File size: 201,415 Bytes
077605d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
[
  {
    "id": "IF01",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the slides use English. Verification method: check whether the titles, body text, chart labels, table text and speaker notes of the 8 slides are all in English. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: create an eight-slide English PPTX (16:9).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G01"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF02",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the PPTX contains exactly 8 slides. Verification method: open artifact/slides.pptx and count the slides, confirming it equals 8. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: create an eight-slide English PPTX (16:9).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G02"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF03",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the PPTX page aspect ratio is 16:9. Verification method: read the slide page width and height and compute whether width/height is 16/9. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: create an eight-slide English PPTX (16:9).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G02"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF04",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: all accompanying PNG images are 1280×720. Verification method: read the pixel dimensions of 01.png–08.png one by one, confirming all are 1280 wide and 720 high. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: with matching 1280×720 PNGs.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G02"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF05",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the output contains 8 PNG files named 01.png through 08.png. Verification method: check whether the number of PNG files in the output directory and their filenames are 01.png, 02.png, …, 08.png. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 13: Deliver artifact/slides.pptx, 01.png–08.png and sources.md.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G03"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF06",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: each PNG matches and corresponds to the PPTX slide of the same order. Verification method: compare 01.png–08.png respectively with the visible content of PPTX pages 1–8, confirming a page-by-page correspondence rather than a wrong order or unrelated images. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: with matching 1280×720 PNGs.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G04"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF07",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: deliver artifact/slides.pptx. Verification method: check that the artifact/slides.pptx file exists and opens normally. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 13: Deliver artifact/slides.pptx, 01.png–08.png and sources.md.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G05"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF08",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: deliver sources.md. Verification method: check that the sources.md file exists and is readable. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 13: Deliver artifact/slides.pptx, 01.png–08.png and sources.md.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G06"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF09",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the content includes an organization of the research problem or a problem statement. Verification method: check whether the opening or a relevant page has a clear problem statement/research question, rather than merely listing methods or results. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 3: Organize the problem, method, experimental evidence and applicability limits yourself..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G07"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF10",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain that W0 is frozen. Verification method: check whether the method page clearly labels W0 as frozen and not involved in training. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain frozen W0, trainable low-rank A/B and addition of the two branches..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF11",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain that the low-rank matrices A/B are trainable. Verification method: check whether the method page clearly states that A/B are trainable low-rank matrices. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain frozen W0, trainable low-rank A/B and addition of the two branches..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF12",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain the update form in which the two branches are added. Verification method: check whether the method page shows an expression, or equivalent wording, in which the W0 branch and the low-rank A/B branch are added. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain frozen W0, trainable low-rank A/B and addition of the two branches..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF13",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain the matrix dimensions. Verification method: check whether the method page marks the matrix dimensions relevant to LoRA, such as d, k, r and their multiplication flow. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain matrix dimensions, rank and scaling with at least one original method diagram..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF14",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain the meaning of rank. Verification method: check whether it explains the significance of the low rank r, for example that r is far smaller than the input/output dimensions or the purpose of the low-rank decomposition. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain matrix dimensions, rank and scaling with at least one original method diagram..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF15",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain the scaling relationship. Verification method: check whether an α/r scaling term or an equivalent scaling explanation appears. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain matrix dimensions, rank and scaling with at least one original method diagram..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF16",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: include at least one method diagram drawn by oneself. Verification method: check whether the method diagram is a self-drawn schematic, rather than being replaced solely by the paper's original figure or a screenshot. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain matrix dimensions, rank and scaling with at least one original method diagram..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF17",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the method diagram is labeled as adapted. Verification method: check whether there is an adapted or equivalent adaptation label near the method diagram. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: draw a method diagram and label it as adapted..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G09"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF18",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain that merging weights can avoid an additional inference branch. Verification method: check whether it explains that merging the low-rank increment back into W0 can remove the additional inference branch. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain why merging weights can avoid an additional inference branch and the resulting limitation on mixed-task batches..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF19",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: explain the resulting limitation on mixed-task batches after merging. Verification method: check whether it explains that merging weights means different tasks/batches cannot directly share the same merged model, or that mixed-batch inference is inconvenient. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Explain why merging weights can avoid an additional inference branch and the resulting limitation on mixed-task batches..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G08"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF20",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: must not imply that fewer trainable parameters cause the base model to be eliminated at deployment. Verification method: read through all statements about deployment, parameter size and model files, confirming that the LoRA incremental parameters are not presented as able to replace the base model. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 5: Do not imply that fewer trainable parameters eliminate the base model at deployment..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G10"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF21",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: the experimental section uses the two models GPT-2 Medium and GPT-2 Large from Table 3. Verification method: check whether the experiment page shows both GPT-2 Medium and GPT-2 Large. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Use GPT-2 Medium and Large from Table 3, comparing full fine-tuning and LoRA separately for each..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF22",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: compare full fine-tuning and LoRA separately for GPT-2 Medium and GPT-2 Large. Verification method: check whether Medium and Large each have a paired comparison of full fine-tuning and LoRA, rather than being merged or comparing only one model. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Use GPT-2 Medium and Large from Table 3, comparing full fine-tuning and LoRA separately for each..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF23",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: show trainable parameter counts. Verification method: check whether the experiment table or text includes a column of/values for trainable parameter counts. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Show trainable parameter counts, BLEU and ROUGE-L, retaining model scale, E2E NLG task and metric names..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF24",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: show BLEU values and retain the BLEU metric name. Verification method: check whether the experiment table/chart has a BLEU label and corresponding values. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Show trainable parameter counts, BLEU and ROUGE-L, retaining model scale, E2E NLG task and metric names..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF25",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: show ROUGE-L values and retain the ROUGE-L metric name. Verification method: check whether the experiment table/chart has a ROUGE-L label and corresponding values. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Show trainable parameter counts, BLEU and ROUGE-L, retaining model scale, E2E NLG task and metric names..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF26",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: retain the model scale labels, such as Medium/Large. Verification method: check whether the experiment table/chart clearly labels the model scale of GPT-2 Medium/Large. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Show trainable parameter counts, BLEU and ROUGE-L, retaining model scale, E2E NLG task and metric names..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF27",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: retain the E2E NLG task name. Verification method: check whether the experiment page shows the E2E NLG task name, rather than only writing the generic \"task\". Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Show trainable parameter counts, BLEU and ROUGE-L, retaining model scale, E2E NLG task and metric names..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF28",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: if a parameter reduction factor is shown, it should be calculated from the specified Table 3 and labeled as a calculated value. Verification method: check only whether the actually used reduction factor is calculated from the training parameter values of the corresponding model in Table 3, with the calculation and source labeled; if no factor is shown, the condition is not triggered, answer Yes and explain. Adding an extra factor is not required. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Calculate any reduction factors from this table and label them; do not reuse GPT-3's 10,000× figure..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF29",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: do not reuse GPT-3's 10,000× figure. Verification method: considering the full semantics of the slides and sources.md, check whether the parameter reduction calculation and conclusion for GPT-2 Medium/Large in this task are based only on the specified Table 3, without borrowing GPT-3's 10,000× or an equivalent reduction value. Explicitly stating that the figure is not adopted and explaining why it does not apply does not count as reuse; No must not be judged merely because the number appears, and adding such an explanation is not required. Pass criterion: fully satisfies this requirement; missing, incorrect or partial satisfaction is No. Evaluation tool failure must not be grounds for No. Verification boundary: verify against the frozen task statement and necessary originals/ground truth; the author's own statement does not replace product evidence. Task basis: prompt.md original line 7: Calculate any reduction factors from this table and label them; do not reuse GPT-3's 10,000× figure..",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF30",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Do not interpret the ± error range as a certain upper bound. Verification method: Inspect wording involving error ranges/± and confirm it is not stated as a certain upper bound or upper-limit guarantee. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 7: Error ranges may be omitted, but do not interpret ± results as certain upper bounds or claim new significance tests.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G11"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF31",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Do not claim that a new significance test was performed. Verification method: Inspect the conclusions and table notes, and confirm no claim of a new significance test or new significance conclusion is made. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 7: Error ranges may be omitted, but do not interpret ± results as certain upper bounds or claim new significance tests.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G10"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF32",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Explain at least one efficiency benefit from §4.2. Verification method: Check whether the efficiency-related pages explicitly select and explain one efficiency benefit from §4.2. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: Also explain one efficiency benefit from §4.2 with its experimental conditions, and at least one limitation.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF33",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: State the experimental conditions corresponding to this §4.2 efficiency benefit. Verification method: Check whether the conditions corresponding to this benefit are given, such as model scale, optimizer, rank/module, or training setup. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: Also explain one efficiency benefit from §4.2 with its experimental conditions, and at least one limitation.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF34",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: State at least one limitation. Verification method: Check whether the limitations page or limitations block has a clear statement of one or more limitations. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: Also explain one efficiency benefit from §4.2 with its experimental conditions, and at least one limitation.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF35",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: If no additional inference latency is mentioned, the weight-merging prerequisite must also be stated. Verification method: Search for latency-related statements; if any exist, check whether the same page or the immediate vicinity states that weights must be merged first. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: If mentioning no additional inference latency, state the weight-merging prerequisite.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF36",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Do not include later work such as QLoRA in the main presentation or comparison scope. Verification method: Search all slides and sources.md, and confirm that QLoRA or other later methods are not presented as main content or as a comparison object. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: Later work such as QLoRA is outside scope.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF37",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Distinguish paper-reported values from self-calculated values. Verification method: Check whether reported values and calculated values in tables/text are distinguished by labels, footnotes, or layout. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 9: Distinguish paper-reported values from your calculations.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G12"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF38",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Every page has an identifiable speaking point. Verification method: Check each page's on-page points or speaker notes, and confirm every page has an identifiable speaking point. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 11: Each slide needs a speaking point.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G13"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF39",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Use editable native tables for experimental data. Verification method: Select the experiment tables in the PPTX and confirm they are native table objects and editable, rather than table screenshots. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 11: Use editable native tables for experiments and editable vector objects for the method diagram.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G13"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF40",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Use editable vector objects for the method diagram. Verification method: Select the method diagram elements in the PPTX and confirm they are editable vector shapes/object groups, rather than bitmaps. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 11: Use editable native tables for experiments and editable vector objects for the method diagram.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G13"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF41",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Avoid using screenshots of long paragraphs. Verification method: Check each page for whether long paragraph text is placed as screenshots rather than distilled key points. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 11: avoid screenshots of long paragraphs.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G13"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF42",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Do not use whole-page paper screenshots as substitutes for slides. Verification method: Check each page for cases where whole-page screenshots of paper pages serve as the main content or substitute for the slide layout. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 11: Local excerpts of paper figures are allowed, but whole-page screenshots must not substitute for slides.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G13"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF43",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Input files remain read-only and must not be rewritten. [v7 verifiable boundary: check that the final inputs match the frozen snapshot file by file; it is not required to prove that no write ever occurred during production. Public runtime inputs are guaranteed by read-only mounts.] Verification method: Compare the content or hashes of the final input files with the frozen originals, and check integrity together with actual input-isolation evidence; do not judge whether content changed solely by timestamps or file permissions, and do not infer from final consistency that no write ever occurred during production. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 13 excerpt: Inputs are read-only[…].",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G14"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF45",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Use the fixed version 2106.09685v2. Verification method: Check whether citations and content basis explicitly correspond to 2106.09685v2. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: Use fixed version 2106.09685v2, 26 pages; do not mix revisions.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G09"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF46",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Do not mix in content from other revisions. Verification method: Check whether the versions cited in sources.md and the slides are uniformly v2, with no mixing of page numbers/content from other revisions. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: Use fixed version 2106.09685v2, 26 pages; do not mix revisions.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G09"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF47",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Page-number citations refer to PDF-file pages, not the paper's printed page numbers. Verification method: Spot-check the correspondence between page numbers in sources.md and PDF-file pages, and confirm PDF-file pages numbering is used. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: Page numbers refer to PDF-file pages.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G09"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF48",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Content focuses on PDF pages 4–5 and page 7; other pages serve only as necessary background. Verification method: Check whether the main method, efficiency, and experimental content comes from PDF pages 4–5 and page 7 where Table 3/§5.4 is located, with only minimal background citation from other pages. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 7: Focus on PDF pages 4–5 (§4.1–4.2) and page 7 (Table 3/§5.4). Use other pages only for necessary context; a review of all experiments is not required.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G15"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF49",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Cite the paper authors. Verification method: Check whether author information for the paper appears in the slides or sources.md. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: Cite authors, title and version.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G16"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF50",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: Cite the paper title. Verification method: Check whether the paper title appears in the slides or sources.md. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Material basis (relative to inputs/): input/sources-and-scope.md original line 5: Cite authors, title and version.",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G16"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF51",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: sources.md records version information 2106.09685v2. Verification method: Check whether sources.md contains a 2106.09685v2 version record. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 13: sources.md (page/table/version/calculation locations).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G06"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF52",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: sources.md marks page locations. Verification method: Check whether sources.md gives the page numbers or page ranges of the content used. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 13: sources.md (page/table/version/calculation locations).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G06"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF53",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: sources.md marks table-number locations. Verification method: Check whether sources.md gives table-number locations such as Table 3. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 13: sources.md (page/table/version/calculation locations).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G06"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "IF54",
    "dimension": "Task Fulfillment / Compliance",
    "criterion": "Requirement: sources.md marks calculation locations. Verification method: Check whether sources.md explains the location of the basis or calculation process for calculations such as the reduction factor. Pass criterion: Fully meets this requirement; missing, incorrect, or partially met is No. Evaluation-tool failure cannot be grounds for No. Verification boundary: Verify against the frozen prompt and necessary originals/ground truth; author self-report does not substitute for artifact evidence. Task basis: prompt.md original line 13: sources.md (page/table/version/calculation locations).",
    "answers": {
      "Yes": "Complete all necessary checks for this criterion, with actual evidence supporting satisfaction of the requirement; equivalent implementations within the task's scope are allowed.",
      "No": "At least one necessary requirement is not satisfied; identify evidence of actual omissions, errors, or partial satisfaction."
    },
    "compliance_groups": [
      "G06"
    ],
    "benchmark_source": "vas"
  },
  {
    "id": "C1",
    "dimension": "Content",
    "criterion": "Accurate expression of LoRA mechanism terminology and formulas.\nObject of measurement: Accurately explain frozen W0, trainable low-rank A/B, and the additive update h=W0x+BAx, and correctly state the matrix dimensions and that r is much smaller than min(d,k); terminology and formulas suit an audience with a deep learning background.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check whether the method page shows W0, A, B, and h=W0x+BAx or equivalent expressions together. Verify whether the dimension labels for d, k, r and the matrix multiplication order are consistent. Check whether it explicitly states that W0 is frozen, A/B are trainable, and r is much smaller than min(d,k).\nScoring boundary: Whether terminology, concepts, and wording let this task's audience understand accurately; IF checks factual implementation, and C1 does not award points based on the number of terms.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most easily confused concepts and qualifications remain concise and accurate, and summary, details, and actual examples show no semantic drift.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: The core formula relationship is explained incorrectly, such as saying A/B replaces W0, multiplying the two branches instead of adding them, or dimensions being inconsistent with the given multiplication. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: The formula is correct but key objects are misplaced, such as not stating that W0 is frozen or describing r as comparable to the original dimensions. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: There is one vague/missing label for dimensions, rank, or scaling, so the audience must guess how the multiplication or scaling holds. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: The mechanism, freezing, and additive relationship are correct, and dimensions are explained; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: the mechanism is expressed clearly, dimension and rank relationships are consistently labeled, and no redundant confusion appears. Quality improvement must be reflected as: easily confused terms, quantities, or references receive brief clarification where they appear, reducing the need to guess. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: formula, dimensions, and diagram corroborate each other, and one or two terms are used to explain why the low-rank update is effective. Quality improvement must be reflected as: definitions, qualifications, and actual labels support each other, and complex meanings require no cross-region assembly. the mechanism is presented maturely and concisely: formula, dimensions, rank, and scaling are all self-consistent and directly serve the subsequent experimental and deployment discussion. The most easily confused concepts and qualifications remain concise and accurate, and summary, details, and actual examples show no semantic drift. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Accurately explain frozen W0, trainable low-rank A/B, and the additive update h=W0x+BAx, and correctly state the matrix dimensions and that r is much smaller than min(d,k); terminology and formulas suit an audience with a deep learning background. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most easily confused concepts and qualifications remain concise and accurate, and summary, details, and actual examples show no semantic drift. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. An unusually economical conceptual formulation makes a genuinely confusable distinction directly understandable, with less interpretation than an equally accurate, concise conventional explanation. Identify the distinction and the wording/representation that produces this extra understanding; merely correct terminology or fluent prose is not exceptional. Apply the exceptional benefit specifically to: Accurately explain frozen W0, trainable low-rank A/B, and the additive update h=W0x+BAx, and correctly state the matrix dimensions and that r is much smaller than min(d,k); terminology and formulas suit an audience with a deep learning background. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most easily confused concepts and qualifications remain concise and accurate, and summary, details, and actual examples show no semantic drift. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 9,
    "benchmark_source": "vas"
  },
  {
    "id": "C2",
    "dimension": "Content",
    "criterion": "Truthfulness of experimental data and efficiency evidence.\nObject of measurement: Table 3's trainable parameters, BLEU, and ROUGE-L for GPT-2 M/L under FT and LoRA must match the paper; §4.2 efficiency benefits and their experimental conditions must be traceable; derived multipliers must be calculated correctly and have formulas.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check cell by cell whether the parameters, BLEU, and ROUGE-L of the four methods in Table 3 are consistent with the given facts. Check whether the E2E NLG task name, GPT-2 Medium/Large scales, and metric names are retained. Check whether the multipliers approximately 1014.06 and 1005.23 are calculated from the table's parameters and labeled as calculated values. Check whether the §4.2 efficiency benefit carries experimental conditions (such as GPT-3 175B, Adam, r=4, query/value only).\nScoring boundary: The support relationship between task claims and evidence/data/actual process; does not repeatedly evaluate graphic encoding or whether source documents are easy to find.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex key comparisons or state chains can still be checked item by item; approximations, limitations, and counterexamples are not obscured by conclusions.\nScenario reference · content quality (adds no new threshold): LoRA freezes W0, trains low-rank A/B, and applies an additive update; dimensions and α/r; merging avoids an additional inference branch and its mixed-batch cost; FT and LoRA parameters, BLEU, ROUGE-L, and multipliers for GPT-2 M/L; §4.2 memory or checkpoint benefits and conditions; at least one limitation.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to reproduce the paper's original layout or do a full-paper translation.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to present all efficiency facts of §4.2 at once or to explain the difference between 25% and 32.5→43.1.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required that every fact appear on screen; facts not required by the prompt to be presented must not be grounds for deduction.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Paper tables and factual values must maintain the prompt-specified conventions: E2E NLG, GPT-2 Medium/Large, trainable parameters M, BLEU, ROUGE-L.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Paper-reported values and self-calculated values must be distinguishable.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Core experimental values are largely wrong or mismatched to the model/method, or fabricated benefits are presented as paper results. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: At least one row's key values are mismatched (such as model scale or FT/LoRA swapped), or a derived multiplier is unfounded and wrong. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: Values are basically correct but units, task, or metric names are missing/miswritten, or efficiency conditions are incompletely stated. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: GPT-2 Medium/Large parameters and BLEU/ROUGE-L correspond accurately, efficiency benefits carry experimental conditions, and derived multipliers can be recomputed from Table 3 and are marked as calculated; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: values and metric names are accurate, efficiency benefits are accompanied by conditions, and derived multipliers are clearly labeled. Quality improvement must be reflected as: key claims correspond nearby to the values, conditions, or process evidence used, making checking more direct. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: all necessary values are precise to display precision, derived multiplier formulas and sources are reviewable, and paper-reported values and calculated values are separated. Quality improvement must be reflected as: evidence selection, comparison conventions, and conclusion strength work together, making clear what can and cannot be inferred. data presentation is mature: tables are precise, units and error ranges are handled honestly, derived calculations are reviewable, and the conditions and conventions of efficiency evidence are unambiguous. The most complex key comparisons or state chains can still be checked item by item; approximations, limitations, and counterexamples are not obscured by conclusions. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Table 3's trainable parameters, BLEU, and ROUGE-L for GPT-2 M/L under FT and LoRA must match the paper; §4.2 efficiency benefits and their experimental conditions must be traceable; derived multipliers must be calculated correctly and have formulas. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex key comparisons or state chains can still be checked item by item; approximations, limitations, and counterexamples are not obscured by conclusions. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A non-obvious presentation of the evidence exposes an important conditional relationship that a flawless fact-and-citation account would leave the reader to reconstruct. Show the additional understanding and its support, without inventing a claim, requiring more data, or rewarding the count of sources. Apply the exceptional benefit specifically to: Table 3's trainable parameters, BLEU, and ROUGE-L for GPT-2 M/L under FT and LoRA must match the paper; §4.2 efficiency benefits and their experimental conditions must be traceable; derived multipliers must be calculated correctly and have formulas. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex key comparisons or state chains can still be checked item by item; approximations, limitations, and counterexamples are not obscured by conclusions. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 9,
    "benchmark_source": "vas"
  },
  {
    "id": "C3",
    "dimension": "Content",
    "criterion": "Scope choices, qualifying tone, and applicable boundaries.\nObject of measurement: Select information according to the seminar's purpose, state the merged-inference prerequisite and mixed-batch limitation, and at least one limitation; do not describe fewer trainable parameters as the disappearance of the base model, and do not introduce later work such as QLoRA.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check whether the base model weights are distinguished from the LoRA incremental checkpoint size. If no additional latency is mentioned, does the same page or immediate vicinity state the weight pre-merging prerequisite? Check whether it states that after merging, mixed batching of different A/B tasks is not directly convenient. Check whether out-of-scope content such as QLoRA appears.\nScoring boundary: Content selection, density, tone, and necessary qualifications around the scenario purpose; not font size or visual hierarchy on the page.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most information-dense places still have appropriate selection, key points and exceptions are understandable, and there is no redundant expansion or oversimplification.\nTask scenario: For a graduate academic seminar with a deep learning background, English 16:9 slides, for oral presentation with speaking-point notes; LoRA v2 paper method, experiments, and applicable boundaries need to be extracted and delivered as an editable PPTX and corresponding PNGs.\nScenario reference · audience (adds no new threshold): Graduate students and researchers who know basic deep learning and are familiar with Transformer and fine-tuning concepts but have not systematically read LoRA.\nScenario reference · purpose (adds no new threshold): About a 15–20 minute group meeting report, the speaker presents page by page, and the audience needs to quickly understand low-rank decomposition, merged inference, and actual experimental comparisons.\nScenario reference · content selection (adds no new threshold): A compact argument of problem—method—experiment—limitations; formulas and table data are checkable; paper-reported values and self-calculated values are clearly distinguished; no piles of long paragraphs and no whole-page screenshots.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to include initialization details (A Gaussian, B zero).\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to show all error ranges or perform significance tests.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to discuss later work such as QLoRA or add the latest external knowledge.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Use only the fixed version arXiv 2106.09685v2, and do not mix in other revisions.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Must not treat ± errors in the table as certain upper bounds, and must not claim new significance tests.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Must not use GPT-3's 10,000× parameter reduction in place of the derived multipliers from GPT-2 Table 3.\nApplicable boundary for this task (all relevant items apply; no new scoring items): Must not imply that fewer trainable parameters mean the base model is unnecessary at deployment.\nApplicable boundary for this task (all relevant items apply; no new scoring items): If no additional inference latency is mentioned, the prerequisite of pre-merging weights must also be stated.\nContent focus: LoRA W0/A/B additive structure and scaling, Table 3 for two GPT-2 scales, parameter ratios, and merging conditions\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Going out of scope to present later work as the main topic or claiming that LoRA deployment does not need the base model, causing core misleading. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Claiming no additional inference latency while omitting the merging prerequisite, or giving no limitations at all. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: The limitation or mixed-batch restriction is only mentioned in passing and vaguely worded, easily causing the audience to misunderstand deployment cost. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Mentions the merging prerequisite and at least one limitation, and does not equate the scaled parameter volume with the entire model volume; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: qualifying tone is clear, limitations and benefits are presented side by side, and conclusions are not exaggerated. Quality improvement must be reflected as: remove restatement that does not aid understanding or place necessary qualifications where needed, making the key points more direct. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: benefit—prerequisite—limitation form a clear boundary, explicitly stating the cost of mixed-task batches and the alternative of dynamically selecting modules. Quality improvement must be reflected as: emphasis, depth of explanation, and tone suit the audience; streamlined content still retains necessary conditions. selection is mature: qualify only where necessary, clearly explain both deployment advantages and their conditions and scenario boundaries, with a tone appropriate to a graduate seminar. Even the most information-dense places still have appropriate selection, key points and exceptions are understandable, and there is no redundant expansion or oversimplification. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Select information according to the seminar's purpose, state the merged-inference prerequisite and mixed-batch limitation, and at least one limitation; do not describe fewer trainable parameters as the disappearance of the base model, and do not introduce later work such as QLoRA. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most information-dense places still have appropriate selection, key points and exceptions are understandable, and there is no redundant expansion or oversimplification. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. An unusually incisive selection resolves the tension between necessary qualifications and the audience’s main purpose: it conveys more useful understanding with no loss of required substance. Explain why a conventionally fully refined summary would still impose more interpretation; brevity, completeness or promotional impact alone is insufficient. Apply the exceptional benefit specifically to: Select information according to the seminar's purpose, state the merged-inference prerequisite and mixed-batch limitation, and at least one limitation; do not describe fewer trainable parameters as the disappearance of the base model, and do not introduce later work such as QLoRA. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most information-dense places still have appropriate selection, key points and exceptions are understandable, and there is no redundant expansion or oversimplification. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 7,
    "benchmark_source": "vas"
  },
  {
    "id": "O1",
    "dimension": "Organization",
    "criterion": "Eight-page argument path.\nObject of measurement: The whole deck should form a coherent path of problem—method—experimental evidence—efficiency—limitations, with each page's selection serving the seminar audience's understanding, and no isolated or repeated pages.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: List the eight page titles and each page's core argument, and check whether problem, method, experiment, efficiency, and limitations are covered. Check whether experiment pages appear after the method explanation. Check whether there are repeated arguments or purely decorative pages.\nScoring boundary: The logic of the overall argument/decision/operation/concept path, not scoring by a fixed structural template.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex branches or comparisons can still be incorporated into the main line, the closure is grounded, and there are no dangling steps or repeated detours.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Page order is chaotic, method and experiment are interleaved so the argument cannot be reconstructed, or a core segment is missing an entire page. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: There is roughly an order but key transitions are broken, such as not establishing the evaluation object before experiments, or a limitations page unrelated to the body. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: The structure is discernible but there are obvious redundant pages or some segment is in an inappropriate position, affecting the rhythm of understanding. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Organizes a reasonable order of problem, method, experiment, and limitations as required by the prompt; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: the path is clear, pages have explicit progression relationships, and no functionless pages appear. Quality improvement must be reflected as: one key transition clearly connects the premise to the next step, making the main line easier to follow. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: each page carries an independent argumentative role, method and experiment echo each other, and limitations naturally follow the efficiency discussion. Quality improvement must be reflected as: the premise, development, and conclusion of each part refer back to each other, and order and length serve this task's explanation or decision. the eight pages form a compact and forceful academic narrative, with decisive selection, front and back pages supporting each other, and no place that could be deleted without damaging the argument. The most complex branches or comparisons can still be incorporated into the main line, the closure is grounded, and there are no dangling steps or repeated detours. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: The whole deck should form a coherent path of problem—method—experimental evidence—efficiency—limitations, with each page's selection serving the seminar audience's understanding, and no isolated or repeated pages. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex branches or comparisons can still be incorporated into the main line, the closure is grounded, and there are no dangling steps or repeated detours. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A non-obvious organization of the existing argument or user path makes separate necessary parts illuminate one another and materially reduces reasoning or navigation steps. Compare the benefit with an equally complete, logically sound conventional order; no additional section or prescribed order is required. Apply the exceptional benefit specifically to: The whole deck should form a coherent path of problem—method—experimental evidence—efficiency—limitations, with each page's selection serving the seminar audience's understanding, and no isolated or repeated pages. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex branches or comparisons can still be incorporated into the main line, the closure is grounded, and there are no dangling steps or repeated detours. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 6,
    "benchmark_source": "vas"
  },
  {
    "id": "O2",
    "dimension": "Organization",
    "criterion": "Within-page semantic development and local connections.\nObject of measurement: Within a single page, formulas, diagrams, tables, and text form a readable local argument; title and body, labels and graphics correspond to each other; multiple unrelated pieces of information are not piled onto one page; cross-page semantic continuity is not judged.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Randomly select a method page and an experiment page, and check whether the title foreshadows the content and whether labels point to the right objects. Check whether there is variable explanation next to formulas and reading guidance next to tables. Check whether there is extra information unrelated to the topic.\nScoring boundary: The semantic order and relationships of explanation within one page, stage, node, or comparison unit.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most complex unit can be understood in the needed order, and local explanations are concise, self-contained, and consistent with the overall main line.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Information within the page does not connect, or graphics and text point in opposite directions, so readers cannot infer the content's purpose from the title. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Multiple items are juxtaposed but lack organization, formulas and explanations are disconnected, and the local argument does not hold. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: The page's main line is discernible but labels are missing or the order jumps, requiring repeated looking back. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Each page has one theme and several supporting pieces of information, and local relationships are readable; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: title, key points, and formulas/diagrams form clear supporting relationships, and label positions are appropriate. Quality improvement must be reflected as: the arrangement of objects, explanations, and results in a local unit reduces one necessary look-back. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the page organizes the argument around a diagram or table, and formulas, dimension explanations, and text complement one another. Quality improvement must be reflected as: the diagrams, text, conditions, and conclusions of each main unit form relationships understandable in place. local development is mature: within one page, progression from concept to evidence is natural, and readers can reach the same conclusion along the visual order without relying on the speaker. Even the most complex unit can be understood in the needed order, and local explanations are concise, self-contained, and consistent with the overall main line. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Within a single page, formulas, diagrams, tables, and text form a readable local argument; title and body, labels and graphics correspond to each other; multiple unrelated pieces of information are not piled onto one page; cross-page semantic continuity is not judged. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most complex unit can be understood in the needed order, and local explanations are concise, self-contained, and consistent with the overall main line. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. An economical local organization makes a difficult relationship among conditions, evidence and objects understandable as one unit rather than assembled mentally from carefully grouped parts. Identify the semantic reconstruction it eliminates; proximity and tidy grouping alone do not qualify. Apply the exceptional benefit specifically to: Within a single page, formulas, diagrams, tables, and text form a readable local argument; title and body, labels and graphics correspond to each other; multiple unrelated pieces of information are not piled onto one page; cross-page semantic continuity is not judged. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the most complex unit can be understood in the needed order, and local explanations are concise, self-contained, and consistent with the overall main line. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 5,
    "benchmark_source": "vas"
  },
  {
    "id": "O3",
    "dimension": "Organization",
    "criterion": "Cross-page identity and terminology inheritance.\nObject of measurement: Maintain consistent terminology and notation across pages for W0, A/B, r, α, trainable parameters, merging, etc.; the naming of the same model/method on experiment pages and efficiency pages does not drift; and judge whether semantics connect between pages.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Spot-check whether W0, A, B, r, α are synonymous on the method page and the limitations page. Check whether GPT-2 M/L is unified in tables and text. Check whether metric names such as BLEU/ROUGE-L are spelled consistently.\nScoring boundary: Cross-page/cross-state semantic inheritance, or closure of a single page's branch paths and identity/action; does not force adding a conclusion.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most demanding transitions in the presented content still maintain identity and causal continuity, and readers do not need later information to repair the current meaning.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nEvaluate meaning carried across the actual slides, not invented software branches or reset states.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: The same symbol changes meaning across pages, or the same method has multiple names causing confusion. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Key terms or model names are inconsistent across multiple pages, affecting comparison. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: An individual abbreviation is not defined at first occurrence, or a later page switches to another spelling. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: W0/A/B, rank, model scale, task, and metrics have stable identity across the eight pages, and key limitations are retained in the conclusion; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: symbols, model names, and method names are stable across pages, and first occurrences are explained. Quality improvement must be reflected as: one cross-page, cross-region, or cross-state continuation clearly retains object identity and existing meaning. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: terminology is unified throughout like a glossary, and symbols on experiment pages correspond one-to-one with those on method pages. Quality improvement must be reflected as: main continuations, branches, and endings form continuous semantics; the positions of repeated objects and necessary exceptions are traceable. cross-page identity inheritance is mature: notation, naming, and visual encoding run through the whole deck, and readers need not rebuild semantics. the most demanding transitions in the presented content still maintain identity and causal continuity, and readers do not need later information to repair the current meaning. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Maintain consistent terminology and notation across pages for W0, A/B, r, α, trainable parameters, merging, etc.; the naming of the same model/method on experiment pages and efficiency pages does not drift; and judge whether semantics connect between pages. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most demanding transitions in the presented content still maintain identity and causal continuity, and readers do not need later information to repair the current meaning. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A purposeful carry-over lets the audience preserve and develop the right meaning through the task’s actual transitions, materially reducing reconstruction beyond fully refined referential consistency. Identify what understanding accumulates; do not invent branches, endings or reset states absent from the task. Apply the exceptional benefit specifically to: Maintain consistent terminology and notation across pages for W0, A/B, r, α, trainable parameters, merging, etc.; the naming of the same model/method on experiment pages and efficiency pages does not drift; and judge whether semantics connect between pages. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most demanding transitions in the presented content still maintain identity and causal continuity, and readers do not need later information to repair the current meaning. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 4,
    "benchmark_source": "vas"
  },
  {
    "id": "V1",
    "dimension": "Visuals",
    "criterion": "Projection hierarchy and focus.\nObject of measurement: Each page has a clear primary visual focus at the 1280×720 projection scale (formula, method diagram, or data table), with distinct hierarchy among title, body, and supporting explanations.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Zoom to projection scale and check where the eye lands first on each page. Check whether the font size/weight/color of title, body, and notes form a hierarchy. Check whether the method diagram or table loses focus by being surrounded by same-level text.\nScoring boundary: The salience relationships among title, body, evidence, and supporting information; evaluate the attention entry point, core focus, and emphasis strength.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the densest region or state shows main information first and then evidence and qualifications, with no substantial attention competition between whole and part.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: The entire page has no focus, title and body have equal weight, and the audience cannot tell where to look first. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: The focus is drowned by large amounts of same-level text or decoration, and the core diagram/table does not stand out. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: The hierarchy is discernible but secondary information is too strong, requiring effort to filter in places. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Title, body, and notes are distinguished by size and position, and necessary content can be found; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: each page has a clear focus, secondary explanations clearly recede, and it can be located at a glance during projection. Quality improvement must be reflected as: a set of position, scale, or contrast choices makes the main focus clearer, and supporting information does not compete for reading. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: focus choices align with the argument's purpose, formulas/diagrams/tables serve as focus on different pages, and transitions are natural. Quality improvement must be reflected as: the strength relationships among main focus, evidence, and necessary qualifications are coordinated and adjusted reasonably with content complexity. hierarchy control is mature: focus is stable, supporting information appears when needed without stealing the scene, and the reading rhythm is consistent across the deck. Even the densest region or state shows main information first and then evidence and qualifications, with no substantial attention competition between whole and part. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Each page has a clear primary visual focus at the 1280×720 projection scale (formula, method diagram, or data table), with distinct hierarchy among title, body, and supporting explanations. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the densest region or state shows main information first and then evidence and qualifications, with no substantial attention competition between whole and part. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A non-obvious hierarchy brings the decisive relationship and its necessary qualification into attention together without competing emphasis. Demonstrate a first-look interpretation benefit beyond an impeccably clear primary/secondary hierarchy, not just stronger contrast or larger type. Apply the exceptional benefit specifically to: Each page has a clear primary visual focus at the 1280×720 projection scale (formula, method diagram, or data table), with distinct hierarchy among title, body, and supporting explanations. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the densest region or state shows main information first and then evidence and qualifications, with no substantial attention competition between whole and part. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 5,
    "benchmark_source": "vas"
  },
  {
    "id": "V2",
    "dimension": "Visuals",
    "criterion": "Block proportions, grid, and whitespace.\nObject of measurement: Block proportions are stable across pages, alignment and whitespace serve reading; tables and method diagrams are not squeezed, and pages do not lose breathing room from being stuffed with information.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check whether the four-side margins and title baselines are consistent across pages. Check whether there is enough whitespace around tables or method diagrams for recognition. Check whether any page content touches the edge or has large areas of purposeless blank space.\nScoring boundary: Area proportions, grid, edges, center of gravity, density, and whitespace between information blocks; evaluate overall composition and spatial organization.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the region with the greatest density or state change still has stable spatial order, and overall balance and local spacing hold simultaneously.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Content is crammed full or there are large areas of useless blank space, and block imbalance affects reading. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Multiple pages have chaotic alignment and inconsistent margins, and diagrams/tables are squeezed and distorted. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: There is obvious misalignment or local crowding, but the whole is still readable. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Pages have a grid and margins, and block proportions are normal; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: alignment is stable, whitespace has rhythm, and diagrams/tables and text each have their place. Quality improvement must be reflected as: block edges, scale, or whitespace have a clear purpose, improving the spatial relationships of one class of content. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the whole deck shares a consistent grid and margin system, and density changes are intentional. Quality improvement must be reflected as: area proportions, alignment, center of gravity, and whitespace jointly adapt to actual content, rather than mechanically applying the same layout. layout proportions are mature: tight but not crowded, whitespace carries grouping and rhythm, and density changes across pages are natural. Even the region with the greatest density or state change still has stable spatial order, and overall balance and local spacing hold simultaneously. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Block proportions are stable across pages, alignment and whitespace serve reading; tables and method diagrams are not squeezed, and pages do not lose breathing room from being stuffed with information. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the region with the greatest density or state change still has stable spatial order, and overall balance and local spacing hold simultaneously. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A content-specific spatial solution reconciles competing demands of overview, detail and available space with exceptional clarity and balance. Identify the real space conflict resolved beyond a flawlessly fitted conventional composition; extra whitespace, density or a fashionable layout is not itself a benefit. Apply the exceptional benefit specifically to: Block proportions are stable across pages, alignment and whitespace serve reading; tables and method diagrams are not squeezed, and pages do not lose breathing room from being stuffed with information. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the region with the greatest density or state change still has stable spatial order, and overall balance and local spacing hold simultaneously. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 8,
    "benchmark_source": "vas"
  },
  {
    "id": "V3",
    "dimension": "Visuals",
    "criterion": "Font size, line spacing, and formula readability.\nObject of measurement: Titles, body text, table values, and formulas are readable at 1280×720 projection; line spacing, line breaks, and sub/superscripts do not cause ambiguity, and long paragraphs are avoided.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check at projection scale whether body text and table values are readable. Check whether formula subscripts and fractions are compressed or clipped. Check whether long paragraphs replace key points.\nScoring boundary: The complete visibility, font levels, contrast, line spacing, and line breaks of text in the actual delivered image; evaluate text presentation and reading at the target scale.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the longest necessary labels, small qualifications, or complex symbols remain complete and easy to read, and typographic quality is consistently stable across regions.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Body text or key values are too small to read in projection, and line breaks cut formulas/terms causing misunderstanding. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Multiple pages have too-small font sizes or overly dense line spacing, and table values require leaning in to identify. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: Local font size or line spacing is inappropriate, and individual formula sub/superscripts are unclear. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Necessary content is readable at projection scale and line spacing is normal; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: font size and line spacing are coordinated, formula sub/superscripts are clear, and long sentences are reasonably split. Quality improvement must be reflected as: the font level, spacing, or line breaks of one class of necessary text reduce reading burden; merely being readable is not the basis. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: readability across the deck has a unified scale, and the typographic details of tables and formulas withstand projected reading. Quality improvement must be reflected as: font levels, line spacing, contrast, and semantic line breaks are coordinated, and dense content remains comfortably readable at the task-specified scale. typography is mature: formulas, tables, and body text are equally reliable at projection scale, and line breaking and punctuation are handled professionally. The longest necessary labels, small qualifications, or complex symbols remain complete and easy to read, and typographic quality is consistently stable across regions. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Titles, body text, table values, and formulas are readable at 1280×720 projection; line spacing, line breaks, and sub/superscripts do not cause ambiguity, and long paragraphs are avoided. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the longest necessary labels, small qualifications, or complex symbols remain complete and easy to read, and typographic quality is consistently stable across regions. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. Typographic structure makes a difficult meaning or relationship apparent as it is read, reducing parsing beyond fully refined legibility, spacing and hierarchy. Identify the specific semantic contribution at the required viewing scale; a distinctive font or decorative lettering is not sufficient. Apply the exceptional benefit specifically to: Titles, body text, table values, and formulas are readable at 1280×720 projection; line spacing, line breaks, and sub/superscripts do not cause ambiguity, and long paragraphs are avoided. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the longest necessary labels, small qualifications, or complex symbols remain complete and easy to read, and typographic quality is consistently stable across regions. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 7,
    "benchmark_source": "vas"
  },
  {
    "id": "V4",
    "dimension": "Visuals",
    "criterion": "Accuracy of graphic decoding in method diagrams and tables.\nObject of measurement: The method diagram should correctly draw the W0 branch and the low-rank A/B additive path and dimension flow; arrows, connectors, and table reading methods should not mislead, and diagrams of merged inference should be consistent with the text.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Check whether the method diagram shows two branches that are added at the output. Verify the d, k, r labels and flow directions in the diagram. Check whether the table is organized by model/method and metrics, with units clearly indicated. Check whether merging-related diagrams are consistent with the textual premise.\nScoring boundary: The legibility and semantic correspondence of graphic marks, lines, arrows, symbols, and dynamic states; evaluate accurate decoding and reliable comparison.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex connections, states, or scale relationships can still be reliably decoded, and legends and actual content have no ambiguity or misleading.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: The method-diagram path contradicts the formula, such as drawing A/B as replacing W0 or arrows pointing the wrong way. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Dimension labels are missing or arrow meanings are unclear, and readers cannot derive BAx from the diagram. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: One connection or label in the diagram is easy to misread, but overall understanding is not changed. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: The method diagram shows a discernible two-branch additive structure, and the table can be read normally; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: the two branches, dimensions, and addition node are clear, and the row-column relationships of the table are explicit. Quality improvement must be reflected as: a set of labels, connections, or contrasts makes object relationships more directly decodable, reducing diagram guessing. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the graphic language is accurate and self-explanatory: dimensions, training/frozen states, and pre/post-merge differences are correctly encoded. Quality improvement must be reflected as: graphic encoding, labels, and actual relationships support one another; complex comparisons do not need extra explanation to correct the diagram's meaning. diagramming is mature: the diagram and formula corroborate each other, and the difference between merged and unmerged states is clearly expressed with consistent geometry and labels. The most complex connections, states, or scale relationships can still be reliably decoded, and legends and actual content have no ambiguity or misleading. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: The method diagram should correctly draw the W0 branch and the low-rank A/B additive path and dimension flow; arrows, connectors, and table reading methods should not mislead, and diagrams of merged inference should be consistent with the text. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex connections, states, or scale relationships can still be reliably decoded, and legends and actual content have no ambiguity or misleading. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A visual encoding makes a difficult relationship directly perceptible instead of requiring mental reconstruction from fully correct labels and marks. Show what becomes easier to understand and verify fidelity to the source or required state; novelty and additional diagrams are not requirements. Apply the exceptional benefit specifically to: The method diagram should correctly draw the W0 branch and the low-rank A/B additive path and dimension flow; arrows, connectors, and table reading methods should not mislead, and diagrams of merged inference should be consistent with the text. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the most complex connections, states, or scale relationships can still be reliably decoded, and legends and actual content have no ambiguity or misleading. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 5,
    "benchmark_source": "vas"
  },
  {
    "id": "V5",
    "dimension": "Visuals",
    "criterion": "Academic visual language and diagram integration.\nObject of measurement: Non-color graphic abstraction, typographic tone, and method-diagram and table components should form a unified academic visual language that serves paper explanation rather than decorative piling.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Compare the line weights, fonts, and label styles of method diagrams, tables, and titles. Check whether there are decorative graphics with no information value. Check whether graphic abstraction is consistent with the formal tone of paper explanation.\nScoring boundary: Audience, purpose, medium temperament, and the integration of graphic abstraction, typography, images, and component surface language; purely color aesthetic is handed to V9.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the hardest-to-integrate content coordinates with the overall visual language; when information is complete, there is still a clear and restrained scenario expression.\nScenario reference · visual language (adds no new threshold): Restrained academic style: clear hierarchy, steady typography, limited accent colors, native tables and vector line diagrams, unified chart language, avoiding decorative visual noise.\nScenario reference · design constraints (adds no new threshold): Within eight pages, it must simultaneously accommodate the method mechanism, dimensions/scaling, merged inference conditions, two sets of GPT-2 experimental data and derived multipliers, efficiency evidence, and limitations; information density and projection readability press against each other, and incremental parameters or merged inference must not be presented as the disappearance of the base model.\nScenario reference · design goal (adds no new threshold): Each page has a single argument and forms a reasoning chain with the previous and next pages; formulas, dimensions, and diagrams corroborate each other; table values are precise and derived calculations are labeled; merging prerequisites and mixed-batch limitations are clearly explained; visual language is unified and reliable at projection scale.\nScoring boundary: V9 independently evaluates the aesthetic harmony and refinement of color; text contrast causing reading impairment belongs to V3, mismatches between color and object meaning belong to V4, and cross-page color-code identity drift belongs to V7. Cross-item attribution only when different functional impairment is proven. V5 cannot merely repeat V9's color quality, and V9 cannot directly award the professional level just because it \"fits the scenario\".\nApplicable boundary for this task (all relevant items apply; no new scoring items): Not required to use a specified color scheme, template, or font family.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: The visual language is chaotic, with diagrams, tables, and icons each doing their own thing, like a collage of templates. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Multiple style conflicts, such as cartoon icons paired with formal formulas and data tables, weakening credibility. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: This item has a clear local substantive defect. Specific manifestation: The whole is acceptable but local components have inconsistent styles, appearing uncoordinated. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "0": "Meets basic requirements: Diagram and text styles are ordinary but unified, with no substantive conflict; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: method diagrams, tables, and typographic tone are coordinated, and decoration is restrained. Quality improvement must be reflected as: specific choices in graphic abstraction, typography, or image processing improve audience understanding and overall impression. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: formulas, diagrams, and tables are integrated with a unified line, label, and component language, and visual abstraction serves the academic argument. Quality improvement must be reflected as: typography, graphics, images, or component language are coordinated to serve this task's purpose, without substituting good-looking colors for analysis. academic visual language is mature: abstraction levels, line weights, and the label system run through the whole deck, and diagrams and tables look like research schematics drawn by the same person. Even the hardest-to-integrate content coordinates with the overall visual language; when information is complete, there is still a clear and restrained scenario expression. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Non-color graphic abstraction, typographic tone, and method-diagram and table components should form a unified academic visual language that serves paper explanation rather than decorative piling. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the hardest-to-integrate content coordinates with the overall visual language; when information is complete, there is still a clear and restrained scenario expression. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A distinctive but appropriate visual language makes the subject or purpose intrinsic to the graphic form, beyond flawless stylistic integration. Identify a non-color design contribution and its communication benefit; personal taste, modernity and impressive decoration are not evidence. Apply the exceptional benefit specifically to: Non-color graphic abstraction, typographic tone, and method-diagram and table components should form a unified academic visual language that serves paper explanation rather than decorative piling. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; even the hardest-to-integrate content coordinates with the overall visual language; when information is complete, there is still a clear and restrained scenario expression. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 7,
    "benchmark_source": "vas"
  },
  {
    "id": "V6",
    "dimension": "Visuals",
    "criterion": "Endpoints, line widths, cropping, and optical alignment.\nObject of measurement: Arrow endpoints, line connectors, table borders, caption cropping, and optical alignment of elements should be clean; local geometric flaws should not interfere with decoding formulas and diagrams.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nObservation basis: Zoom in on the method diagram and check whether arrows touch the target box and point in the right direction. Check whether table line widths are uniform and whether there are redundant borders. Check whether captions or local screenshot edges are cropped cleanly.\nScoring boundary: Graphic outlines, line widths, endpoints, image proportions and cropping, and geometry and optical alignment within components; evaluate local construction precision.\nOriginal full refinement inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the finest connections, complex intersections, or irregular outlines are still handled sufficiently, and local craft does not damage the overall effect.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: The core understanding or use goal of this item cannot be achieved. Specific manifestation: Multiple broken connectors, wrongly pointing arrows, or clipped text directly hinder reading the diagram. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-2": "Clearly unacceptable: The main area or key process of this item is continuously impaired. Specific manifestation: Connectors/arrows in key method diagrams are obviously misaligned, or table borders are disordered. Judge by the actual impact above; harmless minor differences do not trigger this level.",
      "-1": "Below basic requirements: One endpoint, boundary, crop, proportion, or alignment has a clearly visible craft defect that substantially reduces local construction quality. Provide actual-impact evidence in this task's relevant parts of endpoints, line widths, cropping, and optical alignment; ordinary performance, harmless minor differences, or pure personal preference do not trigger a negative level.",
      "0": "Meets basic requirements: Graphic endpoints and table lines are neat; the necessary requirements of this item are fully met, with no substantive defects and ordinary performance.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: arrows, connectors, and alignment are clean, and cropping is standardized. Quality improvement must be reflected as: the handling of one class of endpoints, line widths, cropping, or optical alignment is more precise than ordinary neatness. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: local geometry is refined, line-width levels are consistent, and optical alignment stabilizes the layout. Quality improvement must be reflected as: the geometry, boundaries, and image proportions of different components coordinate with one another and are stable both up close and in overall viewing. craft is mature: endpoints, miters, cropping, and visual alignment withstand magnification checks, without relying on decoration to conceal. The finest connections, complex intersections, or irregular outlines are still handled sufficiently, and local craft does not damage the overall effect. Verify in this item's relevant complex parts: method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions. Task-specific object: Arrow endpoints, line connectors, table borders, caption cropping, and optical alignment of elements should be clean; local geometric flaws should not interfere with decoding formulas and diagrams. Difficult-case and inspection basis: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the finest connections, complex intersections, or irregular outlines are still handled sufficiently, and local craft does not damage the overall effect. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. Fine construction resolves a real perceptual ambiguity at the task’s most demanding local junction or component, beyond spotless alignment and uniform polish. Identify the optical or geometric benefit visible at the intended scale; more detail or tiny changes with no perceptible benefit do not qualify. Apply the exceptional benefit specifically to: Arrow endpoints, line connectors, table borders, caption cropping, and optical alignment of elements should be clean; local geometric flaws should not interfere with decoding formulas and diagrams. In-task verification cases for that benefit: Check only within this item's applicable scope the method-diagram matrix dimensions and Table 3's densest comparison, and the qualification of efficiency conclusions; the finest connections, complex intersections, or irregular outlines are still handled sufficiently, and local craft does not damage the overall effect. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 6,
    "benchmark_source": "vas"
  },
  {
    "id": "V7",
    "dimension": "Visuals",
    "criterion": "Cross-page visual identity and content consistency.\nObject of measurement: the eight pages' title system, footers, color coding and chart styles are consistent; the content of each page corresponds between PNG and PPTX, and the coding of the method figure keeps the same meaning on the relevant pages.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: compare title positions, footers and accent-color usage page by page. Check whether the per-page content of PNG and PPTX is consistent. Check whether the figure coding before/after merging keeps the same meaning on each page.\nScoring boundary: stable correspondence of the same object, identity, marking, style and content across pages, states and carriers; evaluate visual inheritance.\nOriginal full refinement inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; complex states or the densest content remain consistent across all required carriers; identity, layout and qualifications are all stable and traceable.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nCheck recurring objects in the actual slides and required exports; do not demand additional export formats.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: the core understanding or use objective of this item cannot be achieved. Specific manifestations: style breaks across multiple pages, or figures with the same name have contradictory coding on different pages, or PNG and PPTX content clearly does not correspond. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-2": "Clearly unacceptable: the main area or key process of this item is persistently damaged. Specific manifestations: the title/footer system is inconsistent, or the figure coding conflicts before and after merging. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-1": "Below basic requirements: this item has a clear local substantive defect. Specific manifestations: local page identity drift, such as inconsistent title position or accent-color usage. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "0": "Meets basic requirements: the pages look like the same slide deck, PNG and PPTX correspond page by page; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: titles, footers, colors and chart styles are stable across pages. Quality improvement must be shown as: repeated objects or corresponding carriers follow clear visual rules, reducing the burden of re-recognition. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: a unified identity system runs through the whole deck, and the same graphical coding keeps semantic consistency on the method, efficiency and limitations pages. Quality improvement must be shown as: state changes, repeated components and necessary exports correspond to each other; content changes are distinguished while identity does not drift. cross-page identity is mature: the visual system and terminology system are inherited in sync, and readers can transfer understanding seamlessly between pages. Complex states or the densest content remain consistent across all required carriers; identity, layout and qualifications are all stable and traceable. In the complex parts relevant to this item, verify: method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications. Task-specific object: the eight pages' title system, footers, color coding and chart styles are consistent; the content of each page corresponds between PNG and PPTX, and the coding of the method figure keeps the same meaning on the relevant pages. Difficult-case and inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; complex states or the densest content remain consistent across all required carriers; identity, layout and qualifications are all stable and traceable. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A deliberately designed recurring visible identity enables recognition across the task’s genuinely different views or presentations with materially less re-identification effort than fully refined consistency alone. Name the recognition benefit and verify it across actual delivered contexts, without requiring nonexistent extra versions. Apply the exceptional benefit specifically to: the eight pages' title system, footers, color coding and chart styles are consistent; the content of each page corresponds between PNG and PPTX, and the coding of the method figure keeps the same meaning on the relevant pages. In-task verification cases for that benefit: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; complex states or the densest content remain consistent across all required carriers; identity, layout and qualifications are all stable and traceable. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 3,
    "benchmark_source": "vas"
  },
  {
    "id": "V8",
    "dimension": "Visuals",
    "criterion": "Static reading path and speaking-notes key-point rhythm.\nObject of measurement: as static projection material, each page should have a natural reading order (title—body—annotation), and form a rhythm through page density and information sparsity/density; each page has an identifiable oral key point (visible on the page or in speaker notes, either one), but long text is not used to replace speaking. Do not evaluate semantic continuity between pages, and do not require animation or transitions.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: simulate reading each page in title—body—annotation order. Check whether each page has an identifiable oral key point rather than long paragraphs. Check whether density changes rhythmically between key pages and transition pages.\nScoring boundary: eye movement between information units and content reveal, dwell, motion, transition; evaluate scanning and temporal rhythm.\nOriginal full refinement inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the densest content or the most complex handoff can still be followed calmly, and the overall and detailed rhythm of the full reading or playback cycle is consistent.\nApplicable boundary for this task (applies to all relevant entries, adds no scoring items): no additional video, animation or transition deliverables are required.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nFor static slides, assess scanning and backtracking at presentation scale, not motion effects.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: the core understanding or use objective of this item cannot be achieved. Specific manifestations: the reading order is chaotic, or a page has no speakable key point at all and only stacked text remains. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-2": "Clearly unacceptable: the main area or key process of this item is persistently damaged. Specific manifestations: multiple pages lack reading guidance, and speaking-notes key points are disconnected from the visual subject. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-1": "Below basic requirements: this item has a clear local substantive defect. Specific manifestations: local reading paths jump, and individual pages only copy long passages of text. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "0": "Meets basic requirements: each page has a reading order and a tellable core point; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: the reading order is clear, and the arrangement of detail and brevity lets the speaker advance page by page. Quality improvement must be shown as: one type of scanning path, dwell or handoff arrangement makes the necessary information easier to finish viewing and follow. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: page rhythm varies, the visual subject and oral key points complement each other, and listeners can follow dwell and advancement. Quality improvement must be shown as: gaze or time arrangement is coordinated with content difficulty, with no unnecessary backtracking or waiting between entering, understanding and leaving. rhythm is mature: key pages and transition pages have well-proportioned density, and the speaker can finish the argument along visual cues without reading the screen. The densest content or the most complex handoff can still be followed calmly, and the overall and detailed rhythm of the full reading or playback cycle is consistent. In the complex parts relevant to this item, verify: method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications. Task-specific object: as static projection material, each page should have a natural reading order (title—body—annotation), and form a rhythm through page density and information sparsity/density; each page has an identifiable oral key point (visible on the page or in speaker notes, either one), but long text is not used to replace speaking. Do not evaluate semantic continuity between pages, and do not require animation or transitions. Difficult-case and inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the densest content or the most complex handoff can still be followed calmly, and the overall and detailed rhythm of the full reading or playback cycle is consistent. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. A non-obvious reading or temporal organization lets the audience orient, inspect and continue with unusually little backtracking or missed understanding. Identify the timing or scanning tension solved beyond consistently smooth ordinary presentation; add no motion or interaction to a static task. Apply the exceptional benefit specifically to: as static projection material, each page should have a natural reading order (title—body—annotation), and form a rhythm through page density and information sparsity/density; each page has an identifiable oral key point (visible on the page or in speaker notes, either one), but long text is not used to replace speaking. Do not evaluate semantic continuity between pages, and do not require animation or transitions. In-task verification cases for that benefit: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the densest content or the most complex handoff can still be followed calmly, and the overall and detailed rhythm of the full reading or playback cycle is consistent. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 3,
    "benchmark_source": "vas"
  },
  {
    "id": "V9",
    "dimension": "Visuals",
    "criterion": "Color-scheme aesthetics and image-ground integration.\nObject of measurement: the color scheme should be restrained and harmonious; judge from hue, lightness/darkness, saturation and area relationships whether it causes visual conflict or reduces readability; color scheme must not support a high score alone, nor impose a fixed aesthetic; whether the accent color carries semantics is only an optional observation.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: check whether the accent color stably corresponds to a certain type of semantics (such as computed values, trainable parts). Check whether chart ground colors harmonize with the page background. Check whether there are too many high-saturation color blocks competing for attention.\nScoring boundary: aesthetic relationships of hue, lightness/darkness, saturation × area, accent color and image/ground color; and proving quality objectives respectively with V3/V4/V5/V7.\nObservation elements: hue pairing and warm-cool relationships. Lightness hierarchy and color center of gravity. Saturation × area ratio. Number, weight and restraint of accent colors. Integration of images, graphics, panels and ground color.\nOriginal full refinement inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; in the hardest-to-harmonize images, dense areas or changing states, the relationships of lightness, hue and area remain unified, clear and restrained.\nScoring boundary: V9 independently evaluates the aesthetic harmony and refinement of the color scheme; reading impairment caused by text contrast belongs to V3, mismatch between color and object meaning belongs to V4, and cross-page color-code identity drift belongs to V7. Attribute across items only when different functional impairments are proven. V5 cannot merely repeat V9's color-scheme good/bad, and V9 cannot directly award the professional tier for \"fitting the scenario.\"\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.\nJudge negative visual grades by substantive harm and extent within this criterion, not a prerequisite of functional unusability. Harmless style differences are not defects; retain the original concrete defect examples.",
    "answers": {
      "-3": "Severe failure: the core understanding or use objective of this item cannot be achieved. Specific manifestations: the color scheme is cluttered or has high-saturation conflicts, and accent colors are used everywhere and lose semantics. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-2": "Clearly unacceptable: the main area or key process of this item is persistently damaged. Specific manifestations: color imbalance across multiple pages, and conflicts between figure and table ground colors, affecting overall credibility. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-1": "Below basic requirements: this item has a clear local substantive defect. Specific manifestations: the color scheme is basically unified but individual accents are excessive or low-contrast areas are inharmonious. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "0": "Meets basic requirements: the color scheme is ordinarily harmonious, and accent colors are discernible and not harsh; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: colors are restrained, accent colors correspond to semantics, and chart ground colors blend with the page. Quality improvement must be shown as: pointing out how specific hue, lightness or saturation and area relationships improve harmony, rather than merely listing color names. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the deck-wide color system is combined with information coding, and lightness and area allocation serve the hierarchy. Quality improvement must be shown as: the relationships among subject, background, images and accent colors are coordinated, and color center of gravity and area control serve the actual content. color-scheme aesthetics are mature: a limited set of hues and a lightness hierarchy run through the whole deck, accent colors precisely carry semantics and merge with image ground colors. In the hardest-to-harmonize images, dense areas or changing states, the relationships of lightness, hue and area remain unified, clear and restrained. In the complex parts relevant to this item, verify: method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications. Task-specific object: the color scheme should be restrained and harmonious; judge from hue, lightness/darkness, saturation and area relationships whether it causes visual conflict or reduces readability; color scheme must not support a high score alone, nor impose a fixed aesthetic; whether the accent color carries semantics is only an optional observation. Difficult-case and inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; in the hardest-to-harmonize images, dense areas or changing states, the relationships of lightness, hue and area remain unified, clear and restrained. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. Subtle hue, lightness, saturation and area relationships reconcile the hardest color-balance conflict with a distinctive, highly resolved visual character. Explain the perceptual gain beyond a flawlessly harmonious palette; semantic color correctness, fashion and personal liking do not establish exceptional color design. Apply the exceptional benefit specifically to: the color scheme should be restrained and harmonious; judge from hue, lightness/darkness, saturation and area relationships whether it causes visual conflict or reduces readability; color scheme must not support a high score alone, nor impose a fixed aesthetic; whether the accent color carries semantics is only an optional observation. In-task verification cases for that benefit: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; in the hardest-to-harmonize images, dense areas or changing states, the relationships of lightness, hue and area remain unified, clear and restrained. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 6,
    "benchmark_source": "vas"
  },
  {
    "id": "D1",
    "dimension": "Delivery",
    "criterion": "Delivered files can be opened and located directly.\nObject of measurement: artifact/slides.pptx is an openable 8-page 16:9 presentation; 01.png–08.png are corresponding 1280×720 images; file naming matches page order, and they can be used directly for reporting or located by reviewers; only evaluate these files actually required by the task statement.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: open slides.pptx and check the page count and 16:9 ratio. Verify that 01.png–08.png exist and are 1280×720. Compare the correspondence between each page's PNG and PPTX.\nScoring boundary: opening, locating and direct use of the required files; V7 evaluates the consistency of actually visible content, and does not substitute file existence for quality.\nOriginal full refinement inspection basis: within the task's true delivery scope, check the complete viewing flow from locating the main product to the necessary auxiliary materials; a missing item is judged only by actual impact, and no out-of-scope formats or extra files are required.\nScenario reference · medium (adds no threshold): 8-page English PPTX (16:9, 1280×720) plus same-named PNGs and sources.md; projected display, cropped parts of original charts usable, method figure as editable vectors.\nApplicable boundary for this task (applies to all relevant entries, adds no scoring items): no cover page, table-of-contents page or conclusion page beyond the 8 pages is required.\nApplicable boundary for this task (applies to all relevant entries, adds no scoring items): input files are read-only, and deliverables are placed in artifact/.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: A core required deliverable is missing or, because of its own defects, cannot be opened/used as a whole, making the task's core use objective impossible to achieve; first rule out evaluation-tool and environment failures.",
      "-2": "Clearly unacceptable: Required content is missing on a large scale, or key opening/locating processes are persistently obstructed, requiring substantial remediation of the main use workflow; judge according to actual impact, not by mechanically assigning a tier based solely on a single discrepancy in file count.",
      "-1": "Below basic requirements: There is a clear, localized issue involving a missing deliverable, version identification, file correspondence, or opening instructions that affects necessary use under this criterion, while the remaining core content is usable.",
      "0": "Meets basic requirements: the 8-page PPTX and 8 PNGs are complete, and dimensions and naming meet requirements; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: within the task-specified files and entry points, explicit role, sequence, or location cues eliminate one actual search or trial-and-error step; complete files and filenames that comply with the task statement alone qualify only for tier 0. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the roles, sequence, and necessary usage instructions of required files are coordinated, and the viewing path from the main artifact to corresponding previews/supporting materials is streamlined and unambiguous; assess only the files actually required by the task statement. even the required files, versions, or states most likely to be confused can be located accurately, and the complete viewing workflow is smooth; actual opening and required correspondence verification support the judgment, without requiring additional navigation pages or explanatory documents. Task-specific object: artifact/slides.pptx is an openable 8-page 16:9 presentation; 01.png–08.png are corresponding 1280×720 images; file naming matches page order, and they can be used directly for reporting or located by reviewers; only evaluate these files actually required by the task statement. Difficult-case and inspection basis: within the task's true delivery scope, check the complete viewing flow from locating the main product to the necessary auxiliary materials; a missing item is judged only by actual impact, and no out-of-scope formats or extra files are required. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. Within the required deliverables and entry points, an economical handoff removes a real selection, opening or first-use ambiguity that persists in a conventionally flawless delivery. Identify the materially reduced effort; extra files, packaging or software are not required. If no additional benefit is observable, +2 is the ceiling. Apply the exceptional benefit specifically to: artifact/slides.pptx is an openable 8-page 16:9 presentation; 01.png–08.png are corresponding 1280×720 images; file naming matches page order, and they can be used directly for reporting or located by reviewers; only evaluate these files actually required by the task statement. In-task verification cases for that benefit: within the task's true delivery scope, check the complete viewing flow from locating the main product to the necessary auxiliary materials; a missing item is judged only by actual impact, and no out-of-scope formats or extra files are required. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 4,
    "benchmark_source": "vas"
  },
  {
    "id": "D2",
    "dimension": "Delivery",
    "criterion": "Editable native tables and vector method figure.\nObject of measurement: the experiment table should be a PPTX native editable table, and the method figure should be an editable vector object; local paper figures are allowed, but a full-page screenshot must not replace a slide.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: select the table in PPTX and confirm whether it is a native table object. Check whether the method figure is editable shapes/lines rather than a bitmap. Check whether any full-page paper screenshot serves as a slide.\nScoring boundary: editable structure, resources and material organization of objects required by the task statement; no engineering scripts or additional tests are required.\nOriginal full refinement inspection basis: verify the structure, resources and editing entry points of the native experiment table and editable vector method figure in the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; do not modify submitted files, and do not insist on unrequested generation scripts or extra documents.\nApplicable boundary for this task (applies to all relevant entries, adds no scoring items): no source project for the PNGs or extra files beyond the PPTX, PNGs and sources.md listed in the task statement are required.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: the core understanding or use objective of this item cannot be achieved. Specific manifestations: the experiment table or method figure is replaced by a full-page screenshot bitmap and cannot be edited. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-2": "Clearly unacceptable: the main area or key process of this item is persistently damaged. Specific manifestations: the table is an image or most of the method figure is a bitmap, violating the editability requirement. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-1": "Below basic requirements: this item has a clear local substantive defect. Specific manifestations: the table is native but the method figure contains unseparated bitmap elements, or vice versa. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "0": "Meets basic requirements: the native experiment table and editable vector method figure are complete and editable as per the task statement, and necessary resources can be parsed; the boundary between native content and allowed raster assets is correct, and it does not depend on the author's private missing paths; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: in the native experiment table and editable vector method figure, object naming, grouping or data relationships have specific arrangements that make locating one type of necessary content/local maintenance more direct than ordinary editability; object existence alone is insufficient for a positive score. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: the semantic grouping, resources and editing entry points of the native experiment table and editable vector method figure are coordinated, complex objects can be located, and maintaining a local part does not disturb unrelated content; no additional general-purpose engine or refactoring is required. in the required objects corresponding to the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications, both complex structures and simple parts are clearly maintainable, and resource relationships and originally correct content are stable; based on native structure and task-specified construction checks, do not modify the submission or add out-of-scope experiments. Task-specific object: the experiment table should be a PPTX native editable table, and the method figure should be an editable vector object; local paper figures are allowed, but a full-page screenshot must not replace a slide. Difficult-case and inspection basis: verify the structure, resources and editing entry points of the native experiment table and editable vector method figure in the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; do not modify submitted files, and do not insist on unrequested generation scripts or extra documents. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. The native editable organization reveals a meaningful relationship among task objects so directly that a targeted maintenance step requires materially less structural reconstruction than in a flawlessly editable conventional artifact. Show this through native evidence, without requiring an extra refactoring, unrequested source code or speculative editing claims. Apply the exceptional benefit specifically to: the experiment table should be a PPTX native editable table, and the method figure should be an editable vector object; local paper figures are allowed, but a full-page screenshot must not replace a slide. In-task verification cases for that benefit: verify the structure, resources and editing entry points of the native experiment table and editable vector method figure in the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; do not modify submitted files, and do not insist on unrequested generation scripts or extra documents. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 3,
    "benchmark_source": "vas"
  },
  {
    "id": "D3",
    "dimension": "Delivery",
    "criterion": "Verifiability of provenance and computation notes.\nObject of measurement: sources.md should give the page numbers/table numbers/version used and the computation locations, sufficient to verify Table 3 data, derived multiples and §4.2 efficiency facts.\nObservation scope: 8-page 16:9 English PPTX and 1280×720 PNG, for a graduate student group meeting familiar with deep learning.\nBasis for observation: check whether arXiv 2106.09685v2 and the 26-page version are indicated. Verify Table 3's page number/table number and §4.2 page number. Check whether the derived multiple formula and computation location are stated.\nScoring boundary: verifiability of required sources, computation/modification/capture or usage notes; select according to this task's actual requirements, and do not insist on all document types.\nOriginal full refinement inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the most easily confused sources, conditions or modifications can also be verified along the notes, and the complete verification path is concise with no gaps or overclaims.\nRevised grade boundary: the original high-grade descriptions are +2 quality obligations, not sufficient evidence for +3. +2 covers the entire applicable scope; +3 additionally requires the exceptional benefit specified in answers[\"3\"]. Levels inherit quality obligations, not the shortcomings explaining lower grades.",
    "answers": {
      "-3": "Severe failure: the core understanding or use objective of this item cannot be achieved. Specific manifestations: no provenance file, or the content cannot locate any source. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-2": "Clearly unacceptable: the main area or key process of this item is persistently damaged. Specific manifestations: only a generic citation of the paper title, lacking page numbers/table numbers/version, and key computations have no location. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "-1": "Below basic requirements: this item has a clear local substantive defect. Specific manifestations: page numbers and table numbers are present but the version or computation notes are missing in one place, requiring guessing to verify. Judge by the above actual impact; harmless minor differences do not trigger this tier.",
      "0": "Meets basic requirements: the page numbers/table numbers used, the v2 version and the computation notes accurately correspond to the actual claims and can be located and verified; this item's necessary requirements are fully met, with no substantive defects; performance is ordinary.",
      "1": "Accomplished but not yet fully refined: meets tier 0 and shows a substantive observed benefit beyond ordinary correctness, completeness or readability through the following: provenance is specific to pages and tables, and computation step locations are clear. Quality improvement must be shown as: at least one type of key basis has a precise location and corresponds to the actual object, reducing the need to search generic sources. In reason explain the concrete benefit and the observable limitation keeping it below +2. Do not invent a defect to force this grade; missing required observation must be acquired first.",
      "2": "Fully refined within scope: meets tier 0 and achieves complete quality in every applicable part, not only an attractive local example. Retain and jointly verify these quality obligations: every fact and computation can be located and verified via sources.md, and the version and table numbering are clear. Quality improvement must be shown as: claims, materials and the notes required by the task statement correspond to each other, so one can both trace the basis and know where that basis is used. provenance is mature and complete: the correspondence among version, page numbers, table numbers, computation locations and the paper's reported values/computed values is clear at a glance. The most easily confused sources, conditions or modifications can also be verified along the notes, and the complete verification path is concise with no gaps or overclaims. In the complex parts relevant to this item, verify: method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications. Task-specific object: sources.md should give the page numbers/table numbers/version used and the computation locations, sufficient to verify Table 3 data, derived multiples and §4.2 efficiency facts. Difficult-case and inspection basis: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the most easily confused sources, conditions or modifications can also be verified along the notes, and the complete verification path is concise with no gaps or overclaims. Within this criterion and the stated use conditions, there is no identifiable substantive weakness or feasible task-relevant improvement of material benefit. Establish full applicable coverage and actively inspect the strongest plausible shortcoming; finding no error, a single successful example or an author assertion does not prove full refinement.",
      "3": "Exceptional beyond conventional full refinement: first fully meet +2. An economical provenance/handoff organization makes a difficult fact, calculation or asset relationship directly auditable, reducing reconstruction beyond a complete accurate conventional source list. Identify the actual trace and saved ambiguity, without requiring new evidence or new documentation beyond the task. Apply the exceptional benefit specifically to: sources.md should give the page numbers/table numbers/version used and the computation locations, sufficient to verify Table 3 data, derived multiples and §4.2 efficiency facts. In-task verification cases for that benefit: within this item's applicable scope only, check the method figure matrix dimensions and Table 3's densest comparison, efficiency conclusion qualifications; the most easily confused sources, conditions or modifications can also be verified along the notes, and the complete verification path is concise with no gaps or overclaims. Correct completion of these cases is only a foundation, not evidence of exceptional benefit. In reason identify the non-obvious choice, real difficulty and substantial benefit beyond an equally correct, complete and polished conventional solution, with actual evidence. No comparison artifact or unique design is required. Complexity, extra features, ornament, praise or repetition of +2 strengths is insufficient. Without demonstrated extra benefit the ceiling is +2; no score quota applies."
    },
    "weight": 3,
    "benchmark_source": "vas"
  }
]