1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052#!/usr/bin/env node
// Validate the functional image corpus and report progress against the
// acceptance criteria in directives/00-corpus-goals.md.
//
// Usage, from anywhere:
// node tools/validate.mjs validate corpus/functional-images.jsonl
// node tools/validate.mjs --json machine readable report on stdout
// node tools/validate.mjs --corpus FILE validate a different corpus file
// node tools/validate.mjs --rounds DIR read review reports from DIR
// node tools/validate.mjs --target 100 goal of 100 accepted items
// node tools/validate.mjs --selftest run the fixture self-test
//
// The target is the number of accepted items the run is working toward. It
// comes from --target, else from corpus/target.txt, else the 250-item v0.1
// corpus in directive 00. Count targets scale with it; shares do not.
//
// Exit codes:
// 0 corpus is schema clean and every acceptance criterion is met
// 1 corpus is schema clean but not yet complete
// 2 schema errors, or the corpus file is missing
// 3 bad usage, or the self-test failed
//
// No dependencies. Plain text in, plain text out.
import { readFileSync, writeFileSync, existsSync, readdirSync, mkdtempSync,
rmSync } from 'node:fs'
import { tmpdir } from 'node:os'
import { dirname, join, resolve } from 'node:path'
import { fileURLToPath } from 'node:url'
const HERE = dirname(fileURLToPath(import.meta.url))
const PROJECT = resolve(HERE, '..')
// ---------------------------------------------------------------------------
// Schema, mirroring corpus/README.md. Keep the two in step.
// ---------------------------------------------------------------------------
const SUBTYPES = {
'linked-standalone-logo': 1,
'standalone-navigational-link': 1,
'form-control-or-image-button': 2,
'action-or-toggle-icon': 2,
'functional-non-unicode-emoji': 3,
'linked-complex-graphic-or-image-map': 4,
'structural-break-or-reader-control': 5,
}
const STATUSES = ['candidate', 'accepted', 'needs-revision', 'rejected']
const IMPLEMENTATIONS = ['img', 'inline-svg', 'icon-font', 'sprite',
'css-background', 'input-image', 'area']
const IMPLEMENTATIONS_NEEDING_URL = ['img', 'sprite', 'css-background',
'input-image', 'area']
const ROLES = ['link', 'button', 'input-image', 'area', 'custom', 'glyph']
const OBSERVED_VERDICTS = ['correct', 'wrong', 'missing', 'empty-appropriate',
'empty-inappropriate']
const DIFFICULTIES = ['trivial', 'standard', 'ambiguous']
const REQUIRED = ['id', 'status', 'round_added', 'category', 'subtype',
'page_url', 'domain', 'image_url', 'implementation', 'element_role',
'element_html', 'surrounding_text', 'destination', 'observed_alt',
'observed_alt_verdict', 'gold_alt', 'gold_alt_rationale', 'gold_alt_passes',
'adjudication', 'difficulty', 'dual_purpose', 'leakage_check', 'leaky',
'retrieved', 'provenance_note']
const OPTIONAL = ['notes']
const REASON_CODES = ['CLEAN', 'UNVERIFIABLE-SOURCE', 'CONTEXT-INACCURATE',
'NOT-FUNCTIONAL', 'MISCLASSIFIED', 'APPEARANCE-DESCRIPTION',
'REDUNDANCY-MISSED', 'WRONGLY-EMPTY', 'REDUNDANT-STARTER', 'TOO-VERBOSE',
'ASSUMPTION', 'NO-RATIONALE', 'LEAKAGE', 'NON-DISCRIMINATING', 'DUPLICATE',
'MISSING-FIELD', 'NO-SECOND-PASS', 'LICENSE-UNCLEAR']
const REVIEW_REQUIRED = ['item_id', 'round', 'verdict', 'reason_codes',
'evidence', 'required_change', 'blocking']
const VERDICTS = ['accept', 'revise', 'reject']
// Targets from directives/00-corpus-goals.md. The full v0.1 corpus is 250
// accepted items, and the count targets below are stated for that size.
const BASELINE = {
total: 250,
perCategory: 30,
perSubtype: 20,
domainsPerSubtype: 8,
}
// A run may stipulate a smaller corpus. Count targets scale with it, so a
// 100-item goal keeps the same shape as a 250-item one instead of turning into
// 100 items of whatever was easiest to find. Shares are ratios and do not
// scale. The floors keep a small corpus from degenerating: a sub-type
// represented by one item, or drawn from one domain, measures that item rather
// than the sub-type.
function targetsFor(total) {
const k = total / BASELINE.total
const at = (base, floor) => Math.max(floor, Math.round(base * k))
return {
total,
perCategory: at(BASELINE.perCategory, 4),
perSubtype: at(BASELINE.perSubtype, 3),
domainsPerSubtype: at(BASELINE.domainsPerSubtype, 3),
emptyAltShare: 0.15,
dualPurposeShare: 0.10,
ambiguousShare: 0.20,
maxDomainShare: 0.05,
cleanItemShare: 0.95,
quietRounds: 2,
}
}
// Reassigned by main() once the run's target is known.
let TARGETS = targetsFor(BASELINE.total)
// Below this, the floors above collide: seven sub-types at three items each
// already needs 21, and a corpus with no slack over its own minimums cannot
// satisfy the share targets as well.
const MIN_TARGET = 25
// The stipulated goal, one integer on the first line that is not a comment.
// Kept in the corpus directory and committed, so every tool and every round
// agrees on what the run is working toward, and a change to the goal shows up
// in the history rather than in someone's shell.
function readTargetFile(path) {
if (!existsSync(path)) return { value: null }
for (const raw of readFileSync(path, 'utf8').split('\n')) {
const line = raw.trim()
if (line === '' || line.startsWith('#')) continue
if (!/^\d+$/.test(line)) {
return { error: `${path}: expected one whole number, found "${line}"` }
}
return { value: Number(line) }
}
return { error: `${path}: no target found in the file` }
}
const REDUNDANT_STARTERS = ['link to', 'links to', 'button for', 'button to',
'icon of', 'icon for', 'image of', 'graphic of', 'picture of', 'go to link']
const MAX_ALT_CHARS = 125
const MIN_RATIONALE_CHARS = 40
const MIN_LEAKAGE_CHARS = 20
const MIN_EVIDENCE_CHARS = 20
// ---------------------------------------------------------------------------
// Helpers
// ---------------------------------------------------------------------------
const isStr = (v) => typeof v === 'string'
const isBool = (v) => typeof v === 'boolean'
const isInt = (v) => Number.isInteger(v)
const isObj = (v) => v !== null && typeof v === 'object' && !Array.isArray(v)
// Where an item stands on the two-independent-passes rule in directive 00.
//
// one-pass only the first pass exists, so directives/02-second-pass.md
// has not run over it yet
// disagreeing a second pass exists and differs, which is the interesting
// case and needs an adjudication before acceptance
// settled two agreeing passes, or a recorded adjudication
//
// Only `settled` items can be accepted. The other two are ordinary states for a
// candidate, and counting them is how the loop knows what it still owes.
function passState(rec) {
if (isStr(rec.adjudication) && rec.adjudication.trim() !== '') return 'settled'
const passes = Array.isArray(rec.gold_alt_passes)
? rec.gold_alt_passes.filter(isObj) : []
if (passes.length < 2) return 'one-pass'
return new Set(passes.map((p) => p.alt).filter(isStr)).size > 1
? 'disagreeing' : 'settled'
}
function isHttpUrl(v) {
if (!isStr(v)) return false
try {
const u = new URL(v)
return u.protocol === 'http:' || u.protocol === 'https:'
} catch {
return false
}
}
function hostOf(url) {
try {
return new URL(url).host.toLowerCase().replace(/^www\./, '')
} catch {
return null
}
}
function readJsonl(path) {
const errors = []
const rows = []
let text
try {
text = readFileSync(path, 'utf8')
} catch (e) {
errors.push(`${path}: cannot be read, ${e.code ?? e.message}`)
return { rows, errors }
}
text.split('\n').forEach((raw, i) => {
const line = raw.trim()
if (line === '') return
try {
const value = JSON.parse(line)
if (!isObj(value)) {
errors.push(`${path}:${i + 1}: line is not a JSON object`)
return
}
rows.push({ line: i + 1, value })
} catch (e) {
errors.push(`${path}:${i + 1}: invalid JSON, ${e.message}`)
}
})
return { rows, errors }
}
const pct = (n, d) => (d === 0 ? 0 : n / d)
const fmtPct = (x) => `${(x * 100).toFixed(1)}%`
// ---------------------------------------------------------------------------
// Item validation
// ---------------------------------------------------------------------------
function validateItem(rec, line, seenIds, path) {
const errs = []
const bad = (msg) => errs.push(`${path}:${line}: ${msg}`)
const id = isStr(rec.id) ? rec.id : `line ${line}`
for (const field of REQUIRED) {
if (!(field in rec)) bad(`${id}: missing required field \`${field}\``)
}
for (const key of Object.keys(rec)) {
if (!REQUIRED.includes(key) && !OPTIONAL.includes(key)) {
bad(`${id}: unknown field \`${key}\``)
}
}
if (!isStr(rec.id) || !/^fi-\d{4}$/.test(rec.id)) {
bad(`${id}: \`id\` must match fi-0000`)
} else if (seenIds.has(rec.id)) {
bad(`${id}: duplicate \`id\`, first seen on line ${seenIds.get(rec.id)}`)
} else {
seenIds.set(rec.id, line)
}
if (!STATUSES.includes(rec.status)) {
bad(`${id}: \`status\` must be one of ${STATUSES.join(', ')}`)
}
if (!isInt(rec.round_added) || rec.round_added < 1) {
bad(`${id}: \`round_added\` must be an integer of 1 or more`)
}
if (!isInt(rec.category) || rec.category < 1 || rec.category > 5) {
bad(`${id}: \`category\` must be an integer 1 through 5`)
}
if (!(rec.subtype in SUBTYPES)) {
bad(`${id}: \`subtype\` must be one of ${Object.keys(SUBTYPES).join(', ')}`)
} else if (SUBTYPES[rec.subtype] !== rec.category) {
bad(`${id}: \`subtype\` ${rec.subtype} belongs to category ` +
`${SUBTYPES[rec.subtype]}, not ${rec.category}`)
}
if (!isHttpUrl(rec.page_url)) bad(`${id}: \`page_url\` must be an http URL`)
const host = hostOf(rec.page_url)
if (!isStr(rec.domain) || rec.domain === '') {
bad(`${id}: \`domain\` must be a non-empty string`)
} else if (host && rec.domain !== host) {
bad(`${id}: \`domain\` is "${rec.domain}" but \`page_url\` host is "${host}"`)
}
if (!IMPLEMENTATIONS.includes(rec.implementation)) {
bad(`${id}: \`implementation\` must be one of ${IMPLEMENTATIONS.join(', ')}`)
}
if (IMPLEMENTATIONS_NEEDING_URL.includes(rec.implementation)) {
if (!isHttpUrl(rec.image_url)) {
bad(`${id}: \`image_url\` must be an http URL for ` +
`${rec.implementation} implementations`)
}
} else if (rec.image_url !== null && !isHttpUrl(rec.image_url)) {
bad(`${id}: \`image_url\` must be an http URL or null`)
}
if (!ROLES.includes(rec.element_role)) {
bad(`${id}: \`element_role\` must be one of ${ROLES.join(', ')}`)
}
if (!isStr(rec.element_html) || rec.element_html.trim() === '') {
bad(`${id}: \`element_html\` must be non-empty verbatim markup`)
}
if (!isStr(rec.surrounding_text)) {
bad(`${id}: \`surrounding_text\` must be a string, empty if none`)
}
if (!isStr(rec.destination) || rec.destination.trim() === '') {
bad(`${id}: \`destination\` must be non-empty`)
}
if (rec.observed_alt !== null && !isStr(rec.observed_alt)) {
bad(`${id}: \`observed_alt\` must be a string or null`)
}
if (!OBSERVED_VERDICTS.includes(rec.observed_alt_verdict)) {
bad(`${id}: \`observed_alt_verdict\` must be one of ` +
OBSERVED_VERDICTS.join(', '))
}
if (rec.observed_alt === null && rec.observed_alt_verdict !== 'missing') {
bad(`${id}: \`observed_alt\` is null so \`observed_alt_verdict\` must be ` +
`"missing"`)
}
if (!isStr(rec.gold_alt)) {
bad(`${id}: \`gold_alt\` must be a string, empty for alt=""`)
} else {
if (rec.gold_alt.length > MAX_ALT_CHARS) {
bad(`${id}: \`gold_alt\` is ${rec.gold_alt.length} characters, over the ` +
`${MAX_ALT_CHARS} guidance`)
}
const lower = rec.gold_alt.trim().toLowerCase()
for (const starter of REDUNDANT_STARTERS) {
if (lower.startsWith(starter)) {
bad(`${id}: \`gold_alt\` starts with the redundant starter "${starter}"`)
break
}
}
}
if (!isStr(rec.gold_alt_rationale) ||
rec.gold_alt_rationale.trim().length < MIN_RATIONALE_CHARS) {
bad(`${id}: \`gold_alt_rationale\` must be at least ` +
`${MIN_RATIONALE_CHARS} characters`)
}
if (!Array.isArray(rec.gold_alt_passes) || rec.gold_alt_passes.length < 1) {
bad(`${id}: \`gold_alt_passes\` must be an array of one or more passes`)
} else {
rec.gold_alt_passes.forEach((p, i) => {
if (!isObj(p)) {
bad(`${id}: \`gold_alt_passes[${i}]\` must be an object`)
return
}
for (const f of ['author', 'alt', 'rationale']) {
if (!isStr(p[f])) bad(`${id}: \`gold_alt_passes[${i}].${f}\` must be a string`)
}
for (const key of Object.keys(p)) {
if (!['author', 'alt', 'rationale'].includes(key)) {
bad(`${id}: \`gold_alt_passes[${i}]\` has unknown field \`${key}\``)
}
}
})
const alts = new Set(rec.gold_alt_passes
.filter(isObj).map((p) => p.alt).filter(isStr))
const settled = isStr(rec.adjudication) && rec.adjudication.trim() !== ''
// A second pass that disagrees with the first is a normal state for a
// candidate, and the whole point of authoring it blind. The disagreement is
// a finding, not a defect, and the next round resolves it. Only acceptance
// requires it settled, which is checked with the other accepted-item rules
// below and refused by tools/apply-verdicts.mjs.
if (alts.size > 1 && !settled && rec.status === 'accepted') {
bad(`${id}: passes disagree, so \`adjudication\` is required`)
}
if (isStr(rec.gold_alt) && alts.size > 0 && !alts.has(rec.gold_alt) &&
!settled) {
bad(`${id}: \`gold_alt\` matches no pass and there is no adjudication`)
}
}
if (rec.adjudication !== null && !isStr(rec.adjudication)) {
bad(`${id}: \`adjudication\` must be a string or null`)
}
if (!DIFFICULTIES.includes(rec.difficulty)) {
bad(`${id}: \`difficulty\` must be one of ${DIFFICULTIES.join(', ')}`)
}
if (!isBool(rec.dual_purpose)) bad(`${id}: \`dual_purpose\` must be a boolean`)
if (!isBool(rec.leaky)) bad(`${id}: \`leaky\` must be a boolean`)
if (!isStr(rec.leakage_check) ||
rec.leakage_check.trim().length < MIN_LEAKAGE_CHARS) {
bad(`${id}: \`leakage_check\` must be at least ${MIN_LEAKAGE_CHARS} characters`)
}
if (!isStr(rec.retrieved) || !/^\d{4}-\d{2}-\d{2}$/.test(rec.retrieved) ||
Number.isNaN(Date.parse(rec.retrieved))) {
bad(`${id}: \`retrieved\` must be a real date as YYYY-MM-DD`)
}
if (!isStr(rec.provenance_note) || rec.provenance_note.trim() === '') {
bad(`${id}: \`provenance_note\` must be non-empty`)
}
if ('notes' in rec && !isStr(rec.notes)) {
bad(`${id}: \`notes\` must be a string when present`)
}
// Rules that only bind accepted items.
if (rec.status === 'accepted') {
if (rec.leaky === true) {
bad(`${id}: a leaky item cannot be \`accepted\``)
}
if (Array.isArray(rec.gold_alt_passes) && rec.gold_alt_passes.length < 2 &&
!(isStr(rec.adjudication) && rec.adjudication.trim() !== '')) {
bad(`${id}: \`accepted\` needs two independent passes, or an adjudication`)
}
if (Array.isArray(rec.gold_alt_passes) &&
new Set(rec.gold_alt_passes.filter(isObj).map((p) => p.alt)
.filter(isStr)).size > 1 &&
!(isStr(rec.adjudication) && rec.adjudication.trim() !== '')) {
bad(`${id}: \`accepted\` with disagreeing passes needs an \`adjudication\``)
}
}
return errs
}
function validateReview(rec, line, path) {
const errs = []
const bad = (msg) => errs.push(`${path}:${line}: ${msg}`)
const id = isStr(rec.item_id) ? rec.item_id : `line ${line}`
for (const field of REVIEW_REQUIRED) {
if (!(field in rec)) bad(`${id}: missing required field \`${field}\``)
}
for (const key of Object.keys(rec)) {
if (!REVIEW_REQUIRED.includes(key)) bad(`${id}: unknown field \`${key}\``)
}
if (!isStr(rec.item_id) || !/^fi-\d{4}$/.test(rec.item_id)) {
bad(`${id}: \`item_id\` must match fi-0000`)
}
if (!isInt(rec.round) || rec.round < 1) {
bad(`${id}: \`round\` must be an integer of 1 or more`)
}
if (!VERDICTS.includes(rec.verdict)) {
bad(`${id}: \`verdict\` must be one of ${VERDICTS.join(', ')}`)
}
if (!Array.isArray(rec.reason_codes) || rec.reason_codes.length < 1) {
bad(`${id}: \`reason_codes\` must list at least one code`)
} else {
for (const code of rec.reason_codes) {
if (!REASON_CODES.includes(code)) {
bad(`${id}: unknown reason code "${code}"`)
}
}
if (rec.verdict === 'accept' &&
!(rec.reason_codes.length === 1 && rec.reason_codes[0] === 'CLEAN')) {
bad(`${id}: an \`accept\` verdict takes exactly the code CLEAN`)
}
if (rec.verdict !== 'accept' && rec.reason_codes.includes('CLEAN')) {
bad(`${id}: CLEAN is only valid on an \`accept\` verdict`)
}
}
if (!isStr(rec.evidence) || rec.evidence.trim().length < MIN_EVIDENCE_CHARS) {
bad(`${id}: \`evidence\` must be at least ${MIN_EVIDENCE_CHARS} characters`)
}
if (rec.verdict === 'revise') {
if (!isStr(rec.required_change) || rec.required_change.trim() === '') {
bad(`${id}: a \`revise\` verdict needs \`required_change\``)
}
} else if (rec.required_change !== null) {
bad(`${id}: \`required_change\` must be null unless the verdict is revise`)
}
if (!isBool(rec.blocking)) bad(`${id}: \`blocking\` must be a boolean`)
// An accept means no blocking defect was found. The combination would promote
// the item and hold it in the 95 percent criterion at the same time, which no
// later round can resolve because accepted items are not re-reviewed.
else if (rec.verdict === 'accept' && rec.blocking === true) {
bad(`${id}: an \`accept\` verdict cannot also be \`blocking\``)
}
return errs
}
// ---------------------------------------------------------------------------
// Round reports: the two-quiet-rounds gate
// ---------------------------------------------------------------------------
const STATUS_LINE = /^STATUS:\s*new-blocking-findings=(yes|no)\s*$/m
// Round number out of a file name, so ordering is numeric. Sorting these names
// as strings puts round-10 before round-9, which would make the two-quiet-rounds
// gate read the wrong pair of reports.
const roundNumber = (name) => Number(name.match(/^round-(\d+)-/)[1])
function roundFiles(dir, pattern) {
let names
try {
names = readdirSync(dir)
} catch {
return []
}
return names
.filter((n) => pattern.test(n))
.sort((a, b) => roundNumber(a) - roundNumber(b))
}
function readRounds(roundsDir) {
const reports = []
const errors = []
if (!existsSync(roundsDir)) return { reports, errors }
const names = roundFiles(roundsDir, /^round-\d+-report\.md$/)
for (const name of names) {
let text
try {
text = readFileSync(join(roundsDir, name), 'utf8')
} catch (e) {
errors.push(`${join(roundsDir, name)}: cannot be read, ${e.code ?? e.message}`)
reports.push({ name, newBlocking: null })
continue
}
const m = text.match(STATUS_LINE)
if (!m) {
errors.push(`${join(roundsDir, name)}: no ` +
`"STATUS: new-blocking-findings=yes|no" line, required by directive 02`)
reports.push({ name, newBlocking: null })
} else {
reports.push({ name, newBlocking: m[1] === 'yes' })
}
}
return { reports, errors }
}
// ---------------------------------------------------------------------------
// Acceptance criteria
// ---------------------------------------------------------------------------
function assess(items, reviews, reports) {
const accepted = items.filter((r) => r.status === 'accepted')
const total = accepted.length
const criteria = []
const add = (name, met, detail) => criteria.push({ name, met, detail })
add('total accepted items', total >= TARGETS.total,
`${total} of ${TARGETS.total}`)
const catCounts = {}
for (let c = 1; c <= 5; c++) catCounts[c] = 0
const subCounts = {}
for (const s of Object.keys(SUBTYPES)) subCounts[s] = 0
const subDomains = {}
for (const s of Object.keys(SUBTYPES)) subDomains[s] = new Set()
const domainCounts = {}
for (const r of accepted) {
if (r.category in catCounts) catCounts[r.category]++
if (r.subtype in subCounts) {
subCounts[r.subtype]++
subDomains[r.subtype].add(r.domain)
}
domainCounts[r.domain] = (domainCounts[r.domain] || 0) + 1
}
const thinCats = Object.entries(catCounts)
.filter(([, n]) => n < TARGETS.perCategory)
add('per category minimum', thinCats.length === 0,
thinCats.length === 0
? `all five categories at ${TARGETS.perCategory} or more`
: thinCats.map(([c, n]) => `category ${c}: ${n}`).join(', '))
const thinSubs = Object.entries(subCounts)
.filter(([, n]) => n < TARGETS.perSubtype)
add('per sub-type minimum', thinSubs.length === 0,
thinSubs.length === 0
? `all seven sub-types at ${TARGETS.perSubtype} or more`
: thinSubs.map(([s, n]) => `${s}: ${n}`).join(', '))
const emptyAlt = accepted.filter((r) => r.gold_alt === '').length
add('empty-alt share', pct(emptyAlt, total) >= TARGETS.emptyAltShare,
`${emptyAlt} items, ${fmtPct(pct(emptyAlt, total))} of ` +
`${fmtPct(TARGETS.emptyAltShare)}`)
const dual = accepted.filter((r) => r.dual_purpose === true).length
add('dual-purpose share', pct(dual, total) >= TARGETS.dualPurposeShare,
`${dual} items, ${fmtPct(pct(dual, total))} of ` +
`${fmtPct(TARGETS.dualPurposeShare)}`)
const ambiguous = accepted.filter((r) => r.difficulty === 'ambiguous').length
add('ambiguous share', pct(ambiguous, total) >= TARGETS.ambiguousShare,
`${ambiguous} items, ${fmtPct(pct(ambiguous, total))} of ` +
`${fmtPct(TARGETS.ambiguousShare)}`)
const overDomains = Object.entries(domainCounts)
.filter(([, n]) => pct(n, total) > TARGETS.maxDomainShare)
.sort((a, b) => b[1] - a[1])
add('domain concentration', total > 0 && overDomains.length === 0,
total === 0
? 'no accepted items yet'
: overDomains.length === 0
? `no domain over ${fmtPct(TARGETS.maxDomainShare)}`
: overDomains.map(([d, n]) =>
`${d}: ${n}, ${fmtPct(pct(n, total))}`).join(', '))
const thinSpread = Object.entries(subDomains)
.filter(([, set]) => set.size < TARGETS.domainsPerSubtype)
add('domains per sub-type', thinSpread.length === 0,
thinSpread.length === 0
? `every sub-type spans ${TARGETS.domainsPerSubtype} or more domains`
: thinSpread.map(([s, set]) => `${s}: ${set.size}`).join(', '))
const noSecond = accepted.filter((r) => passState(r) !== 'settled').length
add('two independent passes', noSecond === 0,
noSecond === 0 ? 'every accepted item has two agreeing passes or an adjudication'
: `${noSecond} accepted items with a single pass, or disagreeing passes ` +
'and no adjudication')
// Blocking findings still open against accepted items. An item's most recent
// review is the one that counts: a blocking finding stays open until a later
// round says otherwise, so it cannot age out of the gate just by the corpus
// moving on to another round.
const latestRound = reviews.reduce((max, r) =>
isInt(r.round) && r.round > max ? r.round : max, 0)
const latestForItem = new Map()
for (const r of reviews) {
if (!isStr(r.item_id) || !isInt(r.round)) continue
const held = latestForItem.get(r.item_id)
if (held === undefined || r.round >= held.round) latestForItem.set(r.item_id, r)
}
const blockedIds = new Set(accepted
.filter((item) => latestForItem.get(item.id)?.blocking === true)
.map((item) => item.id))
const clean = total - blockedIds.size
add('items clean in latest review',
total > 0 && latestRound >= 1 && pct(clean, total) >= TARGETS.cleanItemShare,
total === 0 ? 'no accepted items yet'
: latestRound === 0
? `${total} accepted items but no review round has run`
: `${clean} of ${total} clean in round ${latestRound}, ` +
`${fmtPct(pct(clean, total))} of ${fmtPct(TARGETS.cleanItemShare)}`)
const tail = reports.slice(-TARGETS.quietRounds)
const quiet = reports.length >= TARGETS.quietRounds &&
tail.every((r) => r.newBlocking === false)
add('two consecutive quiet review rounds', quiet,
reports.length === 0 ? 'no review reports yet'
: tail.map((r) => `${r.name}: new-blocking-findings=` +
(r.newBlocking === null ? 'unstated' : r.newBlocking ? 'yes' : 'no'))
.join(', '))
const byStatus = {}
for (const s of STATUSES) byStatus[s] = items.filter((r) => r.status === s).length
// Not criteria: work owed on items that are still in play. An item with one
// pass is waiting for the second-pass turn, and one with disagreeing passes is
// waiting for a seeking agent to adjudicate. Neither blocks the loop, and both
// block the items themselves, so the loop is told about them every round.
const inPlay = items.filter((r) =>
r.status === 'candidate' || r.status === 'needs-revision')
const pending = {
secondPass: inPlay.filter((r) => passState(r) === 'one-pass').length,
adjudication: inPlay.filter((r) => passState(r) === 'disagreeing').length,
}
return {
criteria,
goalsMet: criteria.every((c) => c.met),
pending,
counts: {
items: items.length,
byStatus,
byCategory: catCounts,
bySubtype: subCounts,
domains: Object.keys(domainCounts).length,
reviewRecords: reviews.length,
reviewRounds: reports.length,
latestRound,
},
}
}
// ---------------------------------------------------------------------------
// Reporting
// ---------------------------------------------------------------------------
function printText(report) {
const { corpusPath, schemaErrors, assessment } = report
const out = []
out.push(`corpus: ${corpusPath}`)
out.push(`goal: ${TARGETS.total} accepted items` +
(TARGETS.total === BASELINE.total
? ' (the full v0.1 corpus)'
: `, scaled from ${BASELINE.total}: ${TARGETS.perCategory} per category, ` +
`${TARGETS.perSubtype} per sub-type, ` +
`${TARGETS.domainsPerSubtype} domains per sub-type`))
if (schemaErrors.length > 0) {
out.push('')
out.push(`schema errors: ${schemaErrors.length}`)
for (const e of schemaErrors) out.push(` ${e}`)
} else {
out.push('schema: clean')
}
if (assessment) {
const c = assessment.counts
out.push('')
out.push(`items: ${c.items} total, ` + STATUSES
.map((s) => `${c.byStatus[s]} ${s}`).join(', '))
out.push(`reviews: ${c.reviewRecords} records across ` +
`${c.reviewRounds} reported rounds`)
const p = assessment.pending
out.push(`pending: ${p.secondPass} item(s) awaiting a second gold standard ` +
`pass, ${p.adjudication} awaiting adjudication`)
out.push('')
out.push('acceptance criteria')
for (const cr of assessment.criteria) {
out.push(` [${cr.met ? 'met' : ' -'}] ${cr.name}: ${cr.detail}`)
}
out.push('')
out.push(assessment.goalsMet
? 'GOALS: met. The corpus satisfies every acceptance criterion.'
: 'GOALS: not met. Keep looping.')
}
process.stdout.write(out.join('\n') + '\n')
}
// ---------------------------------------------------------------------------
// Self-test, run against tools/fixtures/
// ---------------------------------------------------------------------------
function selftest() {
const dir = join(HERE, 'fixtures')
const cases = [
['valid-item.jsonl', 0],
['invalid-items.jsonl', 12],
]
let failures = 0
for (const [name, expected] of cases) {
const path = join(dir, name)
if (!existsSync(path)) {
process.stdout.write(`FAIL ${name}: fixture missing\n`)
failures++
continue
}
const { rows, errors } = readJsonl(path)
const seen = new Map()
const errs = [...errors]
for (const { line, value } of rows) {
errs.push(...validateItem(value, line, seen, path))
}
const ok = expected === 0 ? errs.length === 0 : errs.length >= expected
process.stdout.write(`${ok ? 'PASS' : 'FAIL'} ${name}: ` +
`${errs.length} errors, expected ${expected === 0 ? '0' : expected + ' or more'}\n`)
if (!ok) {
failures++
for (const e of errs) process.stdout.write(` ${e}\n`)
}
}
// Review records.
const reviewPath = join(dir, 'valid-review.jsonl')
if (existsSync(reviewPath)) {
const { rows, errors } = readJsonl(reviewPath)
const errs = [...errors]
for (const { line, value } of rows) {
errs.push(...validateReview(value, line, reviewPath))
}
const ok = errs.length === 0
process.stdout.write(`${ok ? 'PASS' : 'FAIL'} valid-review.jsonl: ` +
`${errs.length} errors, expected 0\n`)
if (!ok) {
failures++
for (const e of errs) process.stdout.write(` ${e}\n`)
}
} else {
process.stdout.write('FAIL valid-review.jsonl: fixture missing\n')
failures++
}
// An accept that also claims to be blocking is incoherent, and would strand
// the item: promoted, yet permanently counted against the clean criterion.
const incoherent = validateReview({ item_id: 'fi-0001', round: 1,
verdict: 'accept', reason_codes: ['CLEAN'], evidence: 'x'.repeat(21),
required_change: null, blocking: true }, 1, 'inline')
if (incoherent.some((e) => e.includes('cannot also be'))) {
process.stdout.write('PASS blocking accept rejected\n')
} else {
process.stdout.write('FAIL blocking accept was allowed\n')
failures++
}
// Two passes that disagree are a legal candidate and an illegal accepted item.
// If a disagreement were a schema error the loop would stop dead the first
// time a blind second pass did its job, so the rule binds acceptance only.
const base = JSON.parse(readFileSync(join(dir, 'valid-item.jsonl'), 'utf8')
.split('\n').filter((l) => l.trim() !== '')[0])
const split = (over) => ({
...base, adjudication: null,
gold_alt_passes: [
{ author: 'pass-a', alt: base.gold_alt, rationale: 'First pass.' },
{ author: 'pass-b', alt: 'Something else entirely', rationale: 'Second.' },
],
...over,
})
const asCandidate = validateItem(split({ status: 'candidate' }), 1, new Map(), 'inline')
const asAccepted = validateItem(split({ status: 'accepted' }), 1, new Map(), 'inline')
const asSettled = validateItem(split({ status: 'accepted',
adjudication: 'Pass A wins: the adjacent text does not name the destination.',
}), 1, new Map(), 'inline')
const states = [passState(split({})), passState({ ...base, gold_alt_passes: [
{ author: 'pass-a', alt: 'x', rationale: 'r' }] }), passState(base)]
if (asCandidate.length === 0 && asSettled.length === 0 &&
asAccepted.some((e) => e.includes('disagreeing passes')) &&
states.join(',') === 'disagreeing,one-pass,settled') {
process.stdout.write('PASS disagreeing passes block acceptance only\n')
} else {
process.stdout.write('FAIL disagreeing passes: candidate ' +
`[${asCandidate.join('; ')}], accepted [${asAccepted.join('; ')}], ` +
`adjudicated [${asSettled.join('; ')}], states ${states.join(',')}\n`)
failures++
}
// An empty corpus must not claim the goals are met.
const empty = assess([], [], [])
if (empty.goalsMet) {
process.stdout.write('FAIL empty corpus reported as complete\n')
failures++
} else {
process.stdout.write('PASS empty corpus is not complete\n')
}
// An unreviewed item must never satisfy the latest-review criterion.
const unreviewed = [JSON.parse(readFileSync(join(dir, 'valid-item.jsonl'), 'utf8')
.split('\n').filter((l) => l.trim() !== '')[1])]
const reviewCrit = (reviews, reports) => assess(unreviewed, reviews, reports)
.criteria.find((c) => c.name === 'items clean in latest review').met
if (reviewCrit([], [])) {
process.stdout.write('FAIL accepted-but-unreviewed item counted as clean\n')
failures++
} else if (!reviewCrit([{ item_id: 'fi-0002', round: 1, verdict: 'accept',
reason_codes: ['CLEAN'], evidence: 'x'.repeat(21), required_change: null,
blocking: false }], [])) {
process.stdout.write('FAIL reviewed clean item not counted as clean\n')
failures++
} else {
process.stdout.write('PASS latest-review criterion needs a review round\n')
}
// Reports must be ordered numerically, not lexicographically: with rounds 9,
// 10 and 11 on disk, the gate has to read 10 and 11 as the latest pair.
const scratch = mkdtempSync(join(tmpdir(), 'alt-corpus-rounds-'))
const writeReport = (n, blocking) => writeFileSync(
join(scratch, `round-${n}-report.md`),
`stub\n\nSTATUS: new-blocking-findings=${blocking ? 'yes' : 'no'}\n`)
writeReport(9, true)
writeReport(10, false)
writeReport(11, false)
const ordered = readRounds(scratch).reports.map((r) => r.name)
const orderOk = ordered.join(',') ===
'round-9-report.md,round-10-report.md,round-11-report.md'
const gateAfterOrder = assess([], [], readRounds(scratch).reports)
.criteria.find((c) => c.name === 'two consecutive quiet review rounds').met
writeReport(11, true)
const gateWhenLatestNoisy = assess([], [], readRounds(scratch).reports)
.criteria.find((c) => c.name === 'two consecutive quiet review rounds').met
rmSync(scratch, { recursive: true, force: true })
if (orderOk && gateAfterOrder === true && gateWhenLatestNoisy === false) {
process.stdout.write('PASS round reports ordered numerically\n')
} else {
process.stdout.write(`FAIL round ordering: [${ordered.join(', ')}], ` +
`gate ${gateAfterOrder}, noisy-latest gate ${gateWhenLatestNoisy}\n`)
failures++
}
// The quiet-rounds gate must need two consecutive quiet rounds.
const gate = (reports) => assess([], [], reports).criteria
.find((c) => c.name === 'two consecutive quiet review rounds').met
const gateCases = [
[[], false],
[[{ name: 'r1', newBlocking: false }], false],
[[{ name: 'r1', newBlocking: false }, { name: 'r2', newBlocking: true }], false],
[[{ name: 'r1', newBlocking: true }, { name: 'r2', newBlocking: false }], false],
[[{ name: 'r1', newBlocking: false }, { name: 'r2', newBlocking: false }], true],
[[{ name: 'r1', newBlocking: false }, { name: 'r2', newBlocking: null }], false],
]
let gateOk = true
gateCases.forEach(([reports, want], i) => {
const got = gate(reports)
if (got !== want) {
gateOk = false
process.stdout.write(`FAIL quiet-rounds gate case ${i}: ` +
`got ${got}, want ${want}\n`)
}
})
if (gateOk) process.stdout.write('PASS quiet-rounds gate\n')
else failures++
// Scaled targets must stay reachable. Seven sub-type minimums and five
// category minimums both have to fit inside the total, or the loop would run
// to the cap chasing a goal that arithmetic forbids.
const scaleCases = [BASELINE.total, 200, 100, 50, MIN_TARGET]
let scaleOk = targetsFor(BASELINE.total).perSubtype === BASELINE.perSubtype &&
targetsFor(BASELINE.total).perCategory === BASELINE.perCategory
for (const n of scaleCases) {
const t = targetsFor(n)
if (t.perSubtype * 7 > n || t.perCategory * 5 > n ||
t.domainsPerSubtype > t.perSubtype) {
scaleOk = false
process.stdout.write(`FAIL target ${n} scales to unreachable targets: ` +
`${t.perSubtype} per sub-type, ${t.perCategory} per category, ` +
`${t.domainsPerSubtype} domains per sub-type\n`)
}
}
if (scaleOk) process.stdout.write('PASS scaled targets stay reachable\n')
else failures++
// A malformed target file must stop the run rather than silently reverting to
// the 250-item goal, which would look like ordinary slow progress.
const tdir = mkdtempSync(join(tmpdir(), 'alt-corpus-target-'))
const targetCase = (body) => {
const p = join(tdir, 'target.txt')
writeFileSync(p, body)
return readTargetFile(p)
}
const targetOk =
readTargetFile(join(tdir, 'absent.txt')).value === null &&
targetCase('# goal for this run\n100\n').value === 100 &&
targetCase(' 100 \n').value === 100 &&
targetCase('one hundred\n').error !== undefined &&
targetCase('# only a comment\n').error !== undefined
rmSync(tdir, { recursive: true, force: true })
if (targetOk) process.stdout.write('PASS target file read strictly\n')
else {
process.stdout.write('FAIL target file parsing\n')
failures++
}
process.stdout.write(failures === 0
? '\nself-test passed\n'
: `\nself-test failed, ${failures} case(s)\n`)
return failures === 0 ? 0 : 3
}
// ---------------------------------------------------------------------------
// Main
// ---------------------------------------------------------------------------
function main(argv) {
let corpusPath = join(PROJECT, 'corpus', 'functional-images.jsonl')
let roundsDir = join(PROJECT, 'rounds')
let asJson = false
let targetArg = null
const needsValue = (arg, value) => {
if (value === undefined || value.startsWith('--')) {
process.stderr.write(`validate.mjs: ${arg} needs a path\n`)
return false
}
return true
}
for (let i = 0; i < argv.length; i++) {
const arg = argv[i]
if (arg === '--selftest') return selftest()
else if (arg === '--json') asJson = true
else if (arg === '--corpus') {
if (!needsValue(arg, argv[i + 1])) return 3
corpusPath = resolve(argv[++i])
} else if (arg === '--rounds') {
if (!needsValue(arg, argv[i + 1])) return 3
roundsDir = resolve(argv[++i])
} else if (arg === '--target') {
const value = argv[i + 1]
if (value === undefined || !/^\d+$/.test(value)) {
process.stderr.write('validate.mjs: --target needs a whole number of ' +
`accepted items, ${MIN_TARGET} or more\n`)
return 3
}
targetArg = Number(value)
i++
} else {
process.stderr.write(`validate.mjs: unknown argument "${arg}"\n` +
'usage: validate.mjs [--json] [--corpus FILE] [--rounds DIR] ' +
'[--target N] [--selftest]\n')
return 3
}
}
// Precedence: the flag, then the committed target file, then directive 00.
let target = targetArg
if (target === null) {
const fromFile = readTargetFile(join(dirname(corpusPath), 'target.txt'))
if (fromFile.error) {
process.stderr.write(`validate.mjs: ${fromFile.error}\n`)
return 3
}
target = fromFile.value === null ? BASELINE.total : fromFile.value
}
if (target < MIN_TARGET) {
process.stderr.write(`validate.mjs: a target of ${target} is too small to ` +
`satisfy the coverage targets. Use ${MIN_TARGET} or more.\n`)
return 3
}
TARGETS = targetsFor(target)
if (!existsSync(corpusPath)) {
const msg = `corpus file not found: ${corpusPath}\n` +
'The seeking agent creates it on the first round. Nothing to validate yet.\n'
if (asJson) {
process.stdout.write(JSON.stringify({
corpusPath, exists: false, goalsMet: false, schemaErrors: [],
}, null, 2) + '\n')
} else {
process.stdout.write(msg)
}
return 2
}
const corpus = readJsonl(corpusPath)
const schemaErrors = [...corpus.errors]
const seen = new Map()
const items = []
for (const { line, value } of corpus.rows) {
const errs = validateItem(value, line, seen, corpusPath)
schemaErrors.push(...errs)
items.push(value)
}
const reviews = []
if (existsSync(roundsDir)) {
const names = roundFiles(roundsDir, /^round-\d+-review\.jsonl$/)
for (const name of names) {
const path = join(roundsDir, name)
const parsed = readJsonl(path)
schemaErrors.push(...parsed.errors)
const fileRound = roundNumber(name)
for (const { line, value } of parsed.rows) {
schemaErrors.push(...validateReview(value, line, path))
// A record filed under the wrong round would be applied to the wrong
// round and would skew the gate, so the file name is authoritative.
if (isInt(value.round) && value.round !== fileRound) {
schemaErrors.push(`${path}:${line}: \`round\` is ${value.round} but ` +
`the file name says round ${fileRound}`)
}
reviews.push(value)
}
}
const knownIds = new Set(items.map((r) => r.id))
for (const r of reviews) {
if (isStr(r.item_id) && !knownIds.has(r.item_id)) {
schemaErrors.push(`${roundsDir}: review references unknown item ` +
`${r.item_id}`)
}
}
}
const { reports, errors: reportErrors } = readRounds(roundsDir)
schemaErrors.push(...reportErrors)
const assessment = assess(items, reviews, reports)
const report = { corpusPath, schemaErrors, assessment }
if (asJson) {
process.stdout.write(JSON.stringify({
corpusPath,
exists: true,
targets: TARGETS,
schemaErrors,
goalsMet: schemaErrors.length === 0 && assessment.goalsMet,
counts: assessment.counts,
criteria: assessment.criteria,
}, null, 2) + '\n')
} else {
printText(report)
}
if (schemaErrors.length > 0) return 2
return assessment.goalsMet ? 0 : 1
}
process.exit(main(process.argv.slice(2)))