OCR Foreclosure Documents: Notes, Allonges and Pay Histories

To OCR foreclosure documents, keep each instrument as its own file in the order it was produced, convert every file to a searchable PDF, and export the servicer's payment history to a spreadsheet. Then check endorsements, allonges and recording stamps against the page images one by one, because OCR is weak on those marks.

A foreclosure file is short on prose and long on small things that matter: a signature stamp on the back of a note, a loose page attached to it, a book and page number in a clerk's stamp, a column of transaction codes. Converting the file makes it searchable. It does not make any of those small things reliable until someone has looked.

Endorsements and recording stamps are easy to miss on a note. Search the converted deed of trust and the mortgage assignments for the book and page, and search the rest of the file for a lien or a lis pendens. In the foreclosure case, line those pages up with the notice and the payment history. Read every figure on the image before you quote it.

What is in a scanned foreclosure file?

A scanned foreclosure file holds the promissory note with its endorsements and any allonges, the mortgage or deed of trust, the assignments of that security instrument, the servicer's payment history, and the default and sale notices. Each proves a different thing and converts differently, so list what you have before converting any of it.

"Allonge" is the term commonly used for a paper attached to a note to carry endorsements. Section 3-204 of the Uniform Commercial Code's uniform text does not use the word. UCC § 3-204(a) defines an indorsement as a signature made on an instrument for one of the listed purposes and ends: "For the purpose of determining whether a signature is made on an instrument, a paper affixed to the instrument is a part of the instrument." Your state's enacted version of Article 3 and its case law govern what that means for a particular note, and they are for the lawyer to apply. For the person handling the scan, the sentence explains why the physical relationship between the note and that extra page is a fact worth preserving.

Borrower-side files have one more source. Under Regulation X, 12 CFR § 1024.36 requires a servicer to respond to a borrower's written request for information about the mortgage loan, and paragraph (d)(2)(i) sets the time limits: not later than "10 days (excluding legal public holidays, Saturdays, and Sundays)" for the identity of the owner or assignee of the loan, and not later than 30 days on the same counting for other requests, subject to the extension and exceptions the section sets out. What comes back from such a request is one more scanned set to log and convert.

A debtor's foreclosure papers can also turn up inside a bankruptcy case. The post on OCR for bankruptcy documents covers the rest of that file.

Converting a foreclosure file without losing the chain of assignments

Convert a foreclosure file without losing the chain by fixing the order first: number the pages of the set as received, log each document with its page range, and keep that sequence in every file name. The chain of assignments is then a table built from the log, with a page reference for every link.

Why order is evidence here

In other kinds of file, order is a convenience. In this one, where a page sat can be a fact in dispute: whether an allonge followed the note, whether an assignment's notary page belongs to it, whether the copy attached to a complaint matches the copy attached to a later affidavit. A conversion that shuffles, merges or drops pages destroys information you cannot rebuild from the text.

A procedure

1. Freeze the set. Save what you received, unchanged, and record where each file came from: the complaint's exhibits, a production, a response to an information request, the county's records.
2. Stamp a working copy. Add sequential page numbers to a duplicate so every page has an address. If the set already carries production numbers, use those and add none of your own.
3. Log every document. One row per document: first page, last page, title, date on its face, source. Count blank and reverse-side pages. The back of a note's signature page may be the page that matters.
4. Split on document boundaries only. Name each file with its page range first, such as 0012-0016 note.pdf, 0017 allonge.pdf, 0018-0019 assignment 2014.pdf. Never split a note from a page that was attached to it without recording the attachment in the log.
5. Convert each file. Upload it, review the parsed pages, and export a searchable PDF. In Schema Law that export is your original PDF with a hidden text layer added, so the page images are the ones you uploaded. Uploads are a PDF, PNG or JPEG of up to 20 MB, and the per-document page limits are on the pricing page.
6. Count pages out. The converted file must have the same number of pages as the file that went in. Record both numbers in the log.

Building the chain table

With the log complete, pull the transfers into one table. Fill it from the page images, using the converted text only to find the pages.

The parties and numbers are fictional. Keep a second, separate table for the note: each endorsement in the order it appears on the paper, who signed, for whom, and to whom it is made payable, if anyone. The two tables answer different questions, and a file can be complete on one and have a gap on the other. Put names in both tables exactly as written. "Harbor First Lending" and "Harbor First Lending, Inc." go in as two spellings, and whether they are one entity is not the table's decision.

A gap in either table is a finding to report with page numbers. What a gap means is a legal question under your jurisdiction's law.

Why do endorsements and allonges need a page-by-page check after OCR?

Endorsements and allonges need a page-by-page check because they are made of things OCR reads poorly: ink stamps set at an angle, signatures written across the stamp, small print, and marks on the reverse of a page. A wrong or missing word in the converted text can go unnoticed.

The words on these pages carry legal weight. Under the uniform text of UCC § 3-205, an indorsement by the holder that "identifies a person to whom it makes the instrument payable" is a "special indorsement", and one that is not special is a "blank indorsement"; "When indorsed in blank, an instrument becomes payable to bearer and may be negotiated by transfer of possession alone until specially indorsed." So the difference between a stamp that names a payee and a stamp with the payee line empty is the difference the statute turns on. Check your own state's enactment for the governing wording. The point for conversion is that a recognizer which skips a faint company name, or reads a blank line as a stray character, has changed a special endorsement into a blank one on your screen, or the reverse.

What goes wrong

• The stamp is not read at all. Light ink on a grey copy produces no text. Search finds nothing, and the page looks like a signature page with no endorsement.
• The signature breaks the words. A signature written across "Pay to the order of" leaves fragments such as "Pay to th rder of", which no search for the phrase will match.
• "Without recourse" disappears. Small words in a stamp's last line can drop out of the text.
• The reverse side was never scanned. If the note's last page was scanned one-sided, an endorsement on its back is absent from the file, and no conversion can supply it.
• A void or cancellation mark is ignored. A line through a stamp, or the word "VOID" written over it, may not appear in the text.

The post on whether OCR can read stamps over text explains the mechanism and which stamps cause the most trouble.

The check

For the note's signature page, its reverse, and every allonge, work from the image and fill in a row for each endorsement:

Then do three things. Compare your transcription with the converted text and note where they differ, so anyone searching the text later knows its limits on that page. Compare each copy of the note in the file with every other copy, since a note attached to a complaint and a note attached to a later filing may not carry the same endorsements. And ask whether the original is available for inspection, because the color of ink, the two sides of a sheet and the way an allonge is attached are not in any scan.

Schema Law reads pages into text and tables. Its homepage says recognition is weaker on stamps over text and handwriting and that nothing marks a passage as uncertain, so this check is done by a person with the image open.

Loan payment histories from a servicer as spreadsheet rows

A servicer's loan payment history is a table, and the useful form of it is a spreadsheet with one row per transaction: date, code, description, amount, and how the amount was split among principal, interest, escrow, fees and suspense. Export the table, stack the pages into one sheet, and prove the balances before you analyze anything.

What the history contains

Layouts differ by servicer and by the system that printed the report, so read the column headings and any code key in your own copy. Two federal sources describe the content you should expect to find in some form. For periodic statements, 12 CFR § 1026.41(d)(4) requires a list of "all the transaction activity that occurred since the last statement", including "the date of the transaction, a brief description of the transaction, and the amount of the transaction for each activity on the list." In an individual debtor's bankruptcy case, Federal Rule of Bankruptcy Procedure 3001(c)(2)(C)(i) requires Form 410A with a proof of claim that asserts a security interest in the debtor's principal residence, and the committee note to the official proof of claim forms says that form "requires the claimant to provide a loan history that reveals when payments were received, how they were applied, when fees and charges were incurred, and when escrow charges were satisfied."

Those four questions, when received, how applied, when fees were incurred, when escrow was satisfied, are the columns your workbook needs whatever the printout calls them.

Converting it

1. Convert the history as its own file and export Excel. The workbook has a sheet for each table found and another for the remaining text. Excel, CSV and JSON exports are on every plan, and the full set is listed under export formats on the homepage.
2. Stack the per-page tables into one sheet, adding a Source page column as you paste each one. The Excel conversion post has the full routine for joining tables that span several pages.
3. Delete repeated header rows from the stacked sheet, and keep the untouched per-page sheets.
4. Format the transaction code column as Text, so a code such as 017 keeps its zero.
5. Convert amounts to numbers, and check how the report shows reversals: a trailing minus, parentheses, or the letters CR. Each of those needs converting to a negative on purpose.
6. Add a Checked column.

Proving the sheet

A payment history carries its own test. If the report prints a running principal balance, compute your own beside it: the prior balance minus the principal portion of each row. Where your figure and the printed one part ways is the row with a misread digit or a missing line. Do the same for the escrow balance and any suspense balance. Then check that for each payment row, the parts add up to the amount received.

Reading the codes

Transaction codes are the servicer's shorthand. Do not assume a code means what it meant in another file, and do not interpret one from memory. Use the key printed with the report, or ask the servicer for one in writing, and add a lookup sheet to the workbook that cites where each meaning came from. Codes with no documented meaning stay labelled "unknown" in your analysis.

With the sheet proved, sorting and filtering answer the working questions: the date of the first missed payment, every fee and when it was charged, each reversal and what it reversed, payments held in suspense and when they were applied. A payment history is a ledger kept by one side. The post on tenant ledgers in Excel works through the same kind of running-balance reconciliation and application-of-payments analysis for a rent ledger.

Recorded foreclosure documents and their book, page and instrument numbers

Recorded foreclosure documents carry a recording stamp or label giving the date and time of recording and a reference number: a book and page, an instrument or document number, or both. Capture those references by hand from the image, because they are how you order the public record, retrieve certified copies and detect a missing instrument.

How documents are recorded and indexed, and what a recording reference looks like, is set by state law and local practice, so learn the format your county uses from a stamp you can read clearly before you interpret a faint one.

Where the reference appears

• A stamp or printed label on the first page, placed in a margin reserved for the recorder
• A cover sheet added by the recording office
• A band of text along the edge of every page
• A return-address block and a "prepared by" line, which are not recording data but sit beside it

A recording mark is small, may be rotated a quarter turn, and may overlap the document's own text. Recognition can miss it, split it across lines, or run it into the title. The post on real estate closing documents covers what OCR does with deeds, legal descriptions and recording stamps in general. For a foreclosure file, the task is narrower: build the index.

An index of recorded instruments

The Refers back to column does the detective work. Where an assignment recites the recording reference of the mortgage it assigns, or a notice of default recites the deed of trust, enter each recited reference from the body of the document, then compare it with the stamped reference on the document it points to. They should match digit for digit. When a recital points to a reference that is not in your index, you have found an instrument to order from the recorder. When a recital and a stamp disagree by one digit, look at both images before deciding whether the error is in the document, the stamp or your typing.

Sort the finished index by recording date and time and compare the sequence with the dates the instruments were signed. An assignment signed after the foreclosure was filed, or recorded years after it was signed, will stand out. Report it with its pages. What it means is for the lawyer.

Notary acknowledgment pages travel with recorded instruments and have their own problems in conversion. The post on scanned notarized documents covers seals, acknowledgments and jurats.

Citing a converted foreclosure document back to its scanned page

Cite a converted foreclosure document by the page of the scan, never by a page of the Word or text export, and quote only what you have read against the image. Each citation should name the document, its page in your numbered set, and its recording reference where it has one.

The searchable PDF is the citation copy. Its pages are the scanned pages, so the page a search lands on is the page a reader will see. Word, text and spreadsheet exports are for working: they reflow, and their page breaks and row numbers mean nothing to anyone else.

A citation that can be followed

For an instrument: document title, date, recording reference, and the page in the produced or filed set. For example, "Assignment of Mortgage dated September 5, 2012, recorded in Book 5107, Page 663 (Ex. C at 0042)". For a payment history row: the report's title and print date, the transaction date, and the page, as in "Loan History printed April 3, 2024, entry of 06/16/2021 (Ex. F at 0153)". Your court's citation and exhibit rules decide the final form.

In the workbook, every row that came from a scan carries its Source page value. When a figure moves from the workbook into a declaration or a brief, the page goes with it. A number without a page is a number someone will have to find again under time pressure.

Three numbering systems on one page

A recorded instrument that was produced in litigation can carry three page numbers at once: the document's own "Page 2 of 3", the recorder's book and page, and the production or exhibit number. Decide which one your citations use for location, state it once, and stay with it. The post on why PDF page numbers do not match explains the gap between a file's page position and the numbers printed on its pages.

Quoting

Copy a clause from the converted text, paste it, and then read it against the image word by word before it goes into anything filed. Rates, dates, dollar amounts and party names get a second reader. The post on quoting text from a scanned PDF sets out a proofreading routine. For endorsements and stamps, quote from your manual transcription and say in the citation that the text is transcribed from the image.

A closing check before the file is used

• Every document in the log has a converted file with a matching page count.
• The assignment table and the endorsement table each cite a page for every entry.
• The payment history sheet reproduces the printed balances, or the rows where it cannot are listed.
• Every recording reference in the index was typed from the image and checked by a second person.
• The untouched set as received is still where you saved it.

Foreclosure files are full of account numbers and borrower identifiers. Before uploading one to an outside service, read its terms against your obligations. Schema Law's commitments are set out under security on the homepage.