class="section-label label-green">Pilot Results</div>
<h2 class="slide-title" style="font-size:34px; margin-bottom:12px;">Leaderboard — 67 cases, Opus-judged, Det≥2</h2>
<p style="font-size:13px; color:var(--text-dim); margin-bottom:16px;">
<strong style="color:var(--text-bright);">Det≥2</strong> = judge confirmed bug found.
<strong style="color:var(--text-bright);">SNR</strong> = true positives / total comments (higher = less noise).
<strong style="color:var(--text-bright);">FP</strong> = false positive comments.
</p>
<div style="overflow-x:auto; border:1px solid var(--border); border-radius:8px;">
<table class="data-table">
<thead>
<tr>
<th>Tool</th>
<th>Context</th>
<th>Det≥2</th>
<th>Det=3</th>
<th>Quality</th>
<th>SNR</th>
<th>FP</th>
<th>Novel</th>
</tr>
</thead>
<tbody>
<tr>
<td class="val-bright">Opus 4.6 <span class="val-dim">(prompt v1)</span></td>
<td>diff-only</td>
<td class="val-green">31%</td>
<td class="val-green">19%</td>
<td class="val-mono">1.99</td>
<td class="val-mono">1.55</td>
<td class="val-mono">22</td>
<td class="val-mono">166</td>
</tr>
<tr>
<td class="val-bright">Opus 4.6 <span class="val-dim">(prompt v1)</span></td>
<td>diff+repo</td>
<td class="val-green">31%</td>
<td class="val-mono">14%</td>
<td class="val-mono">2.25</td>
<td class="val-mono">1.20</td>
<td class="val-mono">30</td>
<td class="val-mono">202</td>
</tr>
<tr>
<td class="val-bright">Opus 4.6 <span class="val-dim">(prompt v2)</span></td>
<td>diff+repo</td>
<td class="val-blue">28%</td>
<td class="val-mono">17%</td>
<td class="val-mono val-green">2.36</td>
<td class="val-mono val-green">2.75</td>
<td class="val-mono val-green">12</td>
<td class="val-mono val-green">298</td>
</tr>
<tr>
<td class="val-bright">Sonnet 4.6</td>
<td>diff-only</td>
<td class="val-blue">27%</td>
<td class="val-mono">15%</td>
<td class="val-mono">1.95</td>
<td class="val-mono">0.57</td>
<td class="val-mono">49</td>
<td class="val-mono">126</td>
</tr>
<tr>
<td class="val-bright">Copilot</td>
<td>PR</td>
<td>23%</td>
<td class="val-mono">7%</td>
<td class="val-mono">1.94</td>
<td class="val-mono">0.75</td>
<td class="val-mono">32</td>
<td class="val-mono">192</td>
</tr>
<tr>
<td class="val-bright">Sonnet 4.6</td>
<td>diff+repo</td>
<td>23%</td>
<td class="val-mono">16%</td>
<td class="val-mono">2.03</td>
<td class="val-mono">0.74</td>
<td class="val-mono">46</td>
<td class="val-mono">168</td>
</tr>
<tr>
<td class="val-bright">Greptile <span class="caveat caveat-red" style="margin-left:0;">Rate limited</span></td>
<td>PR</td>
<td class="val-dim">11%</td>
<td class="val-dim val-mono">—</td>
<td class="val-mono">2.02</td>
<td class="val-dim val-mono">—</td>
<td class="val-dim val-mono">—</td>
<td class="val-dim val-mono">—</td>
</tr>
<tr>
<td class="val-bright">CodeRabbit <span class="caveat caveat-red" style="margin-left:0;">Rate limited</span></td>
<td>PR</td>
<td class="val-red">4%</td>
<td class="val-dim val-mono">—</td>
<td class="val-mono">0.19</td>
<td class="val-dim val-mono">—</td>
<td class="val-dim val-mono">—</td>
<td class="val-dim val-mono">—</td>
</tr>
</tbody>
</table>
</div>
<div class="callout" style="margin-top:12px;">
<p><strong style="color:var(--green);">Opus diff-only matches diff+repo on catch rate (31%)</strong> with fewer false positives (22 vs 30) and better precision on det=3 (19% vs 14%). Repo access adds quality (2.25) and novel findings (202) but also noise. <strong style="color:var(--purple);">Opus (prompt v2) has the best signal-to-noise (2.75) and most novel findings (298)</strong> — the structured approach produces the cleanest reviews.</p>
</div>
<p style="font-size:11px; color:var(--text-muted); margin-top:12px;">Det≥2 = LLM judge confirmed tool identified the bug (stricter than mechanical ±10-line proximity used in pilot report).</p>
</div>
</div>
</section>
<!-- ============================================ -->
<!-- SLIDE 8: More Context Can Hurt -->
<!-- ============================================ -->
<section class="slide">
<div class="content">
<div class="stagger">
<div class="section-label label-blue">Key Finding</div>
<h2 class="slide-title">Repo access adds quality, not catch rate</h2>
<p class="slide-subtitle" style="margin-bottom:28px;">Diff-only matches or beats diff+repo on detection. Repo access adds novel findings and review depth — but also more false positives.</p>
<div class="card-row card-row-2" style="margin-bottom:24px;">
<div class="card" style="text-align:center;">
<div class="section-label label-green">Opus diff-only</div>
<div style="display:flex; gap:24px; justify-content:center; margin:12px 0;">
<div><div class="stat-value val-green" style="font-size:28px;">31%</div><div class="stat-label">Catch</div></div>
<div><div class="stat-value val-green" style="font-size:28px;">19%</div><div class="stat-label">Det=3</div></div>
<div><div class="stat-value val-green" style="font-size:28px;">22</div><div class="stat-label">FP</div></div>
</div>
<p style="font-size:12px; color:var(--text-muted);">Focused. Fewer false positives.<br>Best precision on actionable fixes.</p>
</div>
<div class="card" style="text-align:center;">
<div class="section-label label-blue">Opus diff+repo</div>
<div style="display:flex; gap:24px; justify-content:center; margin:12px 0;">
<div><div class="stat-value val-blue" style="font-size:28px;">31%</div><div class="stat-label">Catch</div></div>
<div><div class="stat-value" style="font-size:28px; color:var(--text-dim);">14%</div><div class="stat-label">Det=3</div></div>
<div><div class="stat-value" style="font-size:28px; color:var(--text-dim);">30</div><div class="stat-label">FP</div></div>
</div>
<p style="font-size:12px; color:var(--text-muted);">More novel findings (202 vs 166).<br>Higher quality (2.25 vs 1.99). More noise.</p>
</div>