diff --git a/evaluation/v2/README.md b/evaluation/v2/README.md
index 97e6064f7c43b13a57d3b40ae43574cb83351969..cde36a32c110268e761942624f6260421f0a6763 100644
--- a/evaluation/v2/README.md
+++ b/evaluation/v2/README.md
@@ -2,21 +2,21 @@
Model revision: `913f3efbf4f0557b1ad5ec1693730d8b7535ac58`. Dataset: `602b8a454ca448713d8a5658650d4feaad8d579c`.
-10/76 cases completed. Gold targets/locations are never passed into CSV inference. Thresholds are frozen from validation. This is a representative synthetic panel, not the full test split.
+20/76 cases completed. Gold targets/locations are never passed into CSV inference. Thresholds are frozen from validation. This is a representative synthetic panel, not the full test split.
```json
{
- "completed": 10,
+ "completed": 20,
"planned": 76,
"totals": {
- "cells": 1219,
- "valid_cells": 1209,
- "exactly_repaired_cells": 10,
+ "cells": 2528,
+ "valid_cells": 2509,
+ "exactly_repaired_cells": 18,
"valid_cells_harmed": 0,
- "detected_faulty_cells": 10,
+ "detected_faulty_cells": 18,
"flagged_valid_cells": 0,
- "faulty_cells": 10,
- "hard_negative_cases": 3,
+ "faulty_cells": 19,
+ "hard_negative_cases": 7,
"hard_negative_cases_harmed": 0
},
"categories": {
@@ -36,28 +36,48 @@ Model revision: `913f3efbf4f0557b1ad5ec1693730d8b7535ac58`. Dataset: `602b8a454c
"exactly_repaired_cells": 2
},
"format.number": {
+ "gold_cells": 2,
+ "detected_cells": 2,
+ "exactly_repaired_cells": 2
+ },
+ "format.date": {
+ "gold_cells": 2,
+ "detected_cells": 2,
+ "exactly_repaired_cells": 2
+ },
+ "format.boolean": {
+ "gold_cells": 2,
+ "detected_cells": 2,
+ "exactly_repaired_cells": 2
+ },
+ "format.case": {
+ "gold_cells": 2,
+ "detected_cells": 2,
+ "exactly_repaired_cells": 2
+ },
+ "format.whitespace": {
"gold_cells": 1,
"detected_cells": 1,
"exactly_repaired_cells": 1
},
- "format.date": {
+ "format.unit": {
"gold_cells": 1,
"detected_cells": 1,
"exactly_repaired_cells": 1
},
- "format.boolean": {},
- "format.case": {},
- "format.whitespace": {},
- "format.unit": {},
"schema.type": {},
"schema.identifier": {},
"schema.category": {
- "gold_cells": 2,
- "detected_cells": 2,
- "exactly_repaired_cells": 2
+ "gold_cells": 3,
+ "detected_cells": 3,
+ "exactly_repaired_cells": 3
},
"consistency.unit": {},
- "consistency.dependency": {},
+ "consistency.dependency": {
+ "gold_cells": 1,
+ "detected_cells": 0,
+ "exactly_repaired_cells": 0
+ },
"consistency.temporal": {},
"missing.required": {},
"missing.conditional": {},
@@ -74,8 +94,8 @@ Model revision: `913f3efbf4f0557b1ad5ec1693730d8b7535ac58`. Dataset: `602b8a454c
"faulty_cells": 1
},
"de": {
- "cells": 229,
- "valid_cells": 226,
+ "cells": 504,
+ "valid_cells": 501,
"exactly_repaired_cells": 3,
"valid_cells_harmed": 0,
"detected_faulty_cells": 3,
@@ -83,13 +103,13 @@ Model revision: `913f3efbf4f0557b1ad5ec1693730d8b7535ac58`. Dataset: `602b8a454c
"faulty_cells": 3
},
"fr": {
- "cells": 308,
- "valid_cells": 306,
- "exactly_repaired_cells": 2,
+ "cells": 594,
+ "valid_cells": 588,
+ "exactly_repaired_cells": 6,
"valid_cells_harmed": 0,
- "detected_faulty_cells": 2,
+ "detected_faulty_cells": 6,
"flagged_valid_cells": 0,
- "faulty_cells": 2
+ "faulty_cells": 6
},
"nl": {
"cells": 352,
@@ -101,17 +121,26 @@ Model revision: `913f3efbf4f0557b1ad5ec1693730d8b7535ac58`. Dataset: `602b8a454c
"faulty_cells": 3
},
"pl": {
- "cells": 242,
- "valid_cells": 241,
- "exactly_repaired_cells": 1,
+ "cells": 836,
+ "valid_cells": 830,
+ "exactly_repaired_cells": 5,
"valid_cells_harmed": 0,
- "detected_faulty_cells": 1,
+ "detected_faulty_cells": 5,
"flagged_valid_cells": 0,
- "faulty_cells": 1
+ "faulty_cells": 6
+ },
+ "en": {
+ "cells": 154,
+ "valid_cells": 154,
+ "exactly_repaired_cells": 0,
+ "valid_cells_harmed": 0,
+ "detected_faulty_cells": 0,
+ "flagged_valid_cells": 0
}
},
"decisions": {
- "applied": 10
+ "applied": 18,
+ "relation contradiction: ambiguous faulty member; unchanged": 1
},
"model_revision": "913f3efbf4f0557b1ad5ec1693730d8b7535ac58",
"dataset_revision": "602b8a454ca448713d8a5658650d4feaad8d579c",
diff --git a/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/input.csv b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/input.csv
new file mode 100644
index 0000000000000000000000000000000000000000..1c7da60015ca6ab3add406f0c1bfc641d2d9bf9e
--- /dev/null
+++ b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/input.csv
@@ -0,0 +1,14 @@
+Kod rekordu,Zewnętrzny kod cyfrowy,Pozycja,Deklarowana masa sztuki,Jednostka deklarowanej masy,Zmierzona masa sztuki,Jednostka pomiaru,Masa wody w gramach,Masa mieszaniny w gramach,Partia produkcyjna,Zarejestrowana nazwa,Uwagi obsługi
+RC-000000,00000100,Mieszanka indygo,0.125,kg,125,g,2375,2500,LT-000200,luma,
+RC-000001,00000101,Mieszanka ochry,1000,g,1,kg,19000,20000,LT-000201,LUMA," ""Przed przeniesieniem Mieszanka ochry sprawdzić dołączoną kartę; stanowisko 2 obsługuje ten rekord.""
+2 "
+RC-000002,00000102,Mieszanka marzanny,1.875,kg,1875,g,35625,37500,LT-000202,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 3 obsługuje ten rekord.
+RC-000038,00000138,Mieszanka marzanny,1.75,kg,1750,g,33250,35000,LT-000238,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 4 obsługuje ten rekord.
+RC-000039,00000139,Mieszanka orzecha,2625,g,2.625,kg,49875,52500,LT-000239,Luma Atelier,Przed przeniesieniem Mieszanka orzecha sprawdzić dołączoną kartę; stanowisko 5 obsługuje ten rekord.
+RC-000040,00000140,Mieszanka ochry,"0,625 h",kg,625,g,11875,12500,LT-000240,luma,
+RC-000041,00000141,Mieszanka marzanny,1500,g,1.5,kg,28500,30000,LT-000241,LUMA," ""Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 7 obsługuje ten rekord.""
+42 "
+RC-000042,00000142,Mieszanka indygo,2.375,kg,2375,g,45125,47500,LT-000242,Luma 東京,Przed przeniesieniem Mieszanka indygo sprawdzić dołączoną kartę; stanowisko 1 obsługuje ten rekord.
+RC-000048,00000148,Mieszanka indygo,1.875,kg,1875,g,35625,37500,LT-000248,luma,
+RC-000050,00000150,Mieszanka marzanny,0.75,kg,750,g,14250,15000,LT-000250,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 2 obsługuje ten rekord.
+RC-000051,00000151,Mieszanka indygo,1625,g,1.625,kg,30875,32500,LT-000251,Luma Atelier,Przed przeniesieniem Mieszanka indygo sprawdzić dołączoną kartę; stanowisko 3 obsługuje ten rekord.
diff --git a/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/audit.json b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/audit.json
new file mode 100644
index 0000000000000000000000000000000000000000..b041172c356afbfd1260f27f87e3b159cfee9532
--- /dev/null
+++ b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/audit.json
@@ -0,0 +1,27 @@
+{
+ "metrics": {
+ "device": "cuda",
+ "rows": 11,
+ "cells": 132,
+ "threshold": 0.9,
+ "load_seconds": 0.0,
+ "inference_seconds": 0.88,
+ "cells_per_second": 150.78,
+ "peak_gpu_bytes": 1927847936,
+ "detected_spans": 1,
+ "changed_cells": 1
+ },
+ "proposals": [
+ {
+ "row": 5,
+ "column": 3,
+ "start": 0,
+ "end": 7,
+ "replacement": "0.625",
+ "category": "format.number",
+ "perplexity": 1.0000343913533796,
+ "status": "applied",
+ "route": "model"
+ }
+ ]
+}
\ No newline at end of file
diff --git a/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/corrected.csv b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/corrected.csv
new file mode 100644
index 0000000000000000000000000000000000000000..0ef1ffe041f2e36df9615ecf0cbdae75eae6c749
--- /dev/null
+++ b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/corrected.csv
@@ -0,0 +1,14 @@
+Kod rekordu,Zewnętrzny kod cyfrowy,Pozycja,Deklarowana masa sztuki,Jednostka deklarowanej masy,Zmierzona masa sztuki,Jednostka pomiaru,Masa wody w gramach,Masa mieszaniny w gramach,Partia produkcyjna,Zarejestrowana nazwa,Uwagi obsługi
+RC-000000,00000100,Mieszanka indygo,0.125,kg,125,g,2375,2500,LT-000200,luma,
+RC-000001,00000101,Mieszanka ochry,1000,g,1,kg,19000,20000,LT-000201,LUMA," ""Przed przeniesieniem Mieszanka ochry sprawdzić dołączoną kartę; stanowisko 2 obsługuje ten rekord.""
+2 "
+RC-000002,00000102,Mieszanka marzanny,1.875,kg,1875,g,35625,37500,LT-000202,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 3 obsługuje ten rekord.
+RC-000038,00000138,Mieszanka marzanny,1.75,kg,1750,g,33250,35000,LT-000238,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 4 obsługuje ten rekord.
+RC-000039,00000139,Mieszanka orzecha,2625,g,2.625,kg,49875,52500,LT-000239,Luma Atelier,Przed przeniesieniem Mieszanka orzecha sprawdzić dołączoną kartę; stanowisko 5 obsługuje ten rekord.
+RC-000040,00000140,Mieszanka ochry,0.625,kg,625,g,11875,12500,LT-000240,luma,
+RC-000041,00000141,Mieszanka marzanny,1500,g,1.5,kg,28500,30000,LT-000241,LUMA," ""Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 7 obsługuje ten rekord.""
+42 "
+RC-000042,00000142,Mieszanka indygo,2.375,kg,2375,g,45125,47500,LT-000242,Luma 東京,Przed przeniesieniem Mieszanka indygo sprawdzić dołączoną kartę; stanowisko 1 obsługuje ten rekord.
+RC-000048,00000148,Mieszanka indygo,1.875,kg,1875,g,35625,37500,LT-000248,luma,
+RC-000050,00000150,Mieszanka marzanny,0.75,kg,750,g,14250,15000,LT-000250,Luma 東京,Przed przeniesieniem Mieszanka marzanny sprawdzić dołączoną kartę; stanowisko 2 obsługuje ten rekord.
+RC-000051,00000151,Mieszanka indygo,1625,g,1.625,kg,30875,32500,LT-000251,Luma Atelier,Przed przeniesieniem Mieszanka indygo sprawdzić dołączoną kartę; stanowisko 3 obsługuje ten rekord.
diff --git a/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/review.html b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/review.html
new file mode 100644
index 0000000000000000000000000000000000000000..834fd61395ce60c801e31b371634c79b9a53a300
--- /dev/null
+++ b/evaluation/v2/cases/0a034e5c80b37ff52a4374db60437743b58d540fafffd8860225c34f4c558a1c/output/review.html
@@ -0,0 +1,18 @@
+TabFix — end-to-end CSV review
+Does TabFix fix this CSV?
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.
+132Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.88 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 5,
+ "column": 3,
+ "start": 0,
+ "end": 7,
+ "replacement": "0.625",
+ "category": "format.number",
+ "perplexity": 1.0000343913533796,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+132Cells scanned
1Errors in input
1Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.88 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 5,
+ "column": 3,
+ "start": 0,
+ "end": 7,
+ "replacement": "0.625",
+ "category": "format.number",
+ "perplexity": 1.0000343913533796,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+132Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.51 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 2,
+ "start": 0,
+ "end": 9,
+ "replacement": "RT-000002",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ },
+ {
+ "row": 4,
+ "column": 7,
+ "start": 0,
+ "end": 3,
+ "replacement": "kg",
+ "category": "schema.category",
+ "perplexity": 1.0000125169191336,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+132Cells scanned
2Errors in input
2Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.51 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 2,
+ "start": 0,
+ "end": 9,
+ "replacement": "RT-000002",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ },
+ {
+ "row": 4,
+ "column": 7,
+ "start": 0,
+ "end": 3,
+ "replacement": "kg",
+ "category": "schema.category",
+ "perplexity": 1.0000125169191336,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+132Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.41 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 9,
+ "replacement": "RC-000024",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]Does TabFix fix this CSV?
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.
+132Cells scanned
1Errors in input
1Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.41 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 9,
+ "replacement": "RC-000024",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]Does TabFix fix this CSV?
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.
+88Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.76 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 7,
+ "column": 2,
+ "start": 0,
+ "end": 10,
+ "replacement": "2024-12-04",
+ "category": "format.date",
+ "perplexity": 1.0021802994372715,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+88Cells scanned
1Errors in input
1Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.76 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 7,
+ "column": 2,
+ "start": 0,
+ "end": 10,
+ "replacement": "2024-12-04",
+ "category": "format.date",
+ "perplexity": 1.0021802994372715,
+ "status": "applied",
+ "route": "model"
+ }
+]Does TabFix fix this CSV?
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.
+121Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.32 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+121Cells scanned
0Errors in input
0Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.32 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.49 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 10,
+ "replacement": "RC-000004",
+ "category": "format.whitespace",
+ "status": "applied",
+ "route": "deterministic"
+ },
+ {
+ "row": 4,
+ "column": 4,
+ "start": 0,
+ "end": 4,
+ "replacement": "true",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]Does TabFix fix this CSV?
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.
+154Cells scanned
2Errors in input
2Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.49 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 10,
+ "replacement": "RC-000004",
+ "category": "format.whitespace",
+ "status": "applied",
+ "route": "deterministic"
+ },
+ {
+ "row": 4,
+ "column": 4,
+ "start": 0,
+ "end": 4,
+ "replacement": "true",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]Does TabFix fix this CSV?
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.
+154Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.47 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
0Errors in input
0Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.47 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.49 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
0Errors in input
0Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.49 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.57 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+154Cells scanned
0Errors in input
0Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.57 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[]
Does TabFix fix this CSV?
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.
+88Cells scanned
—Errors in input
—Errors repaired
—Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.21 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 8,
+ "column": 4,
+ "category": "consistency.dependency",
+ "status": "relation contradiction: ambiguous faulty member; unchanged"
+ },
+ {
+ "row": 8,
+ "column": 6,
+ "start": 0,
+ "end": 3,
+ "replacement": "Nie",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]Does TabFix fix this CSV?
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.
+88Cells scanned
2Errors in input
1Errors repaired
0Valid cells changed
CSV inference review · Threshold 0.9 · CPU inference 0.21 seconds (model loading excluded).
Errors and proposed changes
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.
Complete model output
Every detection and application decision
[
+ {
+ "row": 8,
+ "column": 4,
+ "category": "consistency.dependency",
+ "status": "relation contradiction: ambiguous faulty member; unchanged"
+ },
+ {
+ "row": 8,
+ "column": 6,
+ "start": 0,
+ "end": 3,
+ "replacement": "Nie",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+]TabFix v2 — test de bout en bout
10/76 tableaux. CSV → filtre déterministe/neural → correction → perplexité → validation des règles. Aucune position fautive fournie au modèle.
10/10Cellules fautives réparées
0/1209Cellules valides endommagées
10/10Cellules fautives localisées
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.
Par catégorie
| Catégorie réelle | Cellules fautives | Localisées | Réparées exactement |
|---|
| text.encoding | 2 | 2 | 2 |
| text.invisible | 2 | 2 | 2 |
| text.spelling | 2 | 2 | 2 |
| format.number | 1 | 1 | 1 |
| format.date | 1 | 1 | 1 |
| format.boolean | 0 | 0 | 0 |
| format.case | 0 | 0 | 0 |
| format.whitespace | 0 | 0 | 0 |
| format.unit | 0 | 0 | 0 |
| schema.type | 0 | 0 | 0 |
| schema.identifier | 0 | 0 | 0 |
| schema.category | 2 | 2 | 2 |
| consistency.unit | 0 | 0 | 0 |
| consistency.dependency | 0 | 0 | 0 |
| consistency.temporal | 0 | 0 | 0 |
| missing.required | 0 | 0 | 0 |
| missing.conditional | 0 | 0 | 0 |
| structure.shift | 0 | 0 | 0 |
Corrections et erreurs restantes
| Langue / tableau | Colonne | Entrée | Sortie | Référence | Résultat |
|---|
| ja | 品目 | "麻ã®ããã°" | "麻のバッグ" | "麻のバッグ" | repairedDécisions[
+TabFix v2 — test de bout en boutTabFix v2 — test de bout en bout20/76 tableaux. CSV → filtre déterministe/neural → correction → perplexité → validation des règles. Aucune position fautive fournie au modèle. 18/19Cellules fautives réparées 0/2509Cellules valides endommagées 18/19Cellules fautives localisées 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. Par catégorie | Catégorie réelle | Cellules fautives | Localisées | Réparées exactement |
|---|
| text.encoding | 2 | 2 | 2 | | text.invisible | 2 | 2 | 2 | | text.spelling | 2 | 2 | 2 | | format.number | 2 | 2 | 2 | | format.date | 2 | 2 | 2 | | format.boolean | 2 | 2 | 2 | | format.case | 2 | 2 | 2 | | format.whitespace | 1 | 1 | 1 | | format.unit | 1 | 1 | 1 | | schema.type | 0 | 0 | 0 | | schema.identifier | 0 | 0 | 0 | | schema.category | 3 | 3 | 3 | | consistency.unit | 0 | 0 | 0 | | consistency.dependency | 1 | 0 | 0 | | consistency.temporal | 0 | 0 | 0 | | missing.required | 0 | 0 | 0 | | missing.conditional | 0 | 0 | 0 | | structure.shift | 0 | 0 | 0 |
Corrections et erreurs restantes| Langue / tableau | Colonne | Entrée | Sortie | Référence | Résultat |
|---|
| ja | 品目 | "麻ã®ããã°" | "麻のバッグ" | "麻のバッグ" | repairedDécisions[
{
"row": 7,
"column": 1,
@@ -118,4 +118,102 @@
"status": "applied",
"route": "model"
}
-] |
Tous les tableaux, y compris les contrôles valides |
| pl | Deklarowana masa sztuki | "0,625 h" | "0.625" | "0.625" | repairedDécisions[
+ {
+ "row": 5,
+ "column": 3,
+ "start": 0,
+ "end": 7,
+ "replacement": "0.625",
+ "category": "format.number",
+ "perplexity": 1.0000343913533796,
+ "status": "applied",
+ "route": "model"
+ }
+] |
| pl | Data rozpoczęcia | "04/12/2024" | "2024-12-04" | "2024-12-04" | repairedDécisions[
+ {
+ "row": 7,
+ "column": 2,
+ "start": 0,
+ "end": 10,
+ "replacement": "2024-12-04",
+ "category": "format.date",
+ "perplexity": 1.0021802994372715,
+ "status": "applied",
+ "route": "model"
+ }
+] |
| pl | Zapisany rozmiar (MiB) | "34,82" | "34,82" | "33,82" | remainingDécisions[
+ {
+ "row": 8,
+ "column": 4,
+ "category": "consistency.dependency",
+ "status": "relation contradiction: ambiguous faulty member; unchanged"
+ }
+] |
| pl | Powodzenie | "nIE" | "Nie" | "Nie" | repairedDécisions[
+ {
+ "row": 8,
+ "column": 6,
+ "start": 0,
+ "end": 3,
+ "replacement": "Nie",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+] |
| fr | Code dossier | " RC-000004" | "RC-000004" | "RC-000004" | repairedDécisions[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 10,
+ "replacement": "RC-000004",
+ "category": "format.whitespace",
+ "status": "applied",
+ "route": "deterministic"
+ }
+] |
| fr | Réception confirmée | "TRUE" | "true" | "true" | repairedDécisions[
+ {
+ "row": 4,
+ "column": 4,
+ "start": 0,
+ "end": 4,
+ "replacement": "true",
+ "category": "format.boolean",
+ "status": "applied",
+ "route": "deterministic"
+ }
+] |
| fr | Code trajet | "rt-000002" | "RT-000002" | "RT-000002" | repairedDécisions[
+ {
+ "row": 4,
+ "column": 2,
+ "start": 0,
+ "end": 9,
+ "replacement": "RT-000002",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ }
+] |
| fr | Unité de mesure | "kg?" | "kg" | "kg" | repairedDécisions[
+ {
+ "row": 4,
+ "column": 7,
+ "start": 0,
+ "end": 3,
+ "replacement": "kg",
+ "category": "schema.category",
+ "perplexity": 1.0000125169191336,
+ "status": "applied",
+ "route": "model"
+ }
+] |
| pl | Kod rekordu | "rc-000024" | "RC-000024" | "RC-000024" | repairedDécisions[
+ {
+ "row": 4,
+ "column": 0,
+ "start": 0,
+ "end": 9,
+ "replacement": "RC-000024",
+ "category": "format.case",
+ "status": "applied",
+ "route": "deterministic"
+ }
+] |
Tous les tableaux, y compris les contrôles valides