a3s 0.9.6

a3s — A3S coding agent CLI; `a3s code` launches the interactive TUI
Documentation
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
fn research_report_artifacts_from_output_with_slug(
    output: &str,
    workspace: &Path,
    expected_slug: Option<&str>,
) -> Option<ResearchReportArtifacts> {
    output.lines().rev().find_map(|line| {
        let marker_at = line.find(RESEARCH_VIEW_MARKER)?;
        let raw = &line[marker_at + RESEARCH_VIEW_MARKER.len()..];
        let candidate = clean_research_report_marker_value(raw)?;
        let artifacts = trusted_research_report_artifacts(&candidate, workspace)?;
        match expected_slug {
            Some(slug) if !research_report_artifact_slug_matches(&artifacts, slug) => None,
            _ => Some(artifacts),
        }
    })
}
fn research_report_artifact_slug_matches(
    artifacts: &ResearchReportArtifacts,
    expected_slug: &str,
) -> bool {
    artifacts
        .html
        .parent()
        .and_then(Path::file_name)
        .and_then(|value| value.to_str())
        == Some(expected_slug)
}
fn clean_research_report_marker_value(raw: &str) -> Option<String> {
    let mut value = raw.trim();
    value = value
        .trim_start_matches(['`', '"', '\'', '<'])
        .trim_end_matches(['`', '"', '\'', '>', '.', ',', ';']);
    if value.is_empty() || value.starts_with("file://") {
        return None;
    }
    value
        .split_whitespace()
        .next()
        .filter(|s| !s.is_empty())
        .map(str::to_string)
}

fn trusted_research_report_artifacts(
    candidate: &str,
    workspace: &Path,
) -> Option<ResearchReportArtifacts> {
    let artifacts = trusted_research_report_artifact_paths(candidate, workspace)?;
    completed_research_report_artifacts(&artifacts).then_some(artifacts)
}

fn trusted_research_report_artifact_paths(
    candidate: &str,
    workspace: &Path,
) -> Option<ResearchReportArtifacts> {
    let root = workspace.canonicalize().ok()?;
    let candidate = Path::new(candidate);
    let unresolved_path = if candidate.is_absolute() {
        candidate.to_path_buf()
    } else {
        workspace.join(candidate)
    };
    let unresolved_metadata = std::fs::symlink_metadata(&unresolved_path).ok()?;
    if unresolved_metadata.file_type().is_symlink() || !unresolved_metadata.is_file() {
        return None;
    }
    let path = unresolved_path.canonicalize().ok()?;
    if !is_nonempty_file(&path) || !path.starts_with(&root) || !is_html_path(&path) {
        return None;
    }
    let rel = path.strip_prefix(&root).ok()?;
    let mut components = rel.components();
    let first = components.next()?.as_os_str();
    let second = components.next()?.as_os_str();
    let slug = components.next()?.as_os_str();
    let file = components.next()?.as_os_str();
    if components.next().is_some() {
        return None;
    }
    if first != std::ffi::OsStr::new(".a3s") || second != std::ffi::OsStr::new("research") {
        return None;
    }
    if slug.is_empty() || file != std::ffi::OsStr::new("index.html") {
        return None;
    }
    let markdown_path = path.parent()?.join("report.md");
    let markdown_metadata = std::fs::symlink_metadata(&markdown_path).ok()?;
    if markdown_metadata.file_type().is_symlink() || !markdown_metadata.is_file() {
        return None;
    }
    let markdown = markdown_path.canonicalize().ok()?;
    if !is_nonempty_file(&markdown)
        || markdown.parent() != path.parent()
        || markdown.file_name() != Some(std::ffi::OsStr::new("report.md"))
    {
        return None;
    }
    Some(ResearchReportArtifacts {
        markdown,
        html: path,
    })
}

fn completed_research_report_artifacts(artifacts: &ResearchReportArtifacts) -> bool {
    let markdown = read_small_utf8_file(&artifacts.markdown);
    let html = read_small_utf8_file(&artifacts.html);
    let (Some(markdown), Some(html)) = (markdown, html) else {
        return false;
    };
    !looks_like_deep_research_fallback_draft(&markdown)
        && !looks_like_deep_research_fallback_draft(&html)
        && !looks_like_deep_research_recovery_report(&markdown)
        && !looks_like_deep_research_recovery_report(&html)
        && !is_deep_research_model_failure_text(&markdown)
        && !is_deep_research_model_failure_text(&html)
        && !deep_research_output_has_internal_leak(&markdown)
        && !deep_research_output_has_internal_leak(&html)
        && complete_html_document(&html)
        && has_research_report_substance(&markdown, &html)
}

fn deep_research_report_sources_trace_workflow(
    artifacts: &ResearchReportArtifacts,
    query: &str,
    workflow_output: &str,
    workflow_metadata: Option<&serde_json::Value>,
) -> bool {
    let markdown = read_small_utf8_file(&artifacts.markdown);
    let html = read_small_utf8_file(&artifacts.html);
    let (Some(markdown), Some(html)) = (markdown, html) else {
        return false;
    };

    deep_research_report_content_sources_trace_workflow(
        &markdown,
        &html,
        query,
        workflow_output,
        workflow_metadata,
    )
}

fn deep_research_report_content_sources_trace_workflow(
    markdown: &str,
    html: &str,
    query: &str,
    workflow_output: &str,
    workflow_metadata: Option<&serde_json::Value>,
) -> bool {
    deep_research_report_source_trace_diagnostic(
        markdown,
        html,
        query,
        workflow_output,
        workflow_metadata,
    )
    .is_ok()
}

fn deep_research_report_source_trace_diagnostic(
    markdown: &str,
    html: &str,
    query: &str,
    workflow_output: &str,
    workflow_metadata: Option<&serde_json::Value>,
) -> Result<(), String> {
    let anchors = deep_research_workflow_source_anchors(workflow_output, workflow_metadata);
    if anchors.is_empty() {
        // A DeepResearch report is only "completed" when it can be traced to
        // evidence gathered by this run. Failing open here lets a polished
        // model answer—or an old deterministic-slug report—mask a collection
        // failure that captured no source at all. Callers will materialize an
        // explicit recovery report instead.
        return Err(
            "source trace rejected: the workflow captured no traceable research sources"
                .to_string(),
        );
    }

    let observed = anchors.into_iter().collect::<HashSet<_>>();
    let (cited, has_explicit_source_citation) = markdown_report_source_anchors(markdown, query);
    let html_cited = html_report_source_anchors(html, query);
    if !has_explicit_source_citation {
        return Err(
            "source trace rejected: the report has no explicit citation item under a Sources, References, 来源, 参考文献, or 引用 section"
                .to_string(),
        );
    }
    if html_cited.is_empty() {
        return Err(
            "source trace rejected: the rendered HTML contains no traceable source citation"
                .to_string(),
        );
    }

    let mut unmatched = cited
        .iter()
        .chain(html_cited.iter())
        .filter(|citation| {
            !reported_research_source_candidates(citation)
                .iter()
                .any(|candidate| observed.contains(candidate))
        })
        .cloned()
        .collect::<Vec<_>>();
    unmatched.sort();
    unmatched.dedup();
    if unmatched.is_empty() {
        return Ok(());
    }

    let displayed = unmatched.iter().take(8).cloned().collect::<Vec<_>>();
    let omitted = unmatched.len().saturating_sub(displayed.len());
    let mut message = format!(
        "source trace rejected: {} citation{} were not observed in this run: {}",
        unmatched.len(),
        if unmatched.len() == 1 { "" } else { "s" },
        displayed.join(", ")
    );
    if omitted > 0 {
        message.push_str(&format!(", plus {omitted} more"));
    }
    Err(message)
}

fn sanitize_unobserved_markdown_http_citations(
    markdown: &str,
    query: &str,
    workflow_output: &str,
    workflow_metadata: Option<&serde_json::Value>,
) -> String {
    let observed = deep_research_workflow_source_anchors(workflow_output, workflow_metadata)
        .into_iter()
        .collect::<HashSet<_>>();
    if observed.is_empty() {
        return markdown.to_string();
    }

    let (cited, _) = markdown_report_source_anchors(markdown, query);
    let unobserved = cited
        .into_iter()
        .filter(|citation| {
            !reported_research_source_candidates(citation)
                .iter()
                .any(|candidate| observed.contains(candidate))
        })
        .collect::<HashSet<_>>();
    if unobserved.is_empty() {
        return markdown.to_string();
    }

    let invalid_targets = http_source_targets(markdown)
        .into_iter()
        .filter(|target| {
            reported_research_source_candidates(target)
                .iter()
                .any(|candidate| unobserved.contains(candidate))
        })
        .collect::<HashSet<_>>();
    if invalid_targets.is_empty() {
        return markdown.to_string();
    }

    let mut cleaned = Vec::new();
    let mut in_source_section = false;
    for line in markdown.lines() {
        let trimmed = line.trim();
        if trimmed.starts_with('#') {
            let heading = trimmed.trim_start_matches('#').trim();
            in_source_section = is_source_section_heading(heading);
            cleaned.push(line.to_string());
            continue;
        }

        let line_targets = http_source_targets(line);
        let invalid_count = line_targets
            .iter()
            .filter(|target| invalid_targets.contains(*target))
            .count();
        if in_source_section
            && is_explicit_source_citation_line(trimmed)
            && invalid_count > 0
            && invalid_count == line_targets.len()
        {
            continue;
        }
        cleaned.push(strip_unobserved_http_targets(line, &invalid_targets));
    }

    let mut output = cleaned.join("\n");
    if markdown.ends_with('\n') {
        output.push('\n');
    }
    output
}

fn strip_unobserved_http_targets(line: &str, invalid_targets: &HashSet<String>) -> String {
    let mut output = line.to_string();
    let mut targets = invalid_targets.iter().collect::<Vec<_>>();
    targets.sort_by_key(|target| std::cmp::Reverse(target.len()));
    for target in targets {
        let mut cursor = 0;
        while let Some(offset) = output[cursor..].find(target.as_str()) {
            let start = cursor + offset;
            let end = start + target.len();
            if let Some((link_start, link_end, label)) =
                markdown_link_replacement(&output, start, end)
            {
                output.replace_range(link_start..link_end, &label);
                cursor = link_start + label.len();
            } else {
                output.replace_range(start..end, "");
                cursor = start;
            }
        }
    }
    output
}

fn markdown_link_replacement(
    text: &str,
    target_start: usize,
    target_end: usize,
) -> Option<(usize, usize, String)> {
    if target_start < 2 || text.get(target_start - 2..target_start)? != "](" {
        return None;
    }
    let label_start = text[..target_start - 2].rfind('[')?;
    if text[label_start + 1..target_start - 2]
        .chars()
        .any(|ch| matches!(ch, '\n' | '\r'))
    {
        return None;
    }
    let suffix = &text[target_end..];
    let close_offset = suffix.find(')')?;
    if suffix[..close_offset]
        .chars()
        .any(|ch| matches!(ch, '\n' | '\r'))
    {
        return None;
    }
    let link_end = target_end + close_offset + 1;
    let label = text[label_start + 1..target_start - 2].to_string();
    Some((label_start, link_end, label))
}

fn markdown_report_source_anchors(markdown: &str, query: &str) -> (Vec<String>, bool) {
    let mut anchors = Vec::new();
    let mut seen = HashSet::new();
    let mut in_source_section = false;

    for line in markdown.lines() {
        let trimmed = line.trim();
        if trimmed.starts_with('#') {
            let heading = trimmed.trim_start_matches('#').trim();
            in_source_section = is_source_section_heading(heading);
            continue;
        }
        if in_source_section && is_explicit_source_citation_line(trimmed) {
            collect_source_anchors_from_citation_line(trimmed, &mut anchors, &mut seen);
        }
    }
    let explicit_count = anchors.len();
    let query_heading = format!("# {}", markdown_plain_text(query));
    let query_targets = http_source_targets(&markdown_plain_text(query))
        .into_iter()
        .filter_map(|target| canonical_query_http_target(&target))
        .collect::<HashSet<_>>();
    let mut checked_report_heading = false;
    for line in markdown.lines() {
        let trimmed = line.trim();
        if !checked_report_heading && trimmed.starts_with("# ") {
            checked_report_heading = true;
            let heading_targets = http_source_targets(trimmed);
            let is_query_derived_heading = !heading_targets.is_empty()
                && heading_targets.iter().all(|target| {
                    canonical_query_http_target(target)
                        .is_some_and(|canonical| query_targets.contains(&canonical))
                });
            if trimmed == query_heading || is_query_derived_heading {
                continue;
            }
        }
        collect_http_source_anchors(line, &mut anchors, &mut seen);
        collect_markdown_link_anchors(line, &mut anchors, &mut seen);
    }
    (anchors, explicit_count > 0)
}

fn html_report_source_anchors(html: &str, query: &str) -> Vec<String> {
    let mut anchors = Vec::new();
    let mut seen = HashSet::new();
    let decoded = html.replace("&amp;", "&");
    let without_query_title = remove_matching_html_element(&decoded, "title", query);
    let without_query_heading = remove_matching_html_element(&without_query_title, "h1", query);
    collect_http_source_anchors(&without_query_heading, &mut anchors, &mut seen);
    collect_html_link_anchors(&without_query_heading, &mut anchors, &mut seen);
    // Inline code outside a Sources section commonly names APIs, repository
    // slugs, or unavailable candidate URLs. It is prose, not a citation. The
    // source-section collector below still accepts code-formatted local paths
    // where the author explicitly presents them as sources.
    collect_html_source_section_local_anchors(&without_query_heading, &mut anchors, &mut seen);
    anchors
}

fn remove_matching_html_element(html: &str, tag: &str, query: &str) -> String {
    let lower = html.to_ascii_lowercase();
    let opening = format!("<{tag}");
    let closing = format!("</{tag}>");
    let expected =
        normalize_html_display_text(&markdown_backslash_unescape(&markdown_plain_text(query)));
    let mut cursor = 0;

    while let Some(offset) = lower[cursor..].find(&opening) {
        let start = cursor + offset;
        let name_end = start + opening.len();
        if !lower[name_end..]
            .chars()
            .next()
            .is_some_and(|ch| ch == '>' || ch.is_whitespace())
        {
            cursor = name_end;
            continue;
        }
        let Some(open_end_offset) = lower[name_end..].find('>') else {
            break;
        };
        let content_start = name_end + open_end_offset + 1;
        let Some(close_offset) = lower[content_start..].find(&closing) else {
            break;
        };
        let content_end = content_start + close_offset;
        let display =
            normalize_html_display_text(&strip_html_tags(&html[content_start..content_end]));
        let element_end = content_end + closing.len();
        let (has_http_target, targets_match_query) =
            html_element_targets_match_query(&html[content_start..content_end], &display, query);
        if targets_match_query && (display == expected || has_http_target) {
            let mut filtered = String::with_capacity(html.len());
            filtered.push_str(&html[..start]);
            filtered.push(' ');
            filtered.push_str(&html[element_end..]);
            return filtered;
        }
        cursor = element_end;
    }
    html.to_string()
}

fn html_element_targets_match_query(html: &str, display: &str, query: &str) -> (bool, bool) {
    let query_targets = http_source_targets(&markdown_plain_text(query))
        .into_iter()
        .filter_map(|target| canonical_query_http_target(&target))
        .collect::<HashSet<_>>();
    let targets = http_source_targets(display)
        .into_iter()
        .chain(html_link_targets(html))
        .collect::<Vec<_>>();
    let has_http_target = !targets.is_empty();
    let targets_match_query = targets.iter().all(|target| {
        canonical_query_http_target(target)
            .is_some_and(|canonical| query_targets.contains(&canonical))
    });
    (has_http_target, targets_match_query)
}

fn canonical_query_http_target(value: &str) -> Option<String> {
    let url = reqwest::Url::parse(value).ok()?;
    if !matches!(url.scheme(), "http" | "https") || url.host_str()?.is_empty() {
        return None;
    }
    Some(url.to_string())
}

fn normalize_html_display_text(text: &str) -> String {
    text.replace("&lt;", "<")
        .replace("&gt;", ">")
        .replace("&quot;", "\"")
        .replace("&#39;", "'")
        .replace("&amp;", "&")
        .split_whitespace()
        .collect::<Vec<_>>()
        .join(" ")
}

fn is_source_section_heading(heading: &str) -> bool {
    let lower = heading.to_ascii_lowercase();
    lower.contains("source")
        || lower.contains("reference")
        || heading.contains("来源")
        || heading.contains("参考文献")
        || heading.contains("引用")
}

fn is_explicit_source_citation_line(line: &str) -> bool {
    let lower = line.to_ascii_lowercase();
    if lower.contains("http://")
        || lower.contains("https://")
        || line.contains("](")
        || line.contains('`')
        || line.starts_with('|')
    {
        return true;
    }
    if ["- ", "* ", "+ "]
        .iter()
        .any(|prefix| line.starts_with(prefix))
    {
        return true;
    }
    line.split_once(". ").is_some_and(|(number, _)| {
        !number.is_empty() && number.chars().all(|ch| ch.is_ascii_digit())
    })
}

fn collect_source_anchors_from_citation_line(
    line: &str,
    anchors: &mut Vec<String>,
    seen: &mut HashSet<String>,
) {
    let has_explicit_target = !http_source_targets(line).is_empty() || line.contains("](");
    collect_http_source_anchors(line, anchors, seen);
    collect_markdown_link_anchors(line, anchors, seen);

    // When a citation line already has an explicit link target, the remaining
    // prose is its display label. Treating labels such as `docs.rs` or
    // `crates.io` as local file citations creates false, unobserved sources.
    // A mixed local/web citation can still express the local source as a
    // Markdown link target; otherwise it should use a separate source item.
    if has_explicit_target {
        return;
    }

    let mut code = false;
    for segment in line.split('`') {
        if code {
            push_canonical_source_anchor(segment, anchors, seen);
        }
        code = !code;
    }

    for token in line.split_whitespace() {
        push_local_citation_anchor(token, anchors, seen);
    }
}

fn collect_http_source_anchors(text: &str, anchors: &mut Vec<String>, seen: &mut HashSet<String>) {
    for target in http_source_targets(text) {
        push_canonical_source_anchor(&target, anchors, seen);
    }
}

fn http_source_targets(text: &str) -> Vec<String> {
    let lower = text.to_ascii_lowercase();
    let mut cursor = 0;
    let mut targets = Vec::new();
    while cursor < text.len() {
        let next = ["http://", "https://"]
            .into_iter()
            .filter_map(|prefix| lower[cursor..].find(prefix).map(|index| cursor + index))
            .min();
        let Some(start) = next else {
            break;
        };
        let mut nested_parentheses = 0usize;
        let end = text[start..]
            .char_indices()
            .find_map(|(offset, ch)| {
                if ch.is_whitespace() || matches!(ch, '<' | '>' | '"' | '\'' | '`' | ']' | '}') {
                    return Some(start + offset);
                }
                match ch {
                    '(' => nested_parentheses += 1,
                    ')' if nested_parentheses == 0 => return Some(start + offset),
                    ')' => nested_parentheses -= 1,
                    _ => {}
                }
                None
            })
            .unwrap_or(text.len());
        let candidate = text[start..end].trim_end_matches(['.', ',', ';', ':', '!', '?']);
        targets.push(candidate.to_string());
        cursor = end;
    }
    targets
}

fn collect_markdown_link_anchors(
    line: &str,
    anchors: &mut Vec<String>,
    seen: &mut HashSet<String>,
) {
    let mut cursor = 0;
    while let Some(start) = line[cursor..].find("](") {
        let target_start = cursor + start + 2;
        let mut nested_parentheses = 0usize;
        let mut escaped = false;
        let end = line[target_start..]
            .char_indices()
            .find_map(|(offset, ch)| {
                if escaped {
                    escaped = false;
                    return None;
                }
                if ch == '\\' {
                    escaped = true;
                    return None;
                }
                match ch {
                    '(' => nested_parentheses += 1,
                    ')' if nested_parentheses == 0 => return Some(offset),
                    ')' => nested_parentheses -= 1,
                    _ => {}
                }
                None
            });
        let Some(end) = end else {
            break;
        };
        push_canonical_source_anchor(&line[target_start..target_start + end], anchors, seen);
        cursor = target_start + end + 1;
    }
}

fn collect_html_link_anchors(html: &str, anchors: &mut Vec<String>, seen: &mut HashSet<String>) {
    for target in html_link_targets(html) {
        push_canonical_source_anchor(&target, anchors, seen);
    }
}

fn html_link_targets(html: &str) -> Vec<String> {
    let lower = html.to_ascii_lowercase();
    let mut targets = Vec::new();
    for attribute in ["href", "src"] {
        let mut cursor = 0;
        while let Some(found) = lower[cursor..].find(attribute) {
            let attribute_start = cursor + found;
            let has_name_boundary = attribute_start == 0
                || lower[..attribute_start]
                    .chars()
                    .next_back()
                    .is_some_and(|ch| ch == '<' || ch.is_whitespace());
            if !has_name_boundary {
                cursor = attribute_start + attribute.len();
                continue;
            }
            let mut value_start = attribute_start + attribute.len();
            while html[value_start..]
                .chars()
                .next()
                .is_some_and(char::is_whitespace)
            {
                value_start += html[value_start..].chars().next().unwrap().len_utf8();
            }
            if !html[value_start..].starts_with('=') {
                cursor = value_start;
                continue;
            }
            value_start += 1;
            while html[value_start..]
                .chars()
                .next()
                .is_some_and(char::is_whitespace)
            {
                value_start += html[value_start..].chars().next().unwrap().len_utf8();
            }
            let Some(quote) = html[value_start..].chars().next() else {
                break;
            };
            if !matches!(quote, '"' | '\'') {
                cursor = value_start + quote.len_utf8();
                continue;
            }
            let target_start = value_start + quote.len_utf8();
            let Some(end) = html[target_start..].find(quote) else {
                break;
            };
            targets.push(html[target_start..target_start + end].to_string());
            cursor = target_start + end + quote.len_utf8();
        }
    }
    targets
}

fn collect_html_source_section_local_anchors(
    html: &str,
    anchors: &mut Vec<String>,
    seen: &mut HashSet<String>,
) {
    let lower = html.to_ascii_lowercase();
    let mut cursor = 0;
    let mut in_source_section = false;

    while cursor < html.len() {
        let Some(open_offset) = lower[cursor..].find('<') else {
            if in_source_section {
                for token in html[cursor..].split_whitespace() {
                    push_local_citation_anchor(token, anchors, seen);
                }
            }
            break;
        };
        let open = cursor + open_offset;
        if in_source_section {
            for token in html[cursor..open].split_whitespace() {
                push_local_citation_anchor(token, anchors, seen);
            }
        }
        let Some(close_offset) = lower[open..].find('>') else {
            break;
        };
        let close = open + close_offset;
        let tag = lower[open + 1..close].trim();
        let name = tag
            .split_whitespace()
            .next()
            .unwrap_or_default()
            .trim_end_matches('/');
        if matches!(name, "h1" | "h2" | "h3" | "h4" | "h5" | "h6") {
            let closing = format!("</{name}>");
            let heading_start = close + 1;
            let Some(heading_offset) = lower[heading_start..].find(&closing) else {
                in_source_section = false;
                cursor = heading_start;
                continue;
            };
            let heading_end = heading_start + heading_offset;
            let heading = strip_html_tags(&html[heading_start..heading_end]);
            in_source_section = is_source_section_heading(heading.trim());
            cursor = heading_end + closing.len();
        } else if in_source_section && matches!(name, "li" | "p" | "tr") && !tag.starts_with('/') {
            let closing = format!("</{name}>");
            let block_start = close + 1;
            let Some(block_offset) = lower[block_start..].find(&closing) else {
                cursor = block_start;
                continue;
            };
            let block_end = block_start + block_offset;
            let block = &html[block_start..block_end];
            // Links are collected by `collect_html_link_anchors`. Everything
            // else in the same block is a display label, not another source.
            if html_link_targets(block).is_empty() {
                for token in strip_html_tags(block).split_whitespace() {
                    push_local_citation_anchor(token, anchors, seen);
                }
            }
            cursor = block_end + closing.len();
        } else {
            cursor = close + 1;
        }
    }
}

fn push_local_citation_anchor(
    candidate: &str,
    anchors: &mut Vec<String>,
    seen: &mut HashSet<String>,
) {
    let token = candidate.trim_matches(|ch: char| {
        ch.is_whitespace()
            || matches!(
                ch,
                '`' | '"'
                    | '\''
                    | '<'
                    | '>'
                    | '('
                    | ')'
                    | '['
                    | ']'
                    | '{'
                    | '}'
                    | ','
                    | ';'
                    | '.'
                    | ':'
                    | '!'
                    | '?'
            )
    });
    let has_source_extension = Path::new(token)
        .extension()
        .and_then(|extension| extension.to_str())
        .is_some_and(|extension| extension.chars().any(|ch| ch.is_ascii_alphabetic()));
    if token.contains('/') || token.contains('\\') || has_source_extension {
        push_canonical_source_anchor(token, anchors, seen);
    }
}

fn push_canonical_source_anchor(
    candidate: &str,
    anchors: &mut Vec<String>,
    seen: &mut HashSet<String>,
) {
    let candidate = candidate.trim();
    let lower = candidate.to_ascii_lowercase();
    let mut mark_untrusted = || {
        let marker = "untrusted-report-url".to_string();
        if seen.insert(marker.clone()) {
            anchors.push(marker);
        }
    };
    if candidate.starts_with("//") || lower.contains("&#") || lower.contains("&colon;") {
        mark_untrusted();
        return;
    }
    if let Ok(url) = reqwest::Url::parse(candidate) {
        if !matches!(url.scheme(), "http" | "https") {
            mark_untrusted();
            return;
        }
        if !url.username().is_empty() || url.password().is_some() {
            mark_untrusted();
            return;
        }
    }
    if let Some(anchor) = canonical_research_source_anchor(candidate) {
        if seen.insert(anchor.clone()) {
            anchors.push(anchor);
        }
    }
}

fn read_small_utf8_file(path: &Path) -> Option<String> {
    const MAX_REPORT_VALIDATION_BYTES: u64 = 2 * 1024 * 1024;
    let metadata = std::fs::symlink_metadata(path).ok()?;
    if metadata.file_type().is_symlink()
        || !metadata.is_file()
        || metadata.len() == 0
        || metadata.len() > MAX_REPORT_VALIDATION_BYTES
    {
        return None;
    }
    std::fs::read_to_string(path).ok()
}

fn complete_html_document(html: &str) -> bool {
    let lower = html.to_ascii_lowercase();
    lower.contains("<html")
        && lower.contains("</html>")
        && lower.contains("<body")
        && lower.contains("</body>")
        && lower.contains("name=\"viewport\"")
        && lower.contains("<title>")
        && lower.contains("</title>")
        && lower.contains("<style")
        && lower.matches("<h1").count() == 1
        && lower.contains("@media")
        && lower.contains("max-width")
        && lower.contains("@media print")
        && lower.contains(":focus")
        && lower.contains("overflow-x")
        && lower.contains("clamp(")
        && !lower.contains("<script")
        && !lower.contains("body{display:none")
        && !lower.contains("body {display:none")
        && !lower.contains("font-size:0")
        && !lower.contains("font-size:1px")
        && !lower.contains("width:200vw")
}