Handwritten Changes to a Contract: What OCR Misses

Handwritten changes to a contract usually do not come through OCR as changes. The converted text tends to give you the clause as it was printed, sometimes with stray characters where the pen went. Treat the text as the typed contract only, find the pen edits on the page images yourself, and add them to your working copy by hand.

That matters because the pen edit was made last. If the parties changed a figure at the signing table, the text you search, compare and draft from may state the term they abandoned.

What does OCR do with handwritten changes to a contract?

OCR reads a page as it looks now and has no idea that one layer of ink was added after another. A handwritten change is, to the software, some extra marks near printed words. The usual result is the printed wording, in whole or in part, and no sign that anyone altered it.

General handwriting recognition, including what happens to notes in the margin, is explained in the post on OCR and handwritten notes under margin notes on a typed page. The points below are the ones specific to a signed agreement.

The text has no "deleted" state. A line drawn through "thirty (30) days" does not remove those words from the output. Plain text, Word, JSON and a searchable PDF's hidden layer all hold characters, and a character is either there or not. Unless a person marks the deletion, the struck words read as live contract language.

Search follows the text. If "thirty" is in the text layer, a search for the notice period finds thirty. If the handwritten "sixty" was not recognized, a search for sixty finds nothing. You get a confident wrong answer and no warning.

Comparison follows the text too. Two converted copies of a contract will compare as identical when the only difference between them is in pen.

Nothing tells you where to look. Schema Law's own homepage says of its results that recognition is "strongest on clean print and weaker on faint faxes, stamps over text, and handwriting" and that "nothing marks a passage as uncertain". The same caution applies to any tool until you have tested it on your own pages. The questions section of the homepage has the full answer.

So the practical rule is short. On a contract with any pen on it other than signatures, the converted text is a transcript of the typing. The handwriting is a second source that you read from the image.

When a pen mark rewrites a covenant, log the printed words and the handwritten words on separate lines. If the margin uses the word mistake, quote it with the initials and the date beside the edit. You are recording the scan. Any question about a meeting of the minds belongs with the lawyer.

Strikethroughs, carets and interlineations on a signed page

Each kind of pen mark damages the converted text in its own way. A strikethrough usually leaves the struck words in place, a caret becomes a stray symbol or vanishes, and words written between the lines are dropped or pushed into a neighboring line. Knowing the pattern tells you what to look for.

The dangerous pairs

Two combinations deserve particular care.

A strike with a replacement is two events: words removed and words added. The text may capture one and not the other. If only the replacement is picked up, the output can read "within thirty (30) sixty days", which is at least visibly odd. If only the printed words survive, the output reads perfectly and is out of date.

An overwritten number is the hardest of all. A "3" turned into an "8" with one pen stroke leaves a shape that software has to call something. Whatever it picks will look like an ordinary digit. Check every amount, percentage, date and day count on a page that carries any pen at all.

Blanks are changes too

Form agreements leave blanks for names, dates, rent and addresses. A pen entry in a blank fills in a term of the form, so it is content and not a comment. Check filled blanks with the same care as strikes, and note any blank left empty, because an empty blank on a signed page is something the lawyer will want to know.

Are the initials beside a handwritten contract change readable?

Sometimes to a person, rarely to software. Initials are two or three joined letters with no word around them for context. Do not expect the converted text to report them. Read them on the image, and record what you see without deciding what it means.

Reading signature blocks and signatories in general is covered in the post on extracting signatures and dates from scanned documents. For a pen edit, the facts worth capturing are narrower:

• Whether there are initials at all. An edit with none beside it is recorded as "no initials seen".
• How many sets. One set, two sets, or as many as there are parties.
• Where they sit. Beside the edit, at the foot of the page, or both. Initials at the foot of every page are a different thing from initials next to one change.
• What they appear to be. Write "appears to read JMK", and compare with the signature blocks. Do not write a name unless a second document supports it.
• Whether a date is written with them.
• Ink and hand. The same pen as the edit, or visibly different. A grayscale or color scan shows this. A black-and-white scan does not.

Record these as observations. "Two sets of initials beside the change, one appears to match the Buyer's signatory" is a description. "Both parties agreed to the change" is a conclusion, and it belongs to the lawyer.

If the scan is too poor to tell whether a mark is initials, a tick or a smudge, say so in the log and ask for a better copy or the paper. Guessing here creates a record that later looks like evidence.

Finding every handwritten change before you rely on the OCR text

Look at every page image, one at a time, with the purpose of finding ink that is not print. Searching the text will not find handwriting the text never captured. No search or export does this for you.

A sweep that works:

1. Use the scan as received, at a zoom where a full page fits the screen and small pen marks are still visible.
2. Go through the pages in order and look at the margins first. Initials, brackets and notes cluster there.
3. Then run your eye down the body for lines, carets and anything between the printed lines.
4. Slow down on the pages where edits concentrate: the first page (names, dates), the commercial terms (price, term, quantities), the signature pages, and any schedule with figures.
5. Look at every blank. Filled, empty, or filled and then changed.
6. Check the back of each sheet if the scan includes it, and check that it does.
7. List each mark as you find it in a change log. Do not rely on memory or on a second pass.
8. Record a nil result. "Swept pages 1 to 20, no handwriting other than signatures" is worth writing down.

Clues in the converted text

The text cannot find handwriting, but it sometimes leaves fingerprints. After the sweep, scan the converted text for:

• a line that contains two alternative values side by side, such as two numbers or two dates;
• stray symbols in the middle of a sentence: ^, /, ~, |, or a lone bracket;
• a short run of letters or digits that belongs to no sentence;
• a clause where the wording has an obvious hole or a repeated word.

Each of these is a reason to open that page image again. None of them replaces the sweep.

Compare copies when you have more than one

If you hold two scans of the same signed contract, say one from each party, put the same page of each side by side. Pen edits made on one copy only, or initialed differently on each, show up immediately. A text comparison will not show them, which is one of the limits noted in the post on how to compare two scanned contracts.

Transcribing a handwritten contract change into the working copy

Carry each pen edit into the working copy as a visible, labeled note that shows both the printed wording and the handwritten wording, with the page it came from. Never silently overwrite the printed words. A reader of the working copy should be able to see that a change was made by hand and check it.

The change log

Start with a table, one row per mark. A fictional example, from a distribution agreement between Calloway Instruments Ltd. and Brightwater Supply Co.:

Marking the working copy

In the Word or text file you draft from, put the note at the place of the change:

4.2 Payment is due within thirty (30) days
[HANDWRITTEN CHANGE, scan p. 3: "thirty (30)" struck through; "sixty (60)"
written above; two sets of initials in right margin. Log no. 1.]
of the date of invoice.

A few rules keep the notes trustworthy:

• Keep the printed words. Leave the struck text in the working copy inside the note or beside it. Someone will need to see what was changed from.
• Use bracketed words, not formatting. Strikethrough and color disappear when text is pasted or exported. The post on transcribing handwriting sets out the bracket conventions under strike-outs, insertions and handwritten annotations in a transcript.
• Describe, do not resolve. Where a digit is uncertain, write both readings. Do not pick one.
• Have a second person check each row against the image. They should work from the scan, not from your note.
• Put a line at the top of the working copy saying that it contains handwritten changes transcribed from the signed scan, with the number of log entries.

If you later produce a clean "as changed" version of a clause for drafting, keep it as a separate, labeled paragraph and leave the annotated version in the file. Typed amendments signed as separate documents are handled differently from pen edits on the signed page, and the post on building a conformed copy of a contract covers how those are layered onto the original.

Which handwritten changes need a lawyer's decision, not a transcription?

Any handwritten change whose effect is open to argument goes to the lawyer on the matter. The transcriber's job is to report what is on the page. Whether an edit is part of the agreement, and what it means, is legal analysis that depends on the facts and the governing law.

Escalate a change, with its log row and the page image, when:

• It has no initials, or initials from one side only.
• The two parties' copies differ. The edit is on one scan and not the other, or worded differently.
• The edit touches a commercial or risk term: price, quantity, term, termination, liability, indemnity, governing law, exclusivity.
• You cannot read it with confidence, especially a digit or a date.
• The ink, hand or position suggests it was made at a different time from the signatures. Report the observation, not the inference.
• It conflicts with something else. Another clause, a figure in a schedule, an amount in words, or a later typed amendment that restates the clause without the edit.
• The contract has a clause about how it may be amended. Note the clause number beside the log entry and leave the question to the lawyer.
• A blank that matters was left empty, or filled in after the page appears to have been copied.
• The handwriting adds a whole term that the printed contract does not have, such as a sentence at the foot of a page.

Send the question in a form that can be answered quickly: the clause as printed, the change as you read it, who appears to have initialed it, the page reference and the image. Then record the answer in the log, with the name and date, so the working copy reflects a decision someone made and not an assumption.

Until that answer comes back, keep both readings alive in every document that depends on the clause. If a key dates list or a summary needs a single value, enter the printed one and the handwritten one together and flag the row.