Receipt → JSON
A receipt photo goes in; validated JSON (line items, prices, totals) comes out. Pipeline A: RapidOCR → Groq gpt-oss-20b with schema re-asks and an arithmetic check. Pipeline B: the image → Gemini Flash-Lite. A-naive is A without re-asks or checks (the baseline).
Gemini vision: 0.96 field F1 vs OCR + LLM 0.85 at 2.5x the cost (CORD-v2 test).
Results
CORD-v2 test split, the same 100 receipts for every pipeline. Field F1 with a 95 % bootstrap interval.
| Pipeline | Field F1 [95% CI] | Valid JSON % | Reconcile % | p50 s | p95 s | $ / 1k receipts | n |
|---|---|---|---|---|---|---|---|
| A-naive · OCR + LLM, no checks | 0.820 [0.765, 0.870] | 94 % | 59 % | 2.4 | 8.4 | $0.3523 | 100 |
| A · OCR + LLM (engineered) | 0.846 [0.801, 0.889] | 98 % | 63 % | 7.1 | 31.9 | $0.3930 | 100 |
| B · Gemini vision | 0.957 [0.930, 0.978] | 100 % | 85 % | 4.1 | 17.0 | $0.9751 | 100 |
| gold · scorer calibration | 1.000 [1.000, 1.000] | 100 % | 84 % | 0.0 | 0.0 | $0.0000 | 100 |
Latency excludes time queued for free-tier rate limits. List cost uses dated paid-tier prices; actual spend is $0 (free tiers).


Sample receipts
CORD-v2 validation receipts (never test), each with the stored answer of both pipelines. These are cached results: a static page runs no model. To try your own receipt, run the Gradio demo locally: uv run python app.py (pipeline A only; uploads are never sent to Gemini).
validation-0005 · printed total Rp 8.000 (8000)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| LE MINERAL | 1 | Rp 8.000 (8000) |
{
"items": [
{
"name": "LE MINERAL",
"qty": "1.00",
"unit_price": null,
"price": "8000"
}
],
"subtotal": "7273",
"tax": "727",
"total": "8000",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
⚠ Totals don't add up — needs review (off by Rp 727 (727))
| Item | Qty | Price |
|---|---|---|
| LE MINERAL | Rp 8.000 (8000) |
| Amount | |
|---|---|
| Subtotal | Rp 7.273 (7273) |
| Tax | Rp 727 (727) |
| Total | Rp 8.000 (8000) |
OCR text
LE MINERAL | B.000 1.00 NETI AMT | 7,273 TAX ANT | 727 TOTAL | 8,000 CASH | 8:000
Raw JSON
{
"items": [
{
"name": "LE MINERAL",
"qty": null,
"unit_price": null,
"price": "8000"
}
],
"subtotal": "7273",
"tax": "727",
"total": "8000",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
⚠ Totals don't add up — needs review (off by Rp 727 (727))
| Item | Qty | Price |
|---|---|---|
| LE MINERAL | 1 | Rp 8.000 (8000) |
| Amount | |
|---|---|
| Subtotal | Rp 7.273 (7273) |
| Tax | Rp 727 (727) |
| Total | Rp 8.000 (8000) |
Raw JSON
{
"items": [
{
"name": "LE MINERAL",
"qty": "1.00",
"unit_price": null,
"price": "8000"
}
],
"subtotal": "7273",
"tax": "727",
"total": "8000",
"currency": "IDR"
}validation-0033 · printed total Rp 57.500 (57500)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| GRAINS PAN BREAD | 1 | Rp 20.500 (20500) |
| ICED HIBISCUS LYCHEE TEA | 1 | Rp 37.000 (37000) |
{
"items": [
{
"name": "GRAINS PAN BREAD",
"qty": "1",
"unit_price": null,
"price": "20500"
},
{
"name": "ICED HIBISCUS LYCHEE TEA",
"qty": "1",
"unit_price": null,
"price": "37000"
}
],
"subtotal": null,
"tax": null,
"total": "57500",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| GRAINS PAN BREAD | 1 | Rp 20.500 (20500) |
| ICED HIBISCUS LYCHEE TEA | 1 | Rp 37.000 (37000) |
| Amount | |
|---|---|
| Subtotal | — |
| Tax | — |
| Total | Rp 57.500 (57500) |
OCR text
1 GRAINS PAN BREAD | 20,500 1 ICED HIBISCUS LYCHEE TEA | 37,000 TOTAL | 57,500 GIFT50 | 50,000 811923270311 CASH | 7,500 CHANGE | 0
Raw JSON
{
"items": [
{
"name": "GRAINS PAN BREAD",
"qty": "1",
"unit_price": null,
"price": "20500"
},
{
"name": "ICED HIBISCUS LYCHEE TEA",
"qty": "1",
"unit_price": null,
"price": "37000"
}
],
"subtotal": null,
"tax": null,
"total": "57500",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| GRAINS PAN BREAD | 1 | Rp 20.500 (20500) |
| ICED HIBISCUS LYCHEE TEA | 1 | Rp 37.000 (37000) |
| Amount | |
|---|---|
| Subtotal | — |
| Tax | — |
| Total | Rp 57.500 (57500) |
Raw JSON
{
"items": [
{
"name": "GRAINS PAN BREAD",
"qty": "1",
"unit_price": null,
"price": "20500"
},
{
"name": "ICED HIBISCUS LYCHEE TEA",
"qty": "1",
"unit_price": null,
"price": "37000"
}
],
"subtotal": null,
"tax": null,
"total": "57500",
"currency": "IDR"
}validation-0038 · printed total Rp 20.000 (20000)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| TT | 1 | Rp 20.000 (20000) |
{
"items": [
{
"name": "TT",
"qty": "1",
"unit_price": "20000",
"price": "20000"
}
],
"subtotal": null,
"tax": null,
"total": "20000",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
✓ Totals add up
No line items.
| Amount | |
|---|---|
| Subtotal | — |
| Tax | Rp 20.000 (20000) |
| Total | Rp 20.000 (20000) |
OCR text
820.000 TT | 20,000 TOTAL | 20000 CASH | .100,000 CHANGE | .80,000
Raw JSON
{
"items": [],
"subtotal": null,
"tax": "20000",
"total": "20000",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| 3 | 1 | Rp 20.000 (20000) |
| Amount | |
|---|---|
| Subtotal | — |
| Tax | — |
| Total | Rp 20.000 (20000) |
Raw JSON
{
"items": [
{
"name": "3",
"qty": "1",
"unit_price": "20000",
"price": "20000"
}
],
"subtotal": null,
"tax": null,
"total": "20000",
"currency": "IDR"
}validation-0049 · printed total Rp 81.400 (81400)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| Nasi CampurBali | 1 | Rp 52.000 (52000) |
| Lemon | 1 | Rp 22.000 (22000) |
{
"items": [
{
"name": "Nasi CampurBali",
"qty": "1",
"unit_price": "52000",
"price": "52000"
},
{
"name": "Lemon",
"qty": "1",
"unit_price": "22000",
"price": "22000"
}
],
"subtotal": "74000",
"tax": "7400",
"total": "81400",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| Nasi CampurBali | 1 | Rp 52.000 (52000) |
| Lemon | 1 | Rp 22.000 (22000) |
| Amount | |
|---|---|
| Subtotal | Rp 74.000 (74000) |
| Tax | Rp 7.400 (7400) |
| Total | Rp 81.400 (81400) |
OCR text
1×52.000 | 52.000 Nasi CampurBali 1×22.000 | 22.000 Lemon SubTTL | 74.000 TAX | 7.400 CASH | 81.400
Raw JSON
{
"items": [
{
"name": "Nasi CampurBali",
"qty": "1",
"unit_price": "52000",
"price": "52000"
},
{
"name": "Lemon",
"qty": "1",
"unit_price": "22000",
"price": "22000"
}
],
"subtotal": "74000",
"tax": "7400",
"total": "81400",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| Nasi Campur Bali | 1 | Rp 52.000 (52000) |
| Lemon | 1 | Rp 22.000 (22000) |
| Amount | |
|---|---|
| Subtotal | Rp 74.000 (74000) |
| Tax | Rp 7.400 (7400) |
| Total | Rp 81.400 (81400) |
Raw JSON
{
"items": [
{
"name": "Nasi Campur Bali",
"qty": "1",
"unit_price": "52000",
"price": "52000"
},
{
"name": "Lemon",
"qty": "1",
"unit_price": "22000",
"price": "22000"
}
],
"subtotal": "74000",
"tax": "7400",
"total": "81400",
"currency": "IDR"
}validation-0051 · printed total Rp 57.200 (57200)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| Round Wagyu (1gr) | 118 | Rp 47.200 (47200) |
| Wagyu Rice Box | 1 | Rp 10.000 (10000) |
{
"items": [
{
"name": "Round Wagyu (1gr)",
"qty": "118",
"unit_price": null,
"price": "47200"
},
{
"name": "Wagyu Rice Box",
"qty": "1",
"unit_price": null,
"price": "10000"
}
],
"subtotal": "57200",
"tax": null,
"total": "57200",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| 118 Round Wagyu (1gr) | Rp 47.200 (47200) | |
| 1 Wagyu Rice Box | Rp 10.000 (10000) |
| Amount | |
|---|---|
| Subtotal | Rp 57.200 (57200) |
| Tax | — |
| Total | Rp 57.200 (57200) |
OCR text
118 Round Wagyu (1gr) | 47.200 1 Wagyu Rice Box | 10.000 Subtotal | 57.200 TOTAL | 57.200 BCA | 57.200
Raw JSON
{
"items": [
{
"name": "118 Round Wagyu (1gr)",
"qty": null,
"unit_price": null,
"price": "47200"
},
{
"name": "1 Wagyu Rice Box",
"qty": null,
"unit_price": null,
"price": "10000"
}
],
"subtotal": "57200",
"tax": null,
"total": "57200",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| 118 Round Wagyu (1gr) | 118 | Rp 47.200 (47200) |
| Wagyu Rice Box | 1 | Rp 10.000 (10000) |
| Amount | |
|---|---|
| Subtotal | Rp 57.200 (57200) |
| Tax | — |
| Total | Rp 57.200 (57200) |
Raw JSON
{
"items": [
{
"name": "118 Round Wagyu (1gr)",
"qty": "118",
"unit_price": null,
"price": "47200"
},
{
"name": "Wagyu Rice Box",
"qty": "1",
"unit_price": null,
"price": "10000"
}
],
"subtotal": "57200",
"tax": null,
"total": "57200",
"currency": "IDR"
}validation-0053 · printed total Rp 31.000 (31000)

Gold label (what is printed)
| Item | Qty | Price |
|---|---|---|
| Large 1 | 1 | Rp 11.000 (11000) |
| *Rhum | 1 | Rp 0 (0) |
| Pastry Keju | 1 | Rp 20.000 (20000) |
| *Plastik kcl | 1 | Rp 0 (0) |
{
"items": [
{
"name": "Large 1",
"qty": "1",
"unit_price": null,
"price": "11000"
},
{
"name": "*Rhum",
"qty": "1",
"unit_price": null,
"price": "0"
},
{
"name": "Pastry Keju",
"qty": "1",
"unit_price": null,
"price": "20000"
},
{
"name": "*Plastik kcl",
"qty": "1",
"unit_price": null,
"price": "0"
}
],
"subtotal": null,
"tax": null,
"total": "31000",
"currency": "IDR"
}A · OCR + LLM (RapidOCR → Groq gpt-oss-20b)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| Large 1 | 1 | Rp 11.000 (11000) |
| *Rhum | 1 | Rp 0 (0) |
| Pastry Keju | Rp 20.000 (20000) | |
| *Plastik kcl | 1 | Rp 0 (0) |
| Amount | |
|---|---|
| Subtotal | — |
| Tax | — |
| Total | Rp 31.000 (31000) |
OCR text
1 Large 1 | 11.000 1 *Rhum | 0 Pastry Keju | 20.000 1 *Plastik kcl | 0 4 Total | 31.000 Bayar | 40.000 Kembali | 9.000
Raw JSON
{
"items": [
{
"name": "Large 1",
"qty": "1",
"unit_price": null,
"price": "11000"
},
{
"name": "*Rhum",
"qty": "1",
"unit_price": null,
"price": "0"
},
{
"name": "Pastry Keju",
"qty": null,
"unit_price": null,
"price": "20000"
},
{
"name": "*Plastik kcl",
"qty": "1",
"unit_price": null,
"price": "0"
}
],
"subtotal": null,
"tax": null,
"total": "31000",
"currency": "IDR"
}B · Vision LLM (Gemini Flash-Lite)
✓ Totals add up
| Item | Qty | Price |
|---|---|---|
| Large 1 | 1 | Rp 11.000 (11000) |
| *Rhum | 1 | Rp 0 (0) |
| Pastry Keju | 1 | Rp 20.000 (20000) |
| *Plastik Kcl | 1 | Rp 0 (0) |
| Amount | |
|---|---|
| Subtotal | — |
| Tax | — |
| Total | Rp 31.000 (31000) |
Raw JSON
{
"items": [
{
"name": "Large 1",
"qty": "1",
"unit_price": null,
"price": "11000"
},
{
"name": "*Rhum",
"qty": "1",
"unit_price": null,
"price": "0"
},
{
"name": "Pastry Keju",
"qty": "1",
"unit_price": null,
"price": "20000"
},
{
"name": "*Plastik Kcl",
"qty": "1",
"unit_price": null,
"price": "0"
}
],
"subtotal": null,
"tax": null,
"total": "31000",
"currency": "IDR"
}Failures
Every test receipt a pipeline did not get fully right, hand-labelled by cause.
| Cause | A | A-naive | B |
|---|---|---|---|
| other | 14 | 20 | 10 |
| ocr_misread | 11 | 8 | 0 |
| hallucinated_item | 5 | 6 | 6 |
| wrong_item_split | 7 | 7 | 0 |
| number_format | 5 | 4 | 0 |
| total | 42 | 45 | 16 |
| Pipeline | Case | Cause | Score | Note |
|---|---|---|---|---|
| A | test-0004 | wrong_item_split | 0.4444 | two items merged into one |
| A | test-0005 | hallucinated_item | 0.8889 | 'ITEMS 1.00' summary line taken as an item |
| A | test-0012 | number_format | 0.0000 | OCR fused '108,120,000'; an unparseable amount survived both repairs |
| A | test-0013 | number_format | 0.8000 | tax rate '10.00 %' read as the tax amount |
| A | test-0016 | other | 0.8750 | leading quantity left in item names |
| A | test-0017 | wrong_item_split | 0.6250 | names wrapped over two OCR lines were truncated |
| A | test-0018 | other | 0.8000 | leading quantity left in item name; subtotal added where gold has none |
| A | test-0019 | other | 0.8000 | leading quantity left in names; price cut by OCR noise |
| A | test-0021 | ocr_misread | 0.3333 | OCR merged the first price into '135.000', read as 135,000 |
| A | test-0026 | ocr_misread | 0.8000 | OCR scrambled the 'Teh Tawar' line (qty 2, 8,000 read as 0) |
| A | test-0035 | ocr_misread | 0.8000 | total read as 5,000 instead of 54,000 |
| A | test-0036 | hallucinated_item | 0.7500 | 'ITEMS:1' summary line taken as an item |
| A | test-0038 | other | 0.9565 | quantity read from OCR noise; subtotal missing |
| A | test-0039 | ocr_misread | 0.5000 | OCR garbled item names and total |
| A | test-0040 | other | 0.5000 | item lines dropped; model returned no items |
| A | test-0041 | wrong_item_split | 0.7273 | two items merged into one (twice) |
| A | test-0042 | other | 0.8571 | tax line 'P.Rest 10' not recognised as tax |
| A | test-0043 | other | 0.7500 | leading quantity left in item name |
| A | test-0045 | ocr_misread | 0.6667 | OCR column order scrambled prices and names; one item lost |
| A | test-0047 | hallucinated_item | 0.7273 | barcode lines taken as items |
| A | test-0050 | ocr_misread | 0.5000 | OCR garbled labels and names |
| A | test-0051 | wrong_item_split | 0.8333 | name continuation lines not joined to their item |
| A | test-0052 | other | 0.8571 | two identical lines collapsed into one |
| A | test-0055 | ocr_misread | 0.6667 | total read as 5,000 instead of 54,000 |
| A | test-0056 | number_format | 0.7500 | total read as negative (-8.999) |
| A | test-0057 | ocr_misread | 0.6667 | total read as 42,001 instead of 42,000 |
| A | test-0061 | ocr_misread | 0.0000 | OCR '18.0001' kept as a price; two zero-price items dropped |
| A | test-0063 | wrong_item_split | 0.8000 | modifier line 'Less Ice' split from its item |
| A | test-0066 | number_format | 0.6000 | '10.000%' read as a 10,000 tax amount; 'PLASTIK 25' split into name and price |
| A | test-0068 | other | 0.5000 | cash tendered (50,000) taken as the total |
| A | test-0070 | other | 0.6400 | invalid JSON (no receipt returned) |
| A | test-0071 | other | 0.8000 | leading quantity left in item name; subtotal added where gold has none |
| A | test-0074 | hallucinated_item | 0.8750 | category subtotals ('** FOOD') taken as items |
| A | test-0076 | hallucinated_item | 0.9091 | garbled 'TEMS' line taken as an item with a negative price |
| A | test-0085 | wrong_item_split | 0.4167 | several items merged; OCR row order scrambled |
| A | test-0089 | ocr_misread | 0.3333 | total read as 48,888 (OCR '48:888') |
| A | test-0091 | number_format | 0.2857 | '-6,000' read as a negative price; cash taken as total |
| A | test-0092 | other | 0.6667 | item line dropped; model returned no items |
| A | test-0095 | other | 0.4000 | after the repair, subtotal/tax/total fields shifted by one |
| A | test-0096 | other | 0.6667 | modifier line dropped; subtotal missing |
| A | test-0098 | ocr_misread | 0.1667 | OCR column order scrambled prices, subtotal and names |
| A | test-0099 | wrong_item_split | 0.7000 | names wrapped over two OCR lines were truncated; quantity left in name |
| A-naive | test-0003 | other | 0.8000 | item line dropped; model returned no items |
| A-naive | test-0004 | wrong_item_split | 0.4444 | two items merged into one |
| A-naive | test-0005 | hallucinated_item | 0.8889 | 'ITEMS 1.00' summary line taken as an item |
| A-naive | test-0012 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0013 | number_format | 0.8000 | tax rate '10.00 %' read as the tax amount |
| A-naive | test-0016 | other | 0.8750 | leading quantity left in item names |
| A-naive | test-0017 | wrong_item_split | 0.6250 | names wrapped over two OCR lines were truncated |
| A-naive | test-0018 | other | 0.8000 | leading quantity left in item name; subtotal added where gold has none |
| A-naive | test-0019 | other | 0.8000 | leading quantity left in names; price cut by OCR noise |
| A-naive | test-0021 | other | 0.5000 | item lines dropped; model returned no items |
| A-naive | test-0026 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0033 | hallucinated_item | 0.9231 | discount line taken as an item |
| A-naive | test-0035 | ocr_misread | 0.8000 | total read as 5,000 instead of 54,000 |
| A-naive | test-0036 | hallucinated_item | 0.7500 | 'ITEMS:1' summary line taken as an item |
| A-naive | test-0038 | other | 0.9565 | quantity read from OCR noise; subtotal missing |
| A-naive | test-0039 | ocr_misread | 0.5000 | OCR garbled item names and total |
| A-naive | test-0040 | other | 0.5000 | item lines dropped; model returned no items |
| A-naive | test-0041 | wrong_item_split | 0.7273 | two items merged into one (twice) |
| A-naive | test-0042 | other | 0.8571 | tax line 'P.Rest 10' not recognised as tax |
| A-naive | test-0043 | other | 0.7500 | leading quantity left in item name |
| A-naive | test-0045 | ocr_misread | 0.6667 | OCR column order scrambled prices and names; one item lost |
| A-naive | test-0047 | hallucinated_item | 0.7273 | barcode lines taken as items |
| A-naive | test-0050 | ocr_misread | 0.5000 | OCR garbled labels and names |
| A-naive | test-0051 | wrong_item_split | 0.8333 | name continuation lines not joined to their item |
| A-naive | test-0052 | other | 0.8571 | two identical lines collapsed into one |
| A-naive | test-0055 | ocr_misread | 0.6667 | total read as 5,000 instead of 54,000 |
| A-naive | test-0056 | number_format | 0.7500 | total read as negative (-8.999) |
| A-naive | test-0057 | ocr_misread | 0.6667 | total read as 42,001 instead of 42,000 |
| A-naive | test-0061 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0063 | wrong_item_split | 0.8000 | modifier line 'Less Ice' split from its item |
| A-naive | test-0066 | number_format | 0.6000 | '10.000%' read as a 10,000 tax amount |
| A-naive | test-0068 | other | 0.5000 | cash tendered taken as the total |
| A-naive | test-0070 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0071 | other | 0.8000 | leading quantity left in item name; subtotal added where gold has none |
| A-naive | test-0073 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0074 | hallucinated_item | 0.8750 | category subtotals ('** FOOD') taken as items |
| A-naive | test-0076 | hallucinated_item | 0.9091 | garbled 'TEMS' line taken as an item with a negative price |
| A-naive | test-0085 | wrong_item_split | 0.3478 | several items merged; OCR row order scrambled |
| A-naive | test-0089 | ocr_misread | 0.3333 | total read as 48,888 (OCR '48:888') |
| A-naive | test-0091 | number_format | 0.6667 | '-6,000' read as a negative price; quantity left in names |
| A-naive | test-0092 | other | 0.6667 | item line dropped; model returned no items |
| A-naive | test-0095 | other | 0.0000 | invalid JSON (no receipt returned) |
| A-naive | test-0096 | other | 0.6667 | modifier line dropped; subtotal missing |
| A-naive | test-0098 | ocr_misread | 0.1667 | OCR column order scrambled prices, subtotal and names |
| A-naive | test-0099 | wrong_item_split | 0.7000 | names wrapped over two OCR lines were truncated; quantity left in name |
| B | test-0000 | other | 0.7500 | item code ('901016') kept in the item name |
| B | test-0012 | other | 0.3333 | item codes kept in the item names |
| B | test-0016 | other | 0.8750 | model read the text correctly; the gold names carry OCR typos |
| B | test-0018 | other | 0.4000 | quantity and price text kept in the item name; subtotal added |
| B | test-0031 | hallucinated_item | 0.9091 | discount line taken as an item |
| B | test-0036 | other | 0.8571 | printed subtotal not extracted |
| B | test-0039 | hallucinated_item | 0.7273 | three 'DISKON' lines taken as items |
| B | test-0043 | hallucinated_item | 0.6000 | 'Box of 6' and 'PP Carrier' sub-lines taken as items |
| B | test-0061 | hallucinated_item | 0.8421 | three 'DISC ITEM' lines taken as items |
| B | test-0070 | hallucinated_item | 0.7407 | size labels ('=*LARGE*==') taken as items |
| B | test-0071 | other | 0.8000 | quantity kept in the item name; subtotal added |
| B | test-0074 | hallucinated_item | 0.8750 | two unpriced sub-lines taken as items |
| B | test-0089 | other | 0.8571 | printed subtotal not extracted; '- ' kept in the name |
| B | test-0092 | other | 0.5000 | item code ('901016') kept in the item name |
| B | test-0095 | other | 0.8000 | grand total taken as the rounded 62,000; gold has 61,799 |
| B | test-0096 | other | 0.8571 | printed subtotal not extracted |