Antix5 commited on
Commit
11725dc
·
verified ·
1 Parent(s): bcaee6a

V2 end-to-end test: 60/76 cases (part 2)

Browse files
Files changed (26) hide show
  1. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/input.csv +11 -0
  2. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/audit.json +27 -0
  3. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/corrected.csv +11 -0
  4. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/review.html +18 -0
  5. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/reference.csv +11 -0
  6. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/result.json +71 -0
  7. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/review.html +18 -0
  8. evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/schema.xml +40 -0
  9. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/input.csv +12 -0
  10. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/audit.json +15 -0
  11. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/corrected.csv +12 -0
  12. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/review.html +4 -0
  13. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/reference.csv +12 -0
  14. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/result.json +39 -0
  15. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/review.html +4 -0
  16. evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/schema.xml +47 -0
  17. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/input.csv +13 -0
  18. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/audit.json +38 -0
  19. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/corrected.csv +13 -0
  20. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/review.html +28 -0
  21. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/reference.csv +13 -0
  22. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/result.json +99 -0
  23. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/review.html +28 -0
  24. evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/schema.xml +43 -0
  25. evaluation/v2/review.html +98 -2
  26. evaluation/v2/summary.json +45 -44
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/input.csv ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Identyfikator,Pozycja,Zarezerwowane,Obecni,Nieobecni,Data rozpoczęcia,Uwagi
2
+ 000001,Warsztaty ceramiki,139,26,113,2026-01-31,Oczekuje na ponowną kontrolę
3
+ 000002,Zajęcia biblioteczne,104,19,85,2026-07-06,"Skorzystać z wejścia północnego
4
+ Najpierw zgłosić się do recepcji"
5
+ 000003,Zajęcia biblioteczne,220,100,120,2025-08-06,"Etykieta „priorytet”, pozostawić widoczną"
6
+ 000004,Kurs językowy,33,8,25,2024-05-18,"Skorzystać z wejścia północnego
7
+ Najpierw zgłosić się do recepcji"
8
+ 000007,Zajęcia biblioteczne,178,175,3,2024-09-14,
9
+ 000008,Warsztaty ceramiki,65,38,27,2024-06-30,Sprawdzono przy odbiorze
10
+ 000018,Kurs językowy,176,58,118,2026-01-05,
11
+ 000026,Kurs językowy,193,92,101,2025-01-12,Zachowano odniesienie: UTF-8 / v2.1
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/audit.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metrics": {
3
+ "device": "cuda",
4
+ "rows": 8,
5
+ "cells": 56,
6
+ "threshold": 0.9,
7
+ "load_seconds": 0.0,
8
+ "inference_seconds": 0.39,
9
+ "cells_per_second": 143.28,
10
+ "peak_gpu_bytes": 1925493248,
11
+ "detected_spans": 1,
12
+ "changed_cells": 1
13
+ },
14
+ "proposals": [
15
+ {
16
+ "row": 1,
17
+ "column": 1,
18
+ "start": 3,
19
+ "end": 5,
20
+ "replacement": "Zajęcia biblioteczne",
21
+ "category": "text.encoding",
22
+ "perplexity": 1.0003071924485039,
23
+ "status": "applied",
24
+ "route": "model"
25
+ }
26
+ ]
27
+ }
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/corrected.csv ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Identyfikator,Pozycja,Zarezerwowane,Obecni,Nieobecni,Data rozpoczęcia,Uwagi
2
+ 000001,Warsztaty ceramiki,139,26,113,2026-01-31,Oczekuje na ponowną kontrolę
3
+ 000002,Zajęcia biblioteczne,104,19,85,2026-07-06,"Skorzystać z wejścia północnego
4
+ Najpierw zgłosić się do recepcji"
5
+ 000003,Zajęcia biblioteczne,220,100,120,2025-08-06,"Etykieta „priorytet”, pozostawić widoczną"
6
+ 000004,Kurs językowy,33,8,25,2024-05-18,"Skorzystać z wejścia północnego
7
+ Najpierw zgłosić się do recepcji"
8
+ 000007,Zajęcia biblioteczne,178,175,3,2024-09-14,
9
+ 000008,Warsztaty ceramiki,65,38,27,2024-06-30,Sprawdzono przy odbiorze
10
+ 000018,Kurs językowy,176,58,118,2026-01-05,
11
+ 000026,Kurs językowy,193,92,101,2025-01-12,Zachowano odniesienie: UTF-8 / v2.1
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/output/review.html ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>56</strong><span>Cells scanned</span></div><div class="card"><strong>—</strong><span>Errors in input</span></div><div class="card"><strong>—</strong><span>Errors repaired</span></div><div class="card"><strong>—</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.39 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr><tr><td>2</td><td>Pozycja</td><td>&quot;Zajęcia biblioteczne&quot;</td><td>&quot;Zajęcia biblioteczne&quot;</td><td>—</td><td class="bad">Changed<small>text.encoding</small></td></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>Identyfikator</th><th>Pozycja</th><th>Zarezerwowane</th><th>Obecni</th><th>Nieobecni</th><th>Data rozpoczęcia</th><th>Uwagi</th></tr><tr><td>1</td><td class="">000001</td><td class="">Warsztaty ceramiki</td><td class="">139</td><td class="">26</td><td class="">113</td><td class="">2026-01-31</td><td class="">Oczekuje na ponowną kontrolę</td></tr><tr><td>2</td><td class="">000002</td><td class="changed">Zajęcia biblioteczne</td><td class="">104</td><td class="">19</td><td class="">85</td><td class="">2026-07-06</td><td class="">Skorzystać z wejścia północnego
5
+ Najpierw zgłosić się do recepcji</td></tr><tr><td>3</td><td class="">000003</td><td class="">Zajęcia biblioteczne</td><td class="">220</td><td class="">100</td><td class="">120</td><td class="">2025-08-06</td><td class="">Etykieta „priorytet”, pozostawić widoczną</td></tr><tr><td>4</td><td class="">000004</td><td class="">Kurs językowy</td><td class="">33</td><td class="">8</td><td class="">25</td><td class="">2024-05-18</td><td class="">Skorzystać z wejścia północnego
6
+ Najpierw zgłosić się do recepcji</td></tr><tr><td>5</td><td class="">000007</td><td class="">Zajęcia biblioteczne</td><td class="">178</td><td class="">175</td><td class="">3</td><td class="">2024-09-14</td><td class=""></td></tr><tr><td>6</td><td class="">000008</td><td class="">Warsztaty ceramiki</td><td class="">65</td><td class="">38</td><td class="">27</td><td class="">2024-06-30</td><td class="">Sprawdzono przy odbiorze</td></tr><tr><td>7</td><td class="">000018</td><td class="">Kurs językowy</td><td class="">176</td><td class="">58</td><td class="">118</td><td class="">2026-01-05</td><td class=""></td></tr><tr><td>8</td><td class="">000026</td><td class="">Kurs językowy</td><td class="">193</td><td class="">92</td><td class="">101</td><td class="">2025-01-12</td><td class="">Zachowano odniesienie: UTF-8 / v2.1</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[
7
+ {
8
+ &quot;row&quot;: 1,
9
+ &quot;column&quot;: 1,
10
+ &quot;start&quot;: 3,
11
+ &quot;end&quot;: 5,
12
+ &quot;replacement&quot;: &quot;Zajęcia biblioteczne&quot;,
13
+ &quot;category&quot;: &quot;text.encoding&quot;,
14
+ &quot;perplexity&quot;: 1.0003071924485039,
15
+ &quot;status&quot;: &quot;applied&quot;,
16
+ &quot;route&quot;: &quot;model&quot;
17
+ }
18
+ ]</pre></details>
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/reference.csv ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Identyfikator,Pozycja,Zarezerwowane,Obecni,Nieobecni,Data rozpoczęcia,Uwagi
2
+ 000001,Warsztaty ceramiki,139,26,113,2026-01-31,Oczekuje na ponowną kontrolę
3
+ 000002,Zajęcia biblioteczne,104,19,85,2026-07-06,"Skorzystać z wejścia północnego
4
+ Najpierw zgłosić się do recepcji"
5
+ 000003,Zajęcia biblioteczne,220,100,120,2025-08-06,"Etykieta „priorytet”, pozostawić widoczną"
6
+ 000004,Kurs językowy,33,8,25,2024-05-18,"Skorzystać z wejścia północnego
7
+ Najpierw zgłosić się do recepcji"
8
+ 000007,Zajęcia biblioteczne,178,175,3,2024-09-14,
9
+ 000008,Warsztaty ceramiki,65,38,27,2024-06-30,Sprawdzono przy odbiorze
10
+ 000018,Kurs językowy,176,58,118,2026-01-05,
11
+ 000026,Kurs językowy,193,92,101,2025-01-12,Zachowano odniesienie: UTF-8 / v2.1
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/result.json ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "id": "8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open",
3
+ "language": "pl",
4
+ "roles": [
5
+ "open_positive:text.encoding"
6
+ ],
7
+ "counts": {
8
+ "cells": 56,
9
+ "valid_cells": 55,
10
+ "exactly_repaired_cells": 1,
11
+ "valid_cells_harmed": 0,
12
+ "detected_faulty_cells": 1,
13
+ "flagged_valid_cells": 0,
14
+ "faulty_cells": 1
15
+ },
16
+ "categories": {
17
+ "text.encoding": {
18
+ "gold_cells": 1,
19
+ "detected_cells": 1,
20
+ "exactly_repaired_cells": 1
21
+ },
22
+ "text.invisible": {},
23
+ "text.spelling": {},
24
+ "format.number": {},
25
+ "format.date": {},
26
+ "format.boolean": {},
27
+ "format.case": {},
28
+ "format.whitespace": {},
29
+ "format.unit": {},
30
+ "schema.type": {},
31
+ "schema.identifier": {},
32
+ "schema.category": {},
33
+ "consistency.unit": {},
34
+ "consistency.dependency": {},
35
+ "consistency.temporal": {},
36
+ "missing.required": {},
37
+ "missing.conditional": {},
38
+ "structure.shift": {}
39
+ },
40
+ "decisions": {
41
+ "applied": 1
42
+ },
43
+ "inference_seconds": 0.39,
44
+ "details": [
45
+ {
46
+ "row": 1,
47
+ "column": "Pozycja",
48
+ "original": "Zajęcia biblioteczne",
49
+ "prediction": "Zajęcia biblioteczne",
50
+ "reference": "Zajęcia biblioteczne",
51
+ "categories": [
52
+ "text.encoding"
53
+ ],
54
+ "status": "repaired",
55
+ "decisions": [
56
+ {
57
+ "row": 1,
58
+ "column": 1,
59
+ "start": 3,
60
+ "end": 5,
61
+ "replacement": "Zajęcia biblioteczne",
62
+ "category": "text.encoding",
63
+ "perplexity": 1.0003071924485039,
64
+ "status": "applied",
65
+ "route": "model"
66
+ }
67
+ ]
68
+ }
69
+ ],
70
+ "hard_negative": false
71
+ }
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/review.html ADDED
@@ -0,0 +1,18 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>56</strong><span>Cells scanned</span></div><div class="card"><strong>1</strong><span>Errors in input</span></div><div class="card"><strong>1</strong><span>Errors repaired</span></div><div class="card"><strong>0</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.39 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr><tr><td>2</td><td>Pozycja</td><td>&quot;Zajęcia biblioteczne&quot;</td><td>&quot;Zajęcia biblioteczne&quot;</td><td>&quot;Zajęcia biblioteczne&quot;</td><td class="good">Repaired<small>text.encoding</small></td></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>Identyfikator</th><th>Pozycja</th><th>Zarezerwowane</th><th>Obecni</th><th>Nieobecni</th><th>Data rozpoczęcia</th><th>Uwagi</th></tr><tr><td>1</td><td class="">000001</td><td class="">Warsztaty ceramiki</td><td class="">139</td><td class="">26</td><td class="">113</td><td class="">2026-01-31</td><td class="">Oczekuje na ponowną kontrolę</td></tr><tr><td>2</td><td class="">000002</td><td class="changed">Zajęcia biblioteczne</td><td class="">104</td><td class="">19</td><td class="">85</td><td class="">2026-07-06</td><td class="">Skorzystać z wejścia północnego
5
+ Najpierw zgłosić się do recepcji</td></tr><tr><td>3</td><td class="">000003</td><td class="">Zajęcia biblioteczne</td><td class="">220</td><td class="">100</td><td class="">120</td><td class="">2025-08-06</td><td class="">Etykieta „priorytet”, pozostawić widoczną</td></tr><tr><td>4</td><td class="">000004</td><td class="">Kurs językowy</td><td class="">33</td><td class="">8</td><td class="">25</td><td class="">2024-05-18</td><td class="">Skorzystać z wejścia północnego
6
+ Najpierw zgłosić się do recepcji</td></tr><tr><td>5</td><td class="">000007</td><td class="">Zajęcia biblioteczne</td><td class="">178</td><td class="">175</td><td class="">3</td><td class="">2024-09-14</td><td class=""></td></tr><tr><td>6</td><td class="">000008</td><td class="">Warsztaty ceramiki</td><td class="">65</td><td class="">38</td><td class="">27</td><td class="">2024-06-30</td><td class="">Sprawdzono przy odbiorze</td></tr><tr><td>7</td><td class="">000018</td><td class="">Kurs językowy</td><td class="">176</td><td class="">58</td><td class="">118</td><td class="">2026-01-05</td><td class=""></td></tr><tr><td>8</td><td class="">000026</td><td class="">Kurs językowy</td><td class="">193</td><td class="">92</td><td class="">101</td><td class="">2025-01-12</td><td class="">Zachowano odniesienie: UTF-8 / v2.1</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[
7
+ {
8
+ &quot;row&quot;: 1,
9
+ &quot;column&quot;: 1,
10
+ &quot;start&quot;: 3,
11
+ &quot;end&quot;: 5,
12
+ &quot;replacement&quot;: &quot;Zajęcia biblioteczne&quot;,
13
+ &quot;category&quot;: &quot;text.encoding&quot;,
14
+ &quot;perplexity&quot;: 1.0003071924485039,
15
+ &quot;status&quot;: &quot;applied&quot;,
16
+ &quot;route&quot;: &quot;model&quot;
17
+ }
18
+ ]</pre></details>
evaluation/v2/cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/schema.xml ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <schema>
2
+ <locales encoding="json">[{"currency":null,"date_format":"YYYY-MM-DD","decimal_separator":",","language":"pl"}]</locales>
3
+ <column index="0" name="Identyfikator" type="identifier" language="pl" nullable="forbidden" key="id">
4
+ <accepted_formats encoding="json">["six_digits_leading_zeros"]</accepted_formats>
5
+ <constraints encoding="json">{"pattern":"\\d{6}","unique":true}</constraints>
6
+ <null_markers encoding="json">[]</null_markers>
7
+ </column>
8
+ <column index="1" name="Pozycja" type="text" language="pl" nullable="forbidden" key="item">
9
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
10
+ <constraints encoding="json">{}</constraints>
11
+ <null_markers encoding="json">[]</null_markers>
12
+ </column>
13
+ <column index="2" name="Zarezerwowane" type="integer" language="pl" nullable="forbidden" key="booked">
14
+ <accepted_formats encoding="json">["base10_integer"]</accepted_formats>
15
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
16
+ <null_markers encoding="json">[]</null_markers>
17
+ </column>
18
+ <column index="3" name="Obecni" type="integer" language="pl" nullable="forbidden" key="attendance">
19
+ <accepted_formats encoding="json">["base10_integer"]</accepted_formats>
20
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
21
+ <null_markers encoding="json">[]</null_markers>
22
+ </column>
23
+ <column index="4" name="Nieobecni" type="integer" language="pl" nullable="forbidden" key="absent">
24
+ <accepted_formats encoding="json">["base10_integer"]</accepted_formats>
25
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
26
+ <null_markers encoding="json">[]</null_markers>
27
+ </column>
28
+ <column index="5" name="Data rozpoczęcia" type="date" language="pl" nullable="forbidden" key="start">
29
+ <accepted_formats encoding="json">["YYYY-MM-DD"]</accepted_formats>
30
+ <constraints encoding="json">{}</constraints>
31
+ <null_markers encoding="json">[]</null_markers>
32
+ </column>
33
+ <column index="6" name="Uwagi" type="text" language="pl" nullable="allowed" key="note">
34
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
35
+ <constraints encoding="json">{}</constraints>
36
+ <null_markers encoding="json">[]</null_markers>
37
+ </column>
38
+ <relations encoding="json">[["difference","booked","attendance","absent"]]</relations>
39
+ <functional_dependencies encoding="json">{}</functional_dependencies>
40
+ </schema>
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/input.csv ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Datensatz-ID,Artikel,Größe (MiB),Kopien,Gespeicherte Größe (MiB),IP-Adresse,Erfolgreich,Anmerkung
2
+ 000001,Testserver,"74,48",5,"372,40",192.0.2.1,Ja,
3
+ 000002,Testserver,"6,11",4,"24,44",192.0.2.2,Ja,"Etikett mit „Priorität“, sichtbar lassen"
4
+ 000003,Archivknoten,"65,44",4,"261,76",192.0.2.3,Nein,Bei Ankunft geprüft
5
+ 000004,Archivknoten,"62,29",3,"186,87",192.0.2.4,Ja,Keine zusätzliche Bearbeitung erforderlich
6
+ 000005,Sensor-Gateway,"82,35",3,"247,05",192.0.2.5,Nein,Referenz beibehalten: UTF-8 / v2.1
7
+ 000006,Testserver,"75,02",4,"300,08",192.0.2.6,Nein,Referenz beibehalten: UTF-8 / v2.1
8
+ 000007,Archivknoten,"85,32",2,"170,64",192.0.2.7,Nein,"Etikett mit „Priorität“, sichtbar lassen"
9
+ 000020,Archivknoten,"64,83",5,"324,15",192.0.2.20,Nein,Referenz beibehalten: UTF-8 / v2.1
10
+ 000021,Sensor-Gateway,"70,34",3,"211,02",192.0.2.21,Nein,
11
+ 000028,Sensor-Gateway,"71,34",3,"214,02",192.0.2.28,Ja,Bei Ankunft geprüft
12
+ 000030,Testserver,"33,39",4,"133,56",192.0.2.30,Ja,Bei Ankunft geprüft
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/audit.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metrics": {
3
+ "device": "cuda",
4
+ "rows": 11,
5
+ "cells": 88,
6
+ "threshold": 0.9,
7
+ "load_seconds": 0.0,
8
+ "inference_seconds": 0.2,
9
+ "cells_per_second": 447.67,
10
+ "peak_gpu_bytes": 1583256064,
11
+ "detected_spans": 0,
12
+ "changed_cells": 0
13
+ },
14
+ "proposals": []
15
+ }
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/corrected.csv ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Datensatz-ID,Artikel,Größe (MiB),Kopien,Gespeicherte Größe (MiB),IP-Adresse,Erfolgreich,Anmerkung
2
+ 000001,Testserver,"74,48",5,"372,40",192.0.2.1,Ja,
3
+ 000002,Testserver,"6,11",4,"24,44",192.0.2.2,Ja,"Etikett mit „Priorität“, sichtbar lassen"
4
+ 000003,Archivknoten,"65,44",4,"261,76",192.0.2.3,Nein,Bei Ankunft geprüft
5
+ 000004,Archivknoten,"62,29",3,"186,87",192.0.2.4,Ja,Keine zusätzliche Bearbeitung erforderlich
6
+ 000005,Sensor-Gateway,"82,35",3,"247,05",192.0.2.5,Nein,Referenz beibehalten: UTF-8 / v2.1
7
+ 000006,Testserver,"75,02",4,"300,08",192.0.2.6,Nein,Referenz beibehalten: UTF-8 / v2.1
8
+ 000007,Archivknoten,"85,32",2,"170,64",192.0.2.7,Nein,"Etikett mit „Priorität“, sichtbar lassen"
9
+ 000020,Archivknoten,"64,83",5,"324,15",192.0.2.20,Nein,Referenz beibehalten: UTF-8 / v2.1
10
+ 000021,Sensor-Gateway,"70,34",3,"211,02",192.0.2.21,Nein,
11
+ 000028,Sensor-Gateway,"71,34",3,"214,02",192.0.2.28,Ja,Bei Ankunft geprüft
12
+ 000030,Testserver,"33,39",4,"133,56",192.0.2.30,Ja,Bei Ankunft geprüft
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/output/review.html ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>88</strong><span>Cells scanned</span></div><div class="card"><strong>—</strong><span>Errors in input</span></div><div class="card"><strong>—</strong><span>Errors repaired</span></div><div class="card"><strong>—</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.2 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>Datensatz-ID</th><th>Artikel</th><th>Größe (MiB)</th><th>Kopien</th><th>Gespeicherte Größe (MiB)</th><th>IP-Adresse</th><th>Erfolgreich</th><th>Anmerkung</th></tr><tr><td>1</td><td class="">000001</td><td class="">Testserver</td><td class="">74,48</td><td class="">5</td><td class="">372,40</td><td class="">192.0.2.1</td><td class="">Ja</td><td class=""></td></tr><tr><td>2</td><td class="">000002</td><td class="">Testserver</td><td class="">6,11</td><td class="">4</td><td class="">24,44</td><td class="">192.0.2.2</td><td class="">Ja</td><td class="">Etikett mit „Priorität“, sichtbar lassen</td></tr><tr><td>3</td><td class="">000003</td><td class="">Archivknoten</td><td class="">65,44</td><td class="">4</td><td class="">261,76</td><td class="">192.0.2.3</td><td class="">Nein</td><td class="">Bei Ankunft geprüft</td></tr><tr><td>4</td><td class="">000004</td><td class="">Archivknoten</td><td class="">62,29</td><td class="">3</td><td class="">186,87</td><td class="">192.0.2.4</td><td class="">Ja</td><td class="">Keine zusätzliche Bearbeitung erforderlich</td></tr><tr><td>5</td><td class="">000005</td><td class="">Sensor-Gateway</td><td class="">82,35</td><td class="">3</td><td class="">247,05</td><td class="">192.0.2.5</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>6</td><td class="">000006</td><td class="">Testserver</td><td class="">75,02</td><td class="">4</td><td class="">300,08</td><td class="">192.0.2.6</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>7</td><td class="">000007</td><td class="">Archivknoten</td><td class="">85,32</td><td class="">2</td><td class="">170,64</td><td class="">192.0.2.7</td><td class="">Nein</td><td class="">Etikett mit „Priorität“, sichtbar lassen</td></tr><tr><td>8</td><td class="">000020</td><td class="">Archivknoten</td><td class="">64,83</td><td class="">5</td><td class="">324,15</td><td class="">192.0.2.20</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>9</td><td class="">000021</td><td class="">Sensor-Gateway</td><td class="">70,34</td><td class="">3</td><td class="">211,02</td><td class="">192.0.2.21</td><td class="">Nein</td><td class=""></td></tr><tr><td>10</td><td class="">000028</td><td class="">Sensor-Gateway</td><td class="">71,34</td><td class="">3</td><td class="">214,02</td><td class="">192.0.2.28</td><td class="">Ja</td><td class="">Bei Ankunft geprüft</td></tr><tr><td>11</td><td class="">000030</td><td class="">Testserver</td><td class="">33,39</td><td class="">4</td><td class="">133,56</td><td class="">192.0.2.30</td><td class="">Ja</td><td class="">Bei Ankunft geprüft</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[]</pre></details>
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/reference.csv ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Datensatz-ID,Artikel,Größe (MiB),Kopien,Gespeicherte Größe (MiB),IP-Adresse,Erfolgreich,Anmerkung
2
+ 000001,Testserver,"74,48",5,"372,40",192.0.2.1,Ja,
3
+ 000002,Testserver,"6,11",4,"24,44",192.0.2.2,Ja,"Etikett mit „Priorität“, sichtbar lassen"
4
+ 000003,Archivknoten,"65,44",4,"261,76",192.0.2.3,Nein,Bei Ankunft geprüft
5
+ 000004,Archivknoten,"62,29",3,"186,87",192.0.2.4,Ja,Keine zusätzliche Bearbeitung erforderlich
6
+ 000005,Sensor-Gateway,"82,35",3,"247,05",192.0.2.5,Nein,Referenz beibehalten: UTF-8 / v2.1
7
+ 000006,Testserver,"75,02",4,"300,08",192.0.2.6,Nein,Referenz beibehalten: UTF-8 / v2.1
8
+ 000007,Archivknoten,"85,32",2,"170,64",192.0.2.7,Nein,"Etikett mit „Priorität“, sichtbar lassen"
9
+ 000020,Archivknoten,"64,83",5,"324,15",192.0.2.20,Nein,Referenz beibehalten: UTF-8 / v2.1
10
+ 000021,Sensor-Gateway,"70,34",3,"211,02",192.0.2.21,Nein,
11
+ 000028,Sensor-Gateway,"71,34",3,"214,02",192.0.2.28,Ja,Bei Ankunft geprüft
12
+ 000030,Testserver,"33,39",4,"133,56",192.0.2.30,Ja,Bei Ankunft geprüft
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/result.json ADDED
@@ -0,0 +1,39 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "id": "971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1",
3
+ "language": "de",
4
+ "roles": [
5
+ "clean"
6
+ ],
7
+ "counts": {
8
+ "cells": 88,
9
+ "valid_cells": 88,
10
+ "exactly_repaired_cells": 0,
11
+ "valid_cells_harmed": 0,
12
+ "detected_faulty_cells": 0,
13
+ "flagged_valid_cells": 0
14
+ },
15
+ "categories": {
16
+ "text.encoding": {},
17
+ "text.invisible": {},
18
+ "text.spelling": {},
19
+ "format.number": {},
20
+ "format.date": {},
21
+ "format.boolean": {},
22
+ "format.case": {},
23
+ "format.whitespace": {},
24
+ "format.unit": {},
25
+ "schema.type": {},
26
+ "schema.identifier": {},
27
+ "schema.category": {},
28
+ "consistency.unit": {},
29
+ "consistency.dependency": {},
30
+ "consistency.temporal": {},
31
+ "missing.required": {},
32
+ "missing.conditional": {},
33
+ "structure.shift": {}
34
+ },
35
+ "decisions": {},
36
+ "inference_seconds": 0.2,
37
+ "details": [],
38
+ "hard_negative": false
39
+ }
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/review.html ADDED
@@ -0,0 +1,4 @@
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>88</strong><span>Cells scanned</span></div><div class="card"><strong>0</strong><span>Errors in input</span></div><div class="card"><strong>0</strong><span>Errors repaired</span></div><div class="card"><strong>0</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.2 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>Datensatz-ID</th><th>Artikel</th><th>Größe (MiB)</th><th>Kopien</th><th>Gespeicherte Größe (MiB)</th><th>IP-Adresse</th><th>Erfolgreich</th><th>Anmerkung</th></tr><tr><td>1</td><td class="">000001</td><td class="">Testserver</td><td class="">74,48</td><td class="">5</td><td class="">372,40</td><td class="">192.0.2.1</td><td class="">Ja</td><td class=""></td></tr><tr><td>2</td><td class="">000002</td><td class="">Testserver</td><td class="">6,11</td><td class="">4</td><td class="">24,44</td><td class="">192.0.2.2</td><td class="">Ja</td><td class="">Etikett mit „Priorität“, sichtbar lassen</td></tr><tr><td>3</td><td class="">000003</td><td class="">Archivknoten</td><td class="">65,44</td><td class="">4</td><td class="">261,76</td><td class="">192.0.2.3</td><td class="">Nein</td><td class="">Bei Ankunft geprüft</td></tr><tr><td>4</td><td class="">000004</td><td class="">Archivknoten</td><td class="">62,29</td><td class="">3</td><td class="">186,87</td><td class="">192.0.2.4</td><td class="">Ja</td><td class="">Keine zusätzliche Bearbeitung erforderlich</td></tr><tr><td>5</td><td class="">000005</td><td class="">Sensor-Gateway</td><td class="">82,35</td><td class="">3</td><td class="">247,05</td><td class="">192.0.2.5</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>6</td><td class="">000006</td><td class="">Testserver</td><td class="">75,02</td><td class="">4</td><td class="">300,08</td><td class="">192.0.2.6</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>7</td><td class="">000007</td><td class="">Archivknoten</td><td class="">85,32</td><td class="">2</td><td class="">170,64</td><td class="">192.0.2.7</td><td class="">Nein</td><td class="">Etikett mit „Priorität“, sichtbar lassen</td></tr><tr><td>8</td><td class="">000020</td><td class="">Archivknoten</td><td class="">64,83</td><td class="">5</td><td class="">324,15</td><td class="">192.0.2.20</td><td class="">Nein</td><td class="">Referenz beibehalten: UTF-8 / v2.1</td></tr><tr><td>9</td><td class="">000021</td><td class="">Sensor-Gateway</td><td class="">70,34</td><td class="">3</td><td class="">211,02</td><td class="">192.0.2.21</td><td class="">Nein</td><td class=""></td></tr><tr><td>10</td><td class="">000028</td><td class="">Sensor-Gateway</td><td class="">71,34</td><td class="">3</td><td class="">214,02</td><td class="">192.0.2.28</td><td class="">Ja</td><td class="">Bei Ankunft geprüft</td></tr><tr><td>11</td><td class="">000030</td><td class="">Testserver</td><td class="">33,39</td><td class="">4</td><td class="">133,56</td><td class="">192.0.2.30</td><td class="">Ja</td><td class="">Bei Ankunft geprüft</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[]</pre></details>
evaluation/v2/cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/schema.xml ADDED
@@ -0,0 +1,47 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <schema>
2
+ <locales encoding="json">[{"currency":null,"date_format":"YYYY-MM-DD","decimal_separator":",","language":"de"}]</locales>
3
+ <column index="0" name="Datensatz-ID" type="identifier" language="de" nullable="forbidden" key="id">
4
+ <accepted_formats encoding="json">["six_digits_leading_zeros"]</accepted_formats>
5
+ <constraints encoding="json">{"pattern":"\\d{6}","unique":true}</constraints>
6
+ <null_markers encoding="json">[]</null_markers>
7
+ </column>
8
+ <column index="1" name="Artikel" type="category" language="de" nullable="forbidden" key="item">
9
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
10
+ <constraints encoding="json">{"enum":["Testserver","Archivknoten","Sensor-Gateway","Sicherungsserver"]}</constraints>
11
+ <null_markers encoding="json">[]</null_markers>
12
+ </column>
13
+ <column index="2" name="Größe (MiB)" type="decimal" language="de" nullable="forbidden" key="size">
14
+ <accepted_formats encoding="json">["decimal_comma_2dp"]</accepted_formats>
15
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
16
+ <unit encoding="json">"MiB"</unit>
17
+ <null_markers encoding="json">[]</null_markers>
18
+ </column>
19
+ <column index="3" name="Kopien" type="integer" language="de" nullable="forbidden" key="copies">
20
+ <accepted_formats encoding="json">["base10_integer"]</accepted_formats>
21
+ <constraints encoding="json">{"maximum":12,"minimum":1}</constraints>
22
+ <null_markers encoding="json">[]</null_markers>
23
+ </column>
24
+ <column index="4" name="Gespeicherte Größe (MiB)" type="decimal" language="de" nullable="forbidden" key="stored">
25
+ <accepted_formats encoding="json">["decimal_comma_2dp"]</accepted_formats>
26
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
27
+ <unit encoding="json">"MiB"</unit>
28
+ <null_markers encoding="json">[]</null_markers>
29
+ </column>
30
+ <column index="5" name="IP-Adresse" type="ipv4" language="de" nullable="forbidden" key="ip">
31
+ <accepted_formats encoding="json">["IPv4"]</accepted_formats>
32
+ <constraints encoding="json">{}</constraints>
33
+ <null_markers encoding="json">[]</null_markers>
34
+ </column>
35
+ <column index="6" name="Erfolgreich" type="boolean" language="de" nullable="forbidden" key="success">
36
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
37
+ <constraints encoding="json">{"enum":["Nein","Ja"]}</constraints>
38
+ <null_markers encoding="json">[]</null_markers>
39
+ </column>
40
+ <column index="7" name="Anmerkung" type="text" language="de" nullable="allowed" key="note">
41
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
42
+ <constraints encoding="json">{}</constraints>
43
+ <null_markers encoding="json">[]</null_markers>
44
+ </column>
45
+ <relations encoding="json">[["multiply","size","copies","stored"]]</relations>
46
+ <functional_dependencies encoding="json">{}</functional_dependencies>
47
+ </schema>
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/input.csv ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ レコード番号,品目,投入質量 (kg),換算係数 (g/kg),換算値 (g),単位,備考
2
+ 000001,アルミニウム,26.81,1000.00,26810.00,kg,到着時に確認済み
3
+ 000002,紙,92.29,1000.00,92290.00,kg,到着時に確認済み
4
+ 000003,段ボール,92.91,1000.00,92910.00,kg,参照情報を保持:UTF-8 / v2.1
5
+ 000011,紙,62.01,1000.00,62010.00,kg,
6
+ 000013,紙,19.88,1000.00,19880.00,kg,ラベルは「優先」、見える位置に置く
7
+ 000014,紙,95.22,1000.00,95220.00,kg,
8
+ 000015,アルミニウム,26.28,1000.00,26280.00,kg?,
9
+ 000016,段ボール,1.36,1000.00,1360.00,kg,"北側の入口を利用
10
+ 先に受付へ連絡"
11
+ 000017,紙,63.83,1000.00,63830.00,kg,
12
+ 000025,段ボール,50.98,1000.00,50980.00,kg,再確認待ち
13
+ 000031,紙,0.00,1000.00,0.00,kg,再確認待ち
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/audit.json ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metrics": {
3
+ "device": "cuda",
4
+ "rows": 11,
5
+ "cells": 77,
6
+ "threshold": 0.9,
7
+ "load_seconds": 0.0,
8
+ "inference_seconds": 0.8,
9
+ "cells_per_second": 96.29,
10
+ "peak_gpu_bytes": 2013979136,
11
+ "detected_spans": 2,
12
+ "changed_cells": 2
13
+ },
14
+ "proposals": [
15
+ {
16
+ "row": 6,
17
+ "column": 1,
18
+ "start": 0,
19
+ "end": 18,
20
+ "replacement": "段ボール",
21
+ "category": "text.encoding",
22
+ "perplexity": 1.0002686006337904,
23
+ "status": "applied",
24
+ "route": "model"
25
+ },
26
+ {
27
+ "row": 6,
28
+ "column": 5,
29
+ "start": 0,
30
+ "end": 3,
31
+ "replacement": "kg",
32
+ "category": "schema.category",
33
+ "perplexity": 1.0000396367096813,
34
+ "status": "applied",
35
+ "route": "model"
36
+ }
37
+ ]
38
+ }
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/corrected.csv ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ レコード番号,品目,投入質量 (kg),換算係数 (g/kg),換算値 (g),単位,備考
2
+ 000001,アルミニウム,26.81,1000.00,26810.00,kg,到着時に確認済み
3
+ 000002,紙,92.29,1000.00,92290.00,kg,到着時に確認済み
4
+ 000003,段ボール,92.91,1000.00,92910.00,kg,参照情報を保持:UTF-8 / v2.1
5
+ 000011,紙,62.01,1000.00,62010.00,kg,
6
+ 000013,紙,19.88,1000.00,19880.00,kg,ラベルは「優先」、見える位置に置く
7
+ 000014,紙,95.22,1000.00,95220.00,kg,
8
+ 000015,段ボール,26.28,1000.00,26280.00,kg,
9
+ 000016,段ボール,1.36,1000.00,1360.00,kg,"北側の入口を利用
10
+ 先に受付へ連絡"
11
+ 000017,紙,63.83,1000.00,63830.00,kg,
12
+ 000025,段ボール,50.98,1000.00,50980.00,kg,再確認待ち
13
+ 000031,紙,0.00,1000.00,0.00,kg,再確認待ち
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/output/review.html ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>77</strong><span>Cells scanned</span></div><div class="card"><strong>—</strong><span>Errors in input</span></div><div class="card"><strong>—</strong><span>Errors repaired</span></div><div class="card"><strong>—</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.8 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr><tr><td>7</td><td>品目</td><td>&quot;アルミニウム&quot;</td><td>&quot;段ボール&quot;</td><td>—</td><td class="bad">Changed<small>text.encoding</small></td></tr><tr><td>7</td><td>単位</td><td>&quot;kg?&quot;</td><td>&quot;kg&quot;</td><td>—</td><td class="bad">Changed<small>schema.category</small></td></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>レコード番号</th><th>品目</th><th>投入質量 (kg)</th><th>換算係数 (g/kg)</th><th>換算値 (g)</th><th>単位</th><th>備考</th></tr><tr><td>1</td><td class="">000001</td><td class="">アルミニウム</td><td class="">26.81</td><td class="">1000.00</td><td class="">26810.00</td><td class="">kg</td><td class="">到着時に確認済み</td></tr><tr><td>2</td><td class="">000002</td><td class="">紙</td><td class="">92.29</td><td class="">1000.00</td><td class="">92290.00</td><td class="">kg</td><td class="">到着時に確認済み</td></tr><tr><td>3</td><td class="">000003</td><td class="">段ボール</td><td class="">92.91</td><td class="">1000.00</td><td class="">92910.00</td><td class="">kg</td><td class="">参照情報を保持:UTF-8 / v2.1</td></tr><tr><td>4</td><td class="">000011</td><td class="">紙</td><td class="">62.01</td><td class="">1000.00</td><td class="">62010.00</td><td class="">kg</td><td class=""></td></tr><tr><td>5</td><td class="">000013</td><td class="">紙</td><td class="">19.88</td><td class="">1000.00</td><td class="">19880.00</td><td class="">kg</td><td class="">ラベルは「優先」、見える位置に置く</td></tr><tr><td>6</td><td class="">000014</td><td class="">紙</td><td class="">95.22</td><td class="">1000.00</td><td class="">95220.00</td><td class="">kg</td><td class=""></td></tr><tr><td>7</td><td class="">000015</td><td class="changed">段ボール</td><td class="">26.28</td><td class="">1000.00</td><td class="">26280.00</td><td class="changed">kg</td><td class=""></td></tr><tr><td>8</td><td class="">000016</td><td class="">段ボール</td><td class="">1.36</td><td class="">1000.00</td><td class="">1360.00</td><td class="">kg</td><td class="">北側の入口を利用
5
+ 先に受付へ連絡</td></tr><tr><td>9</td><td class="">000017</td><td class="">紙</td><td class="">63.83</td><td class="">1000.00</td><td class="">63830.00</td><td class="">kg</td><td class=""></td></tr><tr><td>10</td><td class="">000025</td><td class="">段ボール</td><td class="">50.98</td><td class="">1000.00</td><td class="">50980.00</td><td class="">kg</td><td class="">再確認待ち</td></tr><tr><td>11</td><td class="">000031</td><td class="">紙</td><td class="">0.00</td><td class="">1000.00</td><td class="">0.00</td><td class="">kg</td><td class="">再確認待ち</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[
6
+ {
7
+ &quot;row&quot;: 6,
8
+ &quot;column&quot;: 1,
9
+ &quot;start&quot;: 0,
10
+ &quot;end&quot;: 18,
11
+ &quot;replacement&quot;: &quot;段ボール&quot;,
12
+ &quot;category&quot;: &quot;text.encoding&quot;,
13
+ &quot;perplexity&quot;: 1.0002686006337904,
14
+ &quot;status&quot;: &quot;applied&quot;,
15
+ &quot;route&quot;: &quot;model&quot;
16
+ },
17
+ {
18
+ &quot;row&quot;: 6,
19
+ &quot;column&quot;: 5,
20
+ &quot;start&quot;: 0,
21
+ &quot;end&quot;: 3,
22
+ &quot;replacement&quot;: &quot;kg&quot;,
23
+ &quot;category&quot;: &quot;schema.category&quot;,
24
+ &quot;perplexity&quot;: 1.0000396367096813,
25
+ &quot;status&quot;: &quot;applied&quot;,
26
+ &quot;route&quot;: &quot;model&quot;
27
+ }
28
+ ]</pre></details>
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/reference.csv ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ レコード番号,品目,投入質量 (kg),換算係数 (g/kg),換算値 (g),単位,備考
2
+ 000001,アルミニウム,26.81,1000.00,26810.00,kg,到着時に確認済み
3
+ 000002,紙,92.29,1000.00,92290.00,kg,到着時に確認済み
4
+ 000003,段ボール,92.91,1000.00,92910.00,kg,参照情報を保持:UTF-8 / v2.1
5
+ 000011,紙,62.01,1000.00,62010.00,kg,
6
+ 000013,紙,19.88,1000.00,19880.00,kg,ラベルは「優先」、見える位置に置く
7
+ 000014,紙,95.22,1000.00,95220.00,kg,
8
+ 000015,アルミニウム,26.28,1000.00,26280.00,kg,
9
+ 000016,段ボール,1.36,1000.00,1360.00,kg,"北側の入口を利用
10
+ 先に受付へ連絡"
11
+ 000017,紙,63.83,1000.00,63830.00,kg,
12
+ 000025,段ボール,50.98,1000.00,50980.00,kg,再確認待ち
13
+ 000031,紙,0.00,1000.00,0.00,kg,再確認待ち
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/result.json ADDED
@@ -0,0 +1,99 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "id": "e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open",
3
+ "language": "ja",
4
+ "roles": [
5
+ "open_positive:text.encoding"
6
+ ],
7
+ "counts": {
8
+ "cells": 77,
9
+ "valid_cells": 75,
10
+ "exactly_repaired_cells": 1,
11
+ "valid_cells_harmed": 0,
12
+ "detected_faulty_cells": 2,
13
+ "flagged_valid_cells": 0,
14
+ "faulty_cells": 2
15
+ },
16
+ "categories": {
17
+ "text.encoding": {
18
+ "gold_cells": 1,
19
+ "detected_cells": 1,
20
+ "exactly_repaired_cells": 0
21
+ },
22
+ "text.invisible": {},
23
+ "text.spelling": {},
24
+ "format.number": {},
25
+ "format.date": {},
26
+ "format.boolean": {},
27
+ "format.case": {},
28
+ "format.whitespace": {},
29
+ "format.unit": {},
30
+ "schema.type": {},
31
+ "schema.identifier": {},
32
+ "schema.category": {
33
+ "gold_cells": 1,
34
+ "detected_cells": 1,
35
+ "exactly_repaired_cells": 1
36
+ },
37
+ "consistency.unit": {},
38
+ "consistency.dependency": {},
39
+ "consistency.temporal": {},
40
+ "missing.required": {},
41
+ "missing.conditional": {},
42
+ "structure.shift": {}
43
+ },
44
+ "decisions": {
45
+ "applied": 2
46
+ },
47
+ "inference_seconds": 0.8,
48
+ "details": [
49
+ {
50
+ "row": 6,
51
+ "column": "品目",
52
+ "original": "アルミニウム",
53
+ "prediction": "段ボール",
54
+ "reference": "アルミニウム",
55
+ "categories": [
56
+ "text.encoding"
57
+ ],
58
+ "status": "remaining",
59
+ "decisions": [
60
+ {
61
+ "row": 6,
62
+ "column": 1,
63
+ "start": 0,
64
+ "end": 18,
65
+ "replacement": "段ボール",
66
+ "category": "text.encoding",
67
+ "perplexity": 1.0002686006337904,
68
+ "status": "applied",
69
+ "route": "model"
70
+ }
71
+ ]
72
+ },
73
+ {
74
+ "row": 6,
75
+ "column": "単位",
76
+ "original": "kg?",
77
+ "prediction": "kg",
78
+ "reference": "kg",
79
+ "categories": [
80
+ "schema.category"
81
+ ],
82
+ "status": "repaired",
83
+ "decisions": [
84
+ {
85
+ "row": 6,
86
+ "column": 5,
87
+ "start": 0,
88
+ "end": 3,
89
+ "replacement": "kg",
90
+ "category": "schema.category",
91
+ "perplexity": 1.0000396367096813,
92
+ "status": "applied",
93
+ "route": "model"
94
+ }
95
+ ]
96
+ }
97
+ ],
98
+ "hard_negative": false
99
+ }
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/review.html ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix — end-to-end CSV review</title><style>
2
+ .cards{display:flex;gap:16px;flex-wrap:wrap}.card{background:white;padding:18px 24px;border-radius:12px;min-width:130px}.card strong{display:block;font-size:32px}.card span{color:#536175}body{font:16px system-ui;background:#f3f5f8;color:#172238;margin:36px}h1{font-size:30px}p{max-width:950px;line-height:1.6}table{border-collapse:collapse;width:100%;background:white;margin:20px 0}td,th{padding:12px;text-align:left;border-bottom:1px solid #dbe1e8;white-space:pre-wrap;overflow-wrap:anywhere}th{background:#e6ebf2}small{display:block;color:#536175;margin-top:6px}.good{color:#15704b}.bad{color:#b13a37}table.full{min-width:1050px}.changed{background:#fff0c7}.scroll{overflow:auto}pre{white-space:pre-wrap;background:white;padding:20px}details{margin-top:24px}</style>
3
+ <h1>Does TabFix fix this CSV?</h1><p>Actual CSV → schema and context → predicted error locations → generated replacements → corrected CSV. The clean reference is used only after inference. No gold error locations or replacement text are supplied to the model. Row numbers below count data rows, excluding the header.</p>
4
+ <div class="cards"><div class="card"><strong>77</strong><span>Cells scanned</span></div><div class="card"><strong>2</strong><span>Errors in input</span></div><div class="card"><strong>1</strong><span>Errors repaired</span></div><div class="card"><strong>0</strong><span>Valid cells changed</span></div></div><p>CSV inference review · Threshold 0.9 · CPU inference 0.8 seconds (model loading excluded).</p><h2>Errors and proposed changes</h2><div class="scroll"><table><tr><th>Row</th><th>Column</th><th>Input</th><th>Model output</th><th>Reference</th><th>Result</th></tr><tr><td>7</td><td>品目</td><td>&quot;アルミニウム&quot;</td><td>&quot;段ボール&quot;</td><td>&quot;アルミニウム&quot;</td><td class="bad">Missed / incorrect<small>text.encoding</small></td></tr><tr><td>7</td><td>単位</td><td>&quot;kg?&quot;</td><td>&quot;kg&quot;</td><td>&quot;kg&quot;</td><td class="good">Repaired<small>schema.category</small></td></tr></table></div><p>Strings are quoted to make spaces and empty values visible. Yellow cells below were changed. All other cells are included so unwanted edits remain visible.</p><h2>Complete model output</h2><div class="scroll"><table class="full"><tr><th>Data row</th><th>レコード番号</th><th>品目</th><th>投入質量 (kg)</th><th>換算係数 (g/kg)</th><th>換算値 (g)</th><th>単位</th><th>備考</th></tr><tr><td>1</td><td class="">000001</td><td class="">アルミニウム</td><td class="">26.81</td><td class="">1000.00</td><td class="">26810.00</td><td class="">kg</td><td class="">到着時に確認済み</td></tr><tr><td>2</td><td class="">000002</td><td class="">紙</td><td class="">92.29</td><td class="">1000.00</td><td class="">92290.00</td><td class="">kg</td><td class="">到着時に確認済み</td></tr><tr><td>3</td><td class="">000003</td><td class="">段ボール</td><td class="">92.91</td><td class="">1000.00</td><td class="">92910.00</td><td class="">kg</td><td class="">参照情報を保持:UTF-8 / v2.1</td></tr><tr><td>4</td><td class="">000011</td><td class="">紙</td><td class="">62.01</td><td class="">1000.00</td><td class="">62010.00</td><td class="">kg</td><td class=""></td></tr><tr><td>5</td><td class="">000013</td><td class="">紙</td><td class="">19.88</td><td class="">1000.00</td><td class="">19880.00</td><td class="">kg</td><td class="">ラベルは「優先」、見える位置に置く</td></tr><tr><td>6</td><td class="">000014</td><td class="">紙</td><td class="">95.22</td><td class="">1000.00</td><td class="">95220.00</td><td class="">kg</td><td class=""></td></tr><tr><td>7</td><td class="">000015</td><td class="changed">段ボール</td><td class="">26.28</td><td class="">1000.00</td><td class="">26280.00</td><td class="changed">kg</td><td class=""></td></tr><tr><td>8</td><td class="">000016</td><td class="">段ボール</td><td class="">1.36</td><td class="">1000.00</td><td class="">1360.00</td><td class="">kg</td><td class="">北側の入口を利用
5
+ 先に受付へ連絡</td></tr><tr><td>9</td><td class="">000017</td><td class="">紙</td><td class="">63.83</td><td class="">1000.00</td><td class="">63830.00</td><td class="">kg</td><td class=""></td></tr><tr><td>10</td><td class="">000025</td><td class="">段ボール</td><td class="">50.98</td><td class="">1000.00</td><td class="">50980.00</td><td class="">kg</td><td class="">再確認待ち</td></tr><tr><td>11</td><td class="">000031</td><td class="">紙</td><td class="">0.00</td><td class="">1000.00</td><td class="">0.00</td><td class="">kg</td><td class="">再確認待ち</td></tr></table></div><details><summary>Every detection and application decision</summary><pre>[
6
+ {
7
+ &quot;row&quot;: 6,
8
+ &quot;column&quot;: 1,
9
+ &quot;start&quot;: 0,
10
+ &quot;end&quot;: 18,
11
+ &quot;replacement&quot;: &quot;段ボール&quot;,
12
+ &quot;category&quot;: &quot;text.encoding&quot;,
13
+ &quot;perplexity&quot;: 1.0002686006337904,
14
+ &quot;status&quot;: &quot;applied&quot;,
15
+ &quot;route&quot;: &quot;model&quot;
16
+ },
17
+ {
18
+ &quot;row&quot;: 6,
19
+ &quot;column&quot;: 5,
20
+ &quot;start&quot;: 0,
21
+ &quot;end&quot;: 3,
22
+ &quot;replacement&quot;: &quot;kg&quot;,
23
+ &quot;category&quot;: &quot;schema.category&quot;,
24
+ &quot;perplexity&quot;: 1.0000396367096813,
25
+ &quot;status&quot;: &quot;applied&quot;,
26
+ &quot;route&quot;: &quot;model&quot;
27
+ }
28
+ ]</pre></details>
evaluation/v2/cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/schema.xml ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <schema>
2
+ <locales encoding="json">[{"currency":null,"date_format":"YYYY-MM-DD","decimal_separator":".","language":"ja"}]</locales>
3
+ <column index="0" name="レコード番号" type="identifier" language="ja" nullable="forbidden" key="id">
4
+ <accepted_formats encoding="json">["six_digits_leading_zeros"]</accepted_formats>
5
+ <constraints encoding="json">{"pattern":"\\d{6}","unique":true}</constraints>
6
+ <null_markers encoding="json">[]</null_markers>
7
+ </column>
8
+ <column index="1" name="品目" type="text" language="ja" nullable="forbidden" key="item">
9
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
10
+ <constraints encoding="json">{}</constraints>
11
+ <null_markers encoding="json">[]</null_markers>
12
+ </column>
13
+ <column index="2" name="投入質量 (kg)" type="decimal" language="ja" nullable="forbidden" key="weight">
14
+ <accepted_formats encoding="json">["decimal_dot_2dp"]</accepted_formats>
15
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
16
+ <unit encoding="json">"kg"</unit>
17
+ <null_markers encoding="json">[]</null_markers>
18
+ </column>
19
+ <column index="3" name="換算係数 (g/kg)" type="decimal" language="ja" nullable="forbidden" key="factor">
20
+ <accepted_formats encoding="json">["decimal_dot_2dp"]</accepted_formats>
21
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
22
+ <unit encoding="json">"g/kg"</unit>
23
+ <null_markers encoding="json">[]</null_markers>
24
+ </column>
25
+ <column index="4" name="換算値 (g)" type="decimal" language="ja" nullable="forbidden" key="converted">
26
+ <accepted_formats encoding="json">["decimal_dot_2dp"]</accepted_formats>
27
+ <constraints encoding="json">{"maximum":100000000,"minimum":0}</constraints>
28
+ <unit encoding="json">"g"</unit>
29
+ <null_markers encoding="json">[]</null_markers>
30
+ </column>
31
+ <column index="5" name="単位" type="category" language="ja" nullable="forbidden" key="unit">
32
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
33
+ <constraints encoding="json">{"enum":["kg"]}</constraints>
34
+ <null_markers encoding="json">[]</null_markers>
35
+ </column>
36
+ <column index="6" name="備考" type="text" language="ja" nullable="allowed" key="note">
37
+ <accepted_formats encoding="json">["unicode_text"]</accepted_formats>
38
+ <constraints encoding="json">{}</constraints>
39
+ <null_markers encoding="json">[]</null_markers>
40
+ </column>
41
+ <relations encoding="json">[["multiply","weight","factor","converted"],["unit_factor","unit","factor"],"unit_factor(unit,factor): factor is grams per unit; kg=1000, g=1."]</relations>
42
+ <functional_dependencies encoding="json">{}</functional_dependencies>
43
+ </schema>
evaluation/v2/review.html CHANGED
@@ -1,4 +1,4 @@
1
- <!doctype html><meta charset="utf-8"><title>TabFix v2 — test de bout en bout</title><style>body{font:16px system-ui;max-width:1450px;margin:35px auto;padding:0 24px;background:#f6f7fa;color:#182333}table{border-collapse:collapse;width:100%;background:white;margin:25px 0}th,td{text-align:left;padding:10px;border-bottom:1px solid #ddd;vertical-align:top}td code{white-space:pre-wrap;overflow-wrap:anywhere}pre{white-space:pre-wrap;max-width:600px}.repaired{background:#eaf8ee}.harmed{background:#ffe3e3}.remaining{background:#fff4df}.cards{display:flex;gap:18px;flex-wrap:wrap}.card{background:white;padding:18px;border-radius:10px}.card b{font-size:28px;display:block}</style><h1>TabFix v2 — test de bout en bout</h1><p>50/76 tableaux. CSV → filtre déterministe/neural → correction → perplexité → validation des règles. Aucune position fautive fournie au modèle.</p><div class="cards"><div class="card"><b>30/45</b>Cellules fautives réparées</div><div class="card"><b>0/5718</b>Cellules valides endommagées</div><div class="card"><b>38/45</b>Cellules fautives localisées</div></div><p>Panel synthétique représentatif ; pas une précision globale sur tout le corpus. Les cellules de contexte sont incluses. Les résultats par catégorie peuvent se recouper pour les erreurs composées. Les contradictions de groupe sans localisation certaine restent signalées sans correction.</p><h2>Par catégorie</h2><table><tr><th>Catégorie réelle</th><th>Cellules fautives</th><th>Localisées</th><th>Réparées exactement</th></tr><tr><td>text.encoding</td><td>2</td><td>2</td><td>2</td></tr><tr><td>text.invisible</td><td>2</td><td>2</td><td>2</td></tr><tr><td>text.spelling</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.number</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.date</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.boolean</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.case</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.whitespace</td><td>3</td><td>3</td><td>3</td></tr><tr><td>format.unit</td><td>3</td><td>3</td><td>3</td></tr><tr><td>schema.type</td><td>2</td><td>2</td><td>2</td></tr><tr><td>schema.identifier</td><td>2</td><td>2</td><td>2</td></tr><tr><td>schema.category</td><td>7</td><td>7</td><td>7</td></tr><tr><td>consistency.unit</td><td>2</td><td>0</td><td>0</td></tr><tr><td>consistency.dependency</td><td>3</td><td>0</td><td>0</td></tr><tr><td>consistency.temporal</td><td>2</td><td>0</td><td>0</td></tr><tr><td>missing.required</td><td>2</td><td>2</td><td>0</td></tr><tr><td>missing.conditional</td><td>2</td><td>2</td><td>0</td></tr><tr><td>structure.shift</td><td>4</td><td>4</td><td>0</td></tr></table><h2>Corrections et erreurs restantes</h2><table><tr><th>Langue / tableau</th><th>Colonne</th><th>Entrée</th><th>Sortie</th><th>Référence</th><th>Résultat</th></tr><tr class="repaired"><td><a href="cases/3612fb29d718dff78200f14d51e1a860bf3146f0539ece31761207733bc2f8cf/review.html">ja</a></td><td>品目</td><td><code>&quot;麻のバッグ&quot;</code></td><td><code>&quot;麻のバッグ&quot;</code></td><td><code>&quot;麻のバッグ&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
2
  {
3
  &quot;row&quot;: 7,
4
  &quot;column&quot;: 1,
@@ -461,4 +461,100 @@
461
  &quot;status&quot;: &quot;abstained: output violates a declared rule&quot;,
462
  &quot;route&quot;: &quot;model&quot;
463
  }
464
- ]</pre></details></td></tr></table><details><summary>Tous les tableaux, y compris les contrôles valides</summary><ul><li><a href="cases/3612fb29d718dff78200f14d51e1a860bf3146f0539ece31761207733bc2f8cf/review.html">ja · positive:text.encoding</a></li><li><a href="cases/433d9c8af5be2277d006883eb58ce528d287c3fe0431becad819f8997484eee7/review.html">de · positive:text.encoding</a></li><li><a href="cases/11c0221cd3cfa0daa1e68d820026c0a2b76a932215a3e2bb43855ed9b0b468fe/review.html">fr · hard_negative:text.encoding</a></li><li><a href="cases/caf4c48e4fb55ac78afcbd23726eed87ebae8e0836c295731bb50c4f50e43e99/review.html">nl · positive:text.invisible</a></li><li><a href="cases/87e885905fc3606c63cd48369a6a570be6aa2198cf9927b667ceda60401e2e78/review.html">pl · positive:text.invisible</a></li><li><a href="cases/21ad4f83d9ffea1518a6a498f186f6a924179c96d4c8b33dc432806a37327c5e/review.html">pl · hard_negative:text.invisible</a></li><li><a href="cases/6a8776324da9c9e9197bb19fe34cbcd4c499c7da785654a238e1081314249599/review.html">de · positive:text.spelling</a></li><li><a href="cases/e73917739efb46d1ad052025dbbad8ce43535e320c3bfab88dbb2cf070020d9e/review.html">fr · positive:text.spelling, positive:format.date</a></li><li><a href="cases/b46e574a91a93f6642dccfa028c8f0220128094f176d00c06d7dce59b119a198/review.html">nl · hard_negative:text.spelling</a></li><li><a href="cases/a7aadbf7aa1ed87cf17c6d68b7ca3b2b9c98740f2cc7c5be91c987035c7a29a2/review.html">nl · positive:format.number</a></li><li><a href="cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/review.html">pl · positive:format.number</a></li><li><a href="cases/42ae0423efc79bfa2e7e92255c5e94c337c52b95167243cc3497785b15214174/review.html">de · hard_negative:format.number</a></li><li><a href="cases/415bbfbbc5a065d100a535b95d0d4f259fae060cefc532466d412eddfc7e03c7/review.html">pl · positive:format.date</a></li><li><a href="cases/95e9fdf26fc178038d17705c13b1e33edcdd0074c14db6fb67518bfe22725cab/review.html">pl · hard_negative:format.date</a></li><li><a href="cases/ec7e824cd5bf670371a04220237b3ceb64ce873cd260e06fc8e09549b10a5782/review.html">pl · positive:format.boolean</a></li><li><a href="cases/5036ba02f02df9baf2542b0114cf0000fd80e8a9958351cf1b50fca0c70ce42b/review.html">fr · positive:format.boolean</a></li><li><a href="cases/7264aa617c2f6a9abb0463459973e38ba98b12a4589ff77da50876b5fd353be0/review.html">de · hard_negative:format.boolean</a></li><li><a href="cases/299a18168469d268a9f9aca770f04d1410bc17c9bbb077b4eb95badec901239d/review.html">fr · positive:format.case, positive:schema.category</a></li><li><a href="cases/2ec9fe1f256063b86d958b20a9b1c1e811b3096630a38672b31e17811f9f9717/review.html">pl · positive:format.case</a></li><li><a href="cases/6eceda5dd8bccade58555a35ade25610195f33ba3b54df76c348edfd512c5ab7/review.html">en · hard_negative:format.case</a></li><li><a href="cases/5928e2790fc6276004825d19abd24c5230269307e8be597ce5cebc34de1553b2/review.html">de · positive:format.whitespace</a></li><li><a href="cases/85789b30d3fd26084a8927ad3a2195916b57c5fd8951d704b22be3da24bfa0d1/review.html">en · positive:format.whitespace</a></li><li><a href="cases/47256473f1211a06714df6ace7e0d856d17fed3bd4d71f4af522296043382d0e/review.html">ja · hard_negative:format.whitespace</a></li><li><a href="cases/2075d5f5a6a4ccf1843282f9d1a92f011d8682938be7346738c504d0923e3353/review.html">ja · positive:format.unit, positive:schema.type</a></li><li><a href="cases/83aafa579b76e84cf5cb7a44ab4e9724eb1a233cf5f07462bf27a673b495779f/review.html">pl · positive:format.unit, positive:schema.identifier</a></li><li><a href="cases/a5fb4ba42af578df38f103f0d7f7584caf5390d195915472fe340cf085da66fd/review.html">de · hard_negative:format.unit</a></li><li><a href="cases/275a9dadfc0c43c0dd1c1fa953a3daceac2481609dcc76a055704b7b0127fcae/review.html">fr · positive:schema.type</a></li><li><a href="cases/b3dfc51e8a64d2f8b206cd51b42b96b3250738fe276a9559eca738d1d3902adc/review.html">en · hard_negative:schema.type</a></li><li><a href="cases/b3bfd524d8c3ae833e623a5aa58c836c52f067846b5db5fbca44f3bf4ea71852/review.html">en · positive:schema.identifier, positive:consistency.temporal</a></li><li><a href="cases/a78eff0ea659f0315c65f5e3fe1ee1bcc69fd80907b537651b3602b6d9d1263f/review.html">fr · hard_negative:schema.identifier</a></li><li><a href="cases/da7b7de84b71cb8704e63df4a915e0e761a424a12c0f2a0309b5e060c6a96aba/review.html">el · positive:schema.category, positive:missing.conditional</a></li><li><a href="cases/83c6c6000a283cbefb1f55a550d45900dfe1edb8eda981ad8ce47bb9b513c3c7/review.html">pl · hard_negative:schema.category</a></li><li><a href="cases/9ed9eb83de2b77e08cc42088a8079d37463134a454cff1c8459015f380596ca8/review.html">ja · positive:consistency.unit</a></li><li><a href="cases/22e00e06d09d2bd02a565d0592ad8f4afe8169489bc73daa5d46d39cdade28ec/review.html">nl · positive:consistency.unit</a></li><li><a href="cases/9071e3bb96646641aad094c962441e56d5f01695b23420d60ba5e28fc3ac377b/review.html">pl · hard_negative:consistency.unit</a></li><li><a href="cases/f8d8a5333c6d6326bbec66ba279d61d1019dc5594dfbdc9dc83355d3668105c2/review.html">ja · positive:consistency.dependency</a></li><li><a href="cases/cb78edfa963adbebefd0856fd78d5fcb08875803a2bafcf52e94a1fb23200250/review.html">fr · positive:consistency.dependency</a></li><li><a href="cases/83e33197e71ea0876f8c972009b0caadedaf7ab29772d983a0bc40d6b31955e5/review.html">pl · hard_negative:consistency.dependency</a></li><li><a href="cases/cfea35d1a52131df05b5582234701504e0e3eb2c07686371b5c66cece911c175/review.html">el · positive:consistency.temporal</a></li><li><a href="cases/e929315e39db14f2cc01883f4f6bf657e5939c9036ffa7a611cb32d7b0e50d51/review.html">en · hard_negative:consistency.temporal</a></li><li><a href="cases/617c64c47791f492d292e1f15fb5ed3138a59debe8fbb96a601161abe916ce93/review.html">nl · positive:missing.required</a></li><li><a href="cases/fcaa7e23141c64ac0d72ae346f371a49203bfba00bc4b5b1167a42b472e0d335/review.html">en · positive:missing.required</a></li><li><a href="cases/f39bf5dbf6542faf016dc2d2402372abfc3fbbe6f5828072101d838fa9ad3300/review.html">el · hard_negative:missing.required</a></li><li><a href="cases/29cfbb086ac6b31ed80453411061fe7c1afb03697fc75b1b577df0eabc46dfdd/review.html">fr · positive:missing.conditional</a></li><li><a href="cases/186ce4323cdadbcf2d4993dac321e297250d07f2c228c2896c0c7d9d060bf935/review.html">nl · hard_negative:missing.conditional</a></li><li><a href="cases/c334dc1dd0cb8351193fa1b45ff15baca62357f75451ae616c32172f67a727a1/review.html">en · positive:structure.shift</a></li><li><a href="cases/c23ffaaaa99c792fc73679aa45fb432c9c8edce71cdbfe5339715c4d25d8b664/review.html">de · positive:structure.shift</a></li><li><a href="cases/00179669a63111e736586cccbcd6588f3d54ff7c49a67ffb9720f1a3bc7a4f11/review.html">fr · hard_negative:structure.shift</a></li><li><a href="cases/d4f6a7c4c74100babe843014cb1412e497bece2135ff457997650e803cf64e5f/review.html">nl · clean</a></li><li><a href="cases/095635d0edb559ecd27a6203a9cc861ee48b26d4552868f4d817f45a5ecf8374/review.html">el · clean</a></li></ul></details>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ <!doctype html><meta charset="utf-8"><title>TabFix v2 — test de bout en bout</title><style>body{font:16px system-ui;max-width:1450px;margin:35px auto;padding:0 24px;background:#f6f7fa;color:#182333}table{border-collapse:collapse;width:100%;background:white;margin:25px 0}th,td{text-align:left;padding:10px;border-bottom:1px solid #ddd;vertical-align:top}td code{white-space:pre-wrap;overflow-wrap:anywhere}pre{white-space:pre-wrap;max-width:600px}.repaired{background:#eaf8ee}.harmed{background:#ffe3e3}.remaining{background:#fff4df}.cards{display:flex;gap:18px;flex-wrap:wrap}.card{background:white;padding:18px;border-radius:10px}.card b{font-size:28px;display:block}</style><h1>TabFix v2 — test de bout en bout</h1><p>60/76 tableaux. CSV → filtre déterministe/neural → correction → perplexité → validation des règles. Aucune position fautive fournie au modèle.</p><div class="cards"><div class="card"><b>36/53</b>Cellules fautives réparées</div><div class="card"><b>0/6630</b>Cellules valides endommagées</div><div class="card"><b>46/53</b>Cellules fautives localisées</div></div><p>Panel synthétique représentatif ; pas une précision globale sur tout le corpus. Les cellules de contexte sont incluses. Les résultats par catégorie peuvent se recouper pour les erreurs composées. Les contradictions de groupe sans localisation certaine restent signalées sans correction.</p><h2>Par catégorie</h2><table><tr><th>Catégorie réelle</th><th>Cellules fautives</th><th>Localisées</th><th>Réparées exactement</th></tr><tr><td>text.encoding</td><td>7</td><td>7</td><td>5</td></tr><tr><td>text.invisible</td><td>2</td><td>2</td><td>2</td></tr><tr><td>text.spelling</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.number</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.date</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.boolean</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.case</td><td>2</td><td>2</td><td>2</td></tr><tr><td>format.whitespace</td><td>3</td><td>3</td><td>3</td></tr><tr><td>format.unit</td><td>3</td><td>3</td><td>3</td></tr><tr><td>schema.type</td><td>2</td><td>2</td><td>2</td></tr><tr><td>schema.identifier</td><td>4</td><td>4</td><td>4</td></tr><tr><td>schema.category</td><td>8</td><td>8</td><td>8</td></tr><tr><td>consistency.unit</td><td>2</td><td>0</td><td>0</td></tr><tr><td>consistency.dependency</td><td>3</td><td>0</td><td>0</td></tr><tr><td>consistency.temporal</td><td>2</td><td>0</td><td>0</td></tr><tr><td>missing.required</td><td>2</td><td>2</td><td>0</td></tr><tr><td>missing.conditional</td><td>2</td><td>2</td><td>0</td></tr><tr><td>structure.shift</td><td>4</td><td>4</td><td>0</td></tr></table><h2>Corrections et erreurs restantes</h2><table><tr><th>Langue / tableau</th><th>Colonne</th><th>Entrée</th><th>Sortie</th><th>Référence</th><th>Résultat</th></tr><tr class="repaired"><td><a href="cases/3612fb29d718dff78200f14d51e1a860bf3146f0539ece31761207733bc2f8cf/review.html">ja</a></td><td>品目</td><td><code>&quot;麻のバッグ&quot;</code></td><td><code>&quot;麻のバッグ&quot;</code></td><td><code>&quot;麻のバッグ&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
2
  {
3
  &quot;row&quot;: 7,
4
  &quot;column&quot;: 1,
 
461
  &quot;status&quot;: &quot;abstained: output violates a declared rule&quot;,
462
  &quot;route&quot;: &quot;model&quot;
463
  }
464
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/review.html">pl</a></td><td>Pozycja</td><td><code>&quot;Zajęcia biblioteczne&quot;</code></td><td><code>&quot;Zajęcia biblioteczne&quot;</code></td><td><code>&quot;Zajęcia biblioteczne&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
465
+ {
466
+ &quot;row&quot;: 1,
467
+ &quot;column&quot;: 1,
468
+ &quot;start&quot;: 3,
469
+ &quot;end&quot;: 5,
470
+ &quot;replacement&quot;: &quot;Zajęcia biblioteczne&quot;,
471
+ &quot;category&quot;: &quot;text.encoding&quot;,
472
+ &quot;perplexity&quot;: 1.0003071924485039,
473
+ &quot;status&quot;: &quot;applied&quot;,
474
+ &quot;route&quot;: &quot;model&quot;
475
+ }
476
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/6f8933f89b2a2d2894359fef1382ba3094ba903f9f43c14ab6f0b539a4d5f513-open/review.html">el</a></td><td>Αναγνωριστικό</td><td><code>&quot;00004&quot;</code></td><td><code>&quot;000004&quot;</code></td><td><code>&quot;000004&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
477
+ {
478
+ &quot;row&quot;: 3,
479
+ &quot;column&quot;: 0,
480
+ &quot;start&quot;: 0,
481
+ &quot;end&quot;: 5,
482
+ &quot;replacement&quot;: &quot;000004&quot;,
483
+ &quot;category&quot;: &quot;schema.identifier&quot;,
484
+ &quot;perplexity&quot;: 1.000051290442629,
485
+ &quot;status&quot;: &quot;applied&quot;,
486
+ &quot;route&quot;: &quot;model&quot;
487
+ }
488
+ ]</pre></details></td></tr><tr class="remaining"><td><a href="cases/6f8933f89b2a2d2894359fef1382ba3094ba903f9f43c14ab6f0b539a4d5f513-open/review.html">el</a></td><td>Είδος</td><td><code>&quot;Συνεδρία βιβλιοθήκης&quot;</code></td><td><code>&quot;Συνεδρία βιβλιοθήκης&quot;</code></td><td><code>&quot;Συνεδρία βιβλιοθήκης&quot;</code></td><td>remaining<details><summary>Décisions</summary><pre>[
489
+ {
490
+ &quot;row&quot;: 3,
491
+ &quot;column&quot;: 1,
492
+ &quot;start&quot;: 0,
493
+ &quot;end&quot;: 39,
494
+ &quot;replacement&quot;: null,
495
+ &quot;category&quot;: &quot;text.encoding&quot;,
496
+ &quot;perplexity&quot;: 1.0271469638964414,
497
+ &quot;status&quot;: &quot;abstained: perplexity exceeds validation threshold&quot;,
498
+ &quot;route&quot;: &quot;model&quot;
499
+ }
500
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/3e8aa3b55ce0645d85d1ca3a0878e28a476232c7a145710a566640aa3a2974d6-open/review.html">de</a></td><td>Externer Zahlencode</td><td><code>&quot;0000118&quot;</code></td><td><code>&quot;00000118&quot;</code></td><td><code>&quot;00000118&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
501
+ {
502
+ &quot;row&quot;: 6,
503
+ &quot;column&quot;: 1,
504
+ &quot;start&quot;: 0,
505
+ &quot;end&quot;: 7,
506
+ &quot;replacement&quot;: &quot;00000118&quot;,
507
+ &quot;category&quot;: &quot;schema.identifier&quot;,
508
+ &quot;perplexity&quot;: 1.0011254536020566,
509
+ &quot;status&quot;: &quot;applied&quot;,
510
+ &quot;route&quot;: &quot;model&quot;
511
+ }
512
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/3e8aa3b55ce0645d85d1ca3a0878e28a476232c7a145710a566640aa3a2974d6-open/review.html">de</a></td><td>Registrierter Anzeigename</td><td><code>&quot;Luma 東京&quot;</code></td><td><code>&quot;Luma 東京&quot;</code></td><td><code>&quot;Luma 東京&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
513
+ {
514
+ &quot;row&quot;: 6,
515
+ &quot;column&quot;: 12,
516
+ &quot;start&quot;: 4,
517
+ &quot;end&quot;: 11,
518
+ &quot;replacement&quot;: &quot;Luma 東京&quot;,
519
+ &quot;category&quot;: &quot;text.encoding&quot;,
520
+ &quot;perplexity&quot;: 1.0000107884230698,
521
+ &quot;status&quot;: &quot;applied&quot;,
522
+ &quot;route&quot;: &quot;model&quot;
523
+ }
524
+ ]</pre></details></td></tr><tr class="remaining"><td><a href="cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/review.html">ja</a></td><td>品目</td><td><code>&quot;アルミニウム&quot;</code></td><td><code>&quot;段ボール&quot;</code></td><td><code>&quot;アルミニウム&quot;</code></td><td>remaining<details><summary>Décisions</summary><pre>[
525
+ {
526
+ &quot;row&quot;: 6,
527
+ &quot;column&quot;: 1,
528
+ &quot;start&quot;: 0,
529
+ &quot;end&quot;: 18,
530
+ &quot;replacement&quot;: &quot;段ボール&quot;,
531
+ &quot;category&quot;: &quot;text.encoding&quot;,
532
+ &quot;perplexity&quot;: 1.0002686006337904,
533
+ &quot;status&quot;: &quot;applied&quot;,
534
+ &quot;route&quot;: &quot;model&quot;
535
+ }
536
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/review.html">ja</a></td><td>単位</td><td><code>&quot;kg?&quot;</code></td><td><code>&quot;kg&quot;</code></td><td><code>&quot;kg&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
537
+ {
538
+ &quot;row&quot;: 6,
539
+ &quot;column&quot;: 5,
540
+ &quot;start&quot;: 0,
541
+ &quot;end&quot;: 3,
542
+ &quot;replacement&quot;: &quot;kg&quot;,
543
+ &quot;category&quot;: &quot;schema.category&quot;,
544
+ &quot;perplexity&quot;: 1.0000396367096813,
545
+ &quot;status&quot;: &quot;applied&quot;,
546
+ &quot;route&quot;: &quot;model&quot;
547
+ }
548
+ ]</pre></details></td></tr><tr class="repaired"><td><a href="cases/740256b7487307e53e993214201d2e3570fd42bf6d61a469122fd4974ea38b0b-open/review.html">fr</a></td><td>Article</td><td><code>&quot;Accès au jardin&quot;</code></td><td><code>&quot;Accès au jardin&quot;</code></td><td><code>&quot;Accès au jardin&quot;</code></td><td>repaired<details><summary>Décisions</summary><pre>[
549
+ {
550
+ &quot;row&quot;: 4,
551
+ &quot;column&quot;: 1,
552
+ &quot;start&quot;: 3,
553
+ &quot;end&quot;: 5,
554
+ &quot;replacement&quot;: &quot;Accès au jardin&quot;,
555
+ &quot;category&quot;: &quot;text.encoding&quot;,
556
+ &quot;perplexity&quot;: 1.0002408244861347,
557
+ &quot;status&quot;: &quot;applied&quot;,
558
+ &quot;route&quot;: &quot;model&quot;
559
+ }
560
+ ]</pre></details></td></tr></table><details><summary>Tous les tableaux, y compris les contrôles valides</summary><ul><li><a href="cases/3612fb29d718dff78200f14d51e1a860bf3146f0539ece31761207733bc2f8cf/review.html">ja · positive:text.encoding</a></li><li><a href="cases/433d9c8af5be2277d006883eb58ce528d287c3fe0431becad819f8997484eee7/review.html">de · positive:text.encoding</a></li><li><a href="cases/11c0221cd3cfa0daa1e68d820026c0a2b76a932215a3e2bb43855ed9b0b468fe/review.html">fr · hard_negative:text.encoding</a></li><li><a href="cases/caf4c48e4fb55ac78afcbd23726eed87ebae8e0836c295731bb50c4f50e43e99/review.html">nl · positive:text.invisible</a></li><li><a href="cases/87e885905fc3606c63cd48369a6a570be6aa2198cf9927b667ceda60401e2e78/review.html">pl · positive:text.invisible</a></li><li><a href="cases/21ad4f83d9ffea1518a6a498f186f6a924179c96d4c8b33dc432806a37327c5e/review.html">pl · hard_negative:text.invisible</a></li><li><a href="cases/6a8776324da9c9e9197bb19fe34cbcd4c499c7da785654a238e1081314249599/review.html">de · positive:text.spelling</a></li><li><a href="cases/e73917739efb46d1ad052025dbbad8ce43535e320c3bfab88dbb2cf070020d9e/review.html">fr · positive:text.spelling, positive:format.date</a></li><li><a href="cases/b46e574a91a93f6642dccfa028c8f0220128094f176d00c06d7dce59b119a198/review.html">nl · hard_negative:text.spelling</a></li><li><a href="cases/a7aadbf7aa1ed87cf17c6d68b7ca3b2b9c98740f2cc7c5be91c987035c7a29a2/review.html">nl · positive:format.number</a></li><li><a href="cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/review.html">pl · positive:format.number</a></li><li><a href="cases/42ae0423efc79bfa2e7e92255c5e94c337c52b95167243cc3497785b15214174/review.html">de · hard_negative:format.number</a></li><li><a href="cases/415bbfbbc5a065d100a535b95d0d4f259fae060cefc532466d412eddfc7e03c7/review.html">pl · positive:format.date</a></li><li><a href="cases/95e9fdf26fc178038d17705c13b1e33edcdd0074c14db6fb67518bfe22725cab/review.html">pl · hard_negative:format.date</a></li><li><a href="cases/ec7e824cd5bf670371a04220237b3ceb64ce873cd260e06fc8e09549b10a5782/review.html">pl · positive:format.boolean</a></li><li><a href="cases/5036ba02f02df9baf2542b0114cf0000fd80e8a9958351cf1b50fca0c70ce42b/review.html">fr · positive:format.boolean</a></li><li><a href="cases/7264aa617c2f6a9abb0463459973e38ba98b12a4589ff77da50876b5fd353be0/review.html">de · hard_negative:format.boolean</a></li><li><a href="cases/299a18168469d268a9f9aca770f04d1410bc17c9bbb077b4eb95badec901239d/review.html">fr · positive:format.case, positive:schema.category</a></li><li><a href="cases/2ec9fe1f256063b86d958b20a9b1c1e811b3096630a38672b31e17811f9f9717/review.html">pl · positive:format.case</a></li><li><a href="cases/6eceda5dd8bccade58555a35ade25610195f33ba3b54df76c348edfd512c5ab7/review.html">en · hard_negative:format.case</a></li><li><a href="cases/5928e2790fc6276004825d19abd24c5230269307e8be597ce5cebc34de1553b2/review.html">de · positive:format.whitespace</a></li><li><a href="cases/85789b30d3fd26084a8927ad3a2195916b57c5fd8951d704b22be3da24bfa0d1/review.html">en · positive:format.whitespace</a></li><li><a href="cases/47256473f1211a06714df6ace7e0d856d17fed3bd4d71f4af522296043382d0e/review.html">ja · hard_negative:format.whitespace</a></li><li><a href="cases/2075d5f5a6a4ccf1843282f9d1a92f011d8682938be7346738c504d0923e3353/review.html">ja · positive:format.unit, positive:schema.type</a></li><li><a href="cases/83aafa579b76e84cf5cb7a44ab4e9724eb1a233cf5f07462bf27a673b495779f/review.html">pl · positive:format.unit, positive:schema.identifier</a></li><li><a href="cases/a5fb4ba42af578df38f103f0d7f7584caf5390d195915472fe340cf085da66fd/review.html">de · hard_negative:format.unit</a></li><li><a href="cases/275a9dadfc0c43c0dd1c1fa953a3daceac2481609dcc76a055704b7b0127fcae/review.html">fr · positive:schema.type</a></li><li><a href="cases/b3dfc51e8a64d2f8b206cd51b42b96b3250738fe276a9559eca738d1d3902adc/review.html">en · hard_negative:schema.type</a></li><li><a href="cases/b3bfd524d8c3ae833e623a5aa58c836c52f067846b5db5fbca44f3bf4ea71852/review.html">en · positive:schema.identifier, positive:consistency.temporal</a></li><li><a href="cases/a78eff0ea659f0315c65f5e3fe1ee1bcc69fd80907b537651b3602b6d9d1263f/review.html">fr · hard_negative:schema.identifier</a></li><li><a href="cases/da7b7de84b71cb8704e63df4a915e0e761a424a12c0f2a0309b5e060c6a96aba/review.html">el · positive:schema.category, positive:missing.conditional</a></li><li><a href="cases/83c6c6000a283cbefb1f55a550d45900dfe1edb8eda981ad8ce47bb9b513c3c7/review.html">pl · hard_negative:schema.category</a></li><li><a href="cases/9ed9eb83de2b77e08cc42088a8079d37463134a454cff1c8459015f380596ca8/review.html">ja · positive:consistency.unit</a></li><li><a href="cases/22e00e06d09d2bd02a565d0592ad8f4afe8169489bc73daa5d46d39cdade28ec/review.html">nl · positive:consistency.unit</a></li><li><a href="cases/9071e3bb96646641aad094c962441e56d5f01695b23420d60ba5e28fc3ac377b/review.html">pl · hard_negative:consistency.unit</a></li><li><a href="cases/f8d8a5333c6d6326bbec66ba279d61d1019dc5594dfbdc9dc83355d3668105c2/review.html">ja · positive:consistency.dependency</a></li><li><a href="cases/cb78edfa963adbebefd0856fd78d5fcb08875803a2bafcf52e94a1fb23200250/review.html">fr · positive:consistency.dependency</a></li><li><a href="cases/83e33197e71ea0876f8c972009b0caadedaf7ab29772d983a0bc40d6b31955e5/review.html">pl · hard_negative:consistency.dependency</a></li><li><a href="cases/cfea35d1a52131df05b5582234701504e0e3eb2c07686371b5c66cece911c175/review.html">el · positive:consistency.temporal</a></li><li><a href="cases/e929315e39db14f2cc01883f4f6bf657e5939c9036ffa7a611cb32d7b0e50d51/review.html">en · hard_negative:consistency.temporal</a></li><li><a href="cases/617c64c47791f492d292e1f15fb5ed3138a59debe8fbb96a601161abe916ce93/review.html">nl · positive:missing.required</a></li><li><a href="cases/fcaa7e23141c64ac0d72ae346f371a49203bfba00bc4b5b1167a42b472e0d335/review.html">en · positive:missing.required</a></li><li><a href="cases/f39bf5dbf6542faf016dc2d2402372abfc3fbbe6f5828072101d838fa9ad3300/review.html">el · hard_negative:missing.required</a></li><li><a href="cases/29cfbb086ac6b31ed80453411061fe7c1afb03697fc75b1b577df0eabc46dfdd/review.html">fr · positive:missing.conditional</a></li><li><a href="cases/186ce4323cdadbcf2d4993dac321e297250d07f2c228c2896c0c7d9d060bf935/review.html">nl · hard_negative:missing.conditional</a></li><li><a href="cases/c334dc1dd0cb8351193fa1b45ff15baca62357f75451ae616c32172f67a727a1/review.html">en · positive:structure.shift</a></li><li><a href="cases/c23ffaaaa99c792fc73679aa45fb432c9c8edce71cdbfe5339715c4d25d8b664/review.html">de · positive:structure.shift</a></li><li><a href="cases/00179669a63111e736586cccbcd6588f3d54ff7c49a67ffb9720f1a3bc7a4f11/review.html">fr · hard_negative:structure.shift</a></li><li><a href="cases/d4f6a7c4c74100babe843014cb1412e497bece2135ff457997650e803cf64e5f/review.html">nl · clean</a></li><li><a href="cases/095635d0edb559ecd27a6203a9cc861ee48b26d4552868f4d817f45a5ecf8374/review.html">el · clean</a></li><li><a href="cases/690d978ad2cd1404b2e396ca1a398493b9a0bd77e29a3b204b0c6d7ae50931b3/review.html">pl · clean</a></li><li><a href="cases/29ad166c145fdd6e01081a106d4c89e98689f71428440a7d2026d6eb49e7e458/review.html">en · clean</a></li><li><a href="cases/01c4801f223eeba89340c917d1f6c5d467ddfb39fbc83252e943e174c4aaa3fd/review.html">fr · clean</a></li><li><a href="cases/971c609092ceaf3ceee0fb8fc29fd7b4eb6262f72be7d62990d14eef903859d1/review.html">de · clean</a></li><li><a href="cases/5ae28680151872b016f2354b845fce86936d1d259b8fec403006d257e065e18c/review.html">ja · clean</a></li><li><a href="cases/8996ffd684cc285d4b48de3cc565eec5be6b6f18bd818aa8b396b4a14605eb24-open/review.html">pl · open_positive:text.encoding</a></li><li><a href="cases/6f8933f89b2a2d2894359fef1382ba3094ba903f9f43c14ab6f0b539a4d5f513-open/review.html">el · open_positive:text.encoding</a></li><li><a href="cases/3e8aa3b55ce0645d85d1ca3a0878e28a476232c7a145710a566640aa3a2974d6-open/review.html">de · open_positive:text.encoding</a></li><li><a href="cases/e32c5f023d8772f87a89590ae1d26f5699216e8657b7689b9ab3b7bdf58e4942-open/review.html">ja · open_positive:text.encoding</a></li><li><a href="cases/740256b7487307e53e993214201d2e3570fd42bf6d61a469122fd4974ea38b0b-open/review.html">fr · open_positive:text.encoding</a></li></ul></details>
evaluation/v2/summary.json CHANGED
@@ -1,22 +1,22 @@
1
  {
2
- "completed": 50,
3
  "planned": 76,
4
  "totals": {
5
- "cells": 5763,
6
- "valid_cells": 5718,
7
- "exactly_repaired_cells": 30,
8
  "valid_cells_harmed": 0,
9
- "detected_faulty_cells": 38,
10
  "flagged_valid_cells": 1,
11
- "faulty_cells": 45,
12
  "hard_negative_cases": 18,
13
  "hard_negative_cases_harmed": 0
14
  },
15
  "categories": {
16
  "text.encoding": {
17
- "gold_cells": 2,
18
- "detected_cells": 2,
19
- "exactly_repaired_cells": 2
20
  },
21
  "text.invisible": {
22
  "gold_cells": 2,
@@ -64,14 +64,14 @@
64
  "exactly_repaired_cells": 2
65
  },
66
  "schema.identifier": {
67
- "gold_cells": 2,
68
- "detected_cells": 2,
69
- "exactly_repaired_cells": 2
70
  },
71
  "schema.category": {
72
- "gold_cells": 7,
73
- "detected_cells": 7,
74
- "exactly_repaired_cells": 7
75
  },
76
  "consistency.unit": {
77
  "gold_cells": 2,
@@ -106,31 +106,31 @@
106
  },
107
  "languages": {
108
  "ja": {
109
- "cells": 437,
110
- "valid_cells": 432,
111
- "exactly_repaired_cells": 3,
112
  "valid_cells_harmed": 0,
113
- "detected_faulty_cells": 3,
114
  "flagged_valid_cells": 0,
115
- "faulty_cells": 5
116
  },
117
  "de": {
118
- "cells": 866,
119
- "valid_cells": 860,
120
- "exactly_repaired_cells": 4,
121
  "valid_cells_harmed": 0,
122
- "detected_faulty_cells": 6,
123
  "flagged_valid_cells": 0,
124
- "faulty_cells": 6
125
  },
126
  "fr": {
127
- "cells": 1188,
128
- "valid_cells": 1177,
129
- "exactly_repaired_cells": 9,
130
  "valid_cells_harmed": 0,
131
- "detected_faulty_cells": 10,
132
  "flagged_valid_cells": 0,
133
- "faulty_cells": 11
134
  },
135
  "nl": {
136
  "cells": 731,
@@ -142,17 +142,17 @@
142
  "faulty_cells": 5
143
  },
144
  "pl": {
145
- "cells": 1320,
146
- "valid_cells": 1312,
147
- "exactly_repaired_cells": 7,
148
  "valid_cells_harmed": 0,
149
- "detected_faulty_cells": 7,
150
  "flagged_valid_cells": 0,
151
- "faulty_cells": 8
152
  },
153
  "en": {
154
- "cells": 814,
155
- "valid_cells": 807,
156
  "exactly_repaired_cells": 3,
157
  "valid_cells_harmed": 0,
158
  "detected_faulty_cells": 6,
@@ -160,20 +160,21 @@
160
  "faulty_cells": 7
161
  },
162
  "el": {
163
- "cells": 407,
164
- "valid_cells": 404,
165
- "exactly_repaired_cells": 1,
166
  "valid_cells_harmed": 0,
167
- "detected_faulty_cells": 2,
168
  "flagged_valid_cells": 0,
169
- "faulty_cells": 3
170
  }
171
  },
172
  "decisions": {
173
- "applied": 30,
174
  "relation contradiction: ambiguous faulty member; unchanged": 6,
175
  "abstained: output violates a declared rule": 8,
176
- "preserved": 1
 
177
  },
178
  "model_revision": "913f3efbf4f0557b1ad5ec1693730d8b7535ac58",
179
  "dataset_revision": "602b8a454ca448713d8a5658650d4feaad8d579c",
 
1
  {
2
+ "completed": 60,
3
  "planned": 76,
4
  "totals": {
5
+ "cells": 6683,
6
+ "valid_cells": 6630,
7
+ "exactly_repaired_cells": 36,
8
  "valid_cells_harmed": 0,
9
+ "detected_faulty_cells": 46,
10
  "flagged_valid_cells": 1,
11
+ "faulty_cells": 53,
12
  "hard_negative_cases": 18,
13
  "hard_negative_cases_harmed": 0
14
  },
15
  "categories": {
16
  "text.encoding": {
17
+ "gold_cells": 7,
18
+ "detected_cells": 7,
19
+ "exactly_repaired_cells": 5
20
  },
21
  "text.invisible": {
22
  "gold_cells": 2,
 
64
  "exactly_repaired_cells": 2
65
  },
66
  "schema.identifier": {
67
+ "gold_cells": 4,
68
+ "detected_cells": 4,
69
+ "exactly_repaired_cells": 4
70
  },
71
  "schema.category": {
72
+ "gold_cells": 8,
73
+ "detected_cells": 8,
74
+ "exactly_repaired_cells": 8
75
  },
76
  "consistency.unit": {
77
  "gold_cells": 2,
 
106
  },
107
  "languages": {
108
  "ja": {
109
+ "cells": 613,
110
+ "valid_cells": 606,
111
+ "exactly_repaired_cells": 4,
112
  "valid_cells_harmed": 0,
113
+ "detected_faulty_cells": 5,
114
  "flagged_valid_cells": 0,
115
+ "faulty_cells": 7
116
  },
117
  "de": {
118
+ "cells": 1108,
119
+ "valid_cells": 1100,
120
+ "exactly_repaired_cells": 6,
121
  "valid_cells_harmed": 0,
122
+ "detected_faulty_cells": 8,
123
  "flagged_valid_cells": 0,
124
+ "faulty_cells": 8
125
  },
126
  "fr": {
127
+ "cells": 1384,
128
+ "valid_cells": 1372,
129
+ "exactly_repaired_cells": 10,
130
  "valid_cells_harmed": 0,
131
+ "detected_faulty_cells": 11,
132
  "flagged_valid_cells": 0,
133
+ "faulty_cells": 12
134
  },
135
  "nl": {
136
  "cells": 731,
 
142
  "faulty_cells": 5
143
  },
144
  "pl": {
145
+ "cells": 1457,
146
+ "valid_cells": 1448,
147
+ "exactly_repaired_cells": 8,
148
  "valid_cells_harmed": 0,
149
+ "detected_faulty_cells": 8,
150
  "flagged_valid_cells": 0,
151
+ "faulty_cells": 9
152
  },
153
  "en": {
154
+ "cells": 913,
155
+ "valid_cells": 906,
156
  "exactly_repaired_cells": 3,
157
  "valid_cells_harmed": 0,
158
  "detected_faulty_cells": 6,
 
160
  "faulty_cells": 7
161
  },
162
  "el": {
163
+ "cells": 477,
164
+ "valid_cells": 472,
165
+ "exactly_repaired_cells": 2,
166
  "valid_cells_harmed": 0,
167
+ "detected_faulty_cells": 4,
168
  "flagged_valid_cells": 0,
169
+ "faulty_cells": 5
170
  }
171
  },
172
  "decisions": {
173
+ "applied": 37,
174
  "relation contradiction: ambiguous faulty member; unchanged": 6,
175
  "abstained: output violates a declared rule": 8,
176
+ "preserved": 1,
177
+ "abstained: perplexity exceeds validation threshold": 1
178
  },
179
  "model_revision": "913f3efbf4f0557b1ad5ec1693730d8b7535ac58",
180
  "dataset_revision": "602b8a454ca448713d8a5658650d4feaad8d579c",