Test set

Test set A · CFPB disclosures, forms and tables

20 values on 2 pages, each checked by a person together with the place it is printed. Every page below shows a numbered bounding box on each checked value, then the answers.

Closing Disclosure, cost details

Source: page 3 of 201403_cfpb_closing-disclosure_cover-H25B.pdf, CFPB sample form for a fictional loan (U.S. federal government work, public domain). The test reads it as a digital PDF page, with its text layer.

#	What it is	Section	Label	Value
1	Table cell	S1 · Loan Costs	row: A. Origination Charges · column: Borrower-Paid	$1,802.00
2	Table cell	S1 · Loan Costs	row: A. Origination Charges · 01 0.25 % of Loan Amount (Points) · column: Borrower-Paid · At Closing	$405.00
3	Table cell	S1 · Loan Costs	row: B. Services Borrower Did Not Shop For · 01 Appraisal Fee to John Smith Appraisers Inc. · column: Paid by Others	$405.00
4	Table cell	S1 · Loan Costs	row: B. Services Borrower Did Not Shop For · 02 Credit Report Fee to Information Inc. · column: Borrower-Paid · Before Closing	$29.80
5	Table cell	S1 · Loan Costs	row: C. Services Borrower Did Shop For · 04 Title – Lender’s Title Insurance to Epsilon Title Co. · column: Borrower-Paid · At Closing	$500.00
6	Table cell	S2 · Other Costs	row: E. Taxes and Other Government Fees · 02 Transfer Tax to Any State · column: Seller-Paid · At Closing	$950.00
7	Table cell	S2 · Other Costs	row: G. Initial Escrow Payment at Closing · 08 Aggregate Adjustment · column: Borrower-Paid · At Closing	– 0.01
8	Table cell	S2 · Other Costs	row: H. Other · 03 Home Inspection Fee to Engineers Inc. · column: Seller-Paid · Before Closing	$750.00
9	Table cell	S3 · J. Total Closing Costs	row: J. TOTAL CLOSING COSTS (Borrower-Paid) · column: Borrower-Paid	$9,712.10
10	Table cell	S3 · J. Total Closing Costs	row: J. TOTAL CLOSING COSTS (Borrower-Paid) · Closing Costs Subtotals (D + I) · column: Seller-Paid · At Closing	$12,800.00

Loan Estimate, first page

Source: page 2 of 201403_cfpb_loan-estimate_fixed-rate-loan-sample-H24B.pdf, CFPB sample form for a fictional loan (U.S. federal government work, public domain). The test reads it as an image of the page, with no text layer.

#	What it is	Section	Label	Value
1	Form field	S1 · Loan Estimate header	DATE ISSUED	2/15/2013
2	Form field	S1 · Loan Estimate header	SALE PRICE	$180,000
3	Choice	S1 · Loan Estimate header	LOAN TYPE	Conventional
4	Table cell	S2 · Loan Terms	Loan Amount	$162,000
5	Table cell	S2 · Loan Terms	row: Interest Rate · column: Can this amount increase after closing?	NO
6	Table cell	S2 · Loan Terms	row: Balloon Payment · column: Does the loan have these features?	NO
7	Table cell	S3 · Projected Payments	row: Payment Calculation · Principal & Interest · column: Years 8-30	$761.78
8	Table cell	S3 · Projected Payments	row: Estimated Total Monthly Payment · column: Years 1-7	$1,050
9	Checkbox	S3 · Projected Payments	Estimated Taxes, Insurance & Assessments · This estimate includes · Property Taxes	checked
10	Table cell	S4 · Costs at Closing	Estimated Cash to Close	$16,054

Glossary

Term	Definition
OCR	Optical character recognition: reading the text, tables and marks of a page from its image or PDF.
engine	An OCR product or model that reads a page and returns its text, such as Mistral OCR, Landing AI or Reducto.
block	One part of a page as an engine returns it: a paragraph, a table, a heading or a figure.
bounding box	The rectangle that marks where something is printed on the page.
table cell	One value of a table, where a row and a column meet.
form field	A label on a form and the value written for it.
grounding	Linking each value of a reading to its bounding box on the page.
ground truth	The correct answers for a page, one fact per value, checkbox, choice or signature. Every reading is scored against them.
test set	The pages a scoreboard is measured on, with their ground truth.
grader	The program or model that compares a reading with the ground truth and counts what is right.