# CapturePoint extracted a blank or wrong field

Find which of five layers dropped the value — classification, location, recognition, validation, or export — fix only that layer, and prove the fix on several real documents.

Product: capturepoint · Audience: system-administrator · Time: 25 minutes · Last verified: 2026-08-30

Canonical: https://help.ademero.com/capturepoint/templates/capturepoint-extracted-blank-or-wrong-field

**By the end of this guide you'll know which of five layers dropped the value — classification, location, recognition, validation, or export — you'll have fixed it at that layer only, and you'll have proven the fix on several real documents instead of just the one that failed.**

> **NOTE:** A blank field on a page whose text you can read perfectly is almost never an OCR problem. The usual culprit is a field box reading the wrong spot — or the whole page matched to the wrong layout. Work the layers in order, and resist fixing two things at once.

## Work down the layers — fix only the one that failed

A value passes through five layers on its way out of CapturePoint. Find the first layer where it goes wrong:

| Layer | Question | Where to look |
| --- | --- | --- |
| 1. Classification | Did the right Document Layout match? | **Classify** view |
| 2. Location | Is the box on the value on *this* document? | **Rule Properties** > **Location & Position** |
| 3. Recognition | Can OCR read what's inside the box? | The page image itself |
| 4. Validation | Is a pattern or mask rejecting a good read? | **Identification**, **Output Translation**, **Field Properties** |
| 5. Export | Did the value drop after indexing? | **Export** view |

## Layer 1: Confirm the right Document Layout matched

1. Open the failing document in the **Classify** view. Next to it you'll see **Classified using Document Layout** and a drop-down naming the layout that matched.

2. If the wrong layout matched — or none did — stop here. The field boxes were drawn against one layout's geometry, so a wrong match makes *every* box read the wrong place. Fix classification, not the field boxes.

3. To make a layout harder to confuse with a similar one, add a second anchor: re-run the classify rule wizard, and when it asks `Would you like to improve accuracy by adding another location?`, take it.

> **NOTE:** Classification is rule-based — CapturePoint shows no classification confidence percentage. The percentage badges on fields in the **Index** view are extraction confidence, a different thing (layer 3).

*[Screenshot: Classify view with a document selected, showing the "Classified using Document Layout" label and the layout drop-down beside it.]*

## Layer 2: Check the box against this document, not the trained sample

Vendors move things — a date shifts when a logo grows, a total drops when line items overflow. The box was drawn on the trained sample; the failing page is what matters now.

1. In the **Index** view, open the **Field Rules** configuration screen and open the failing field's rule. In the **Rule Properties** dialog, set **What would you like to configure?** to **Location & Position**.

2. With **Location Type** set to **Fixed Location**, the box is the **Left**, **Top**, **Width**, and **Height** values, in inches.

3. Click **USE WIZARD** to see and redraw the box on the page — `Drag and size a box to the location of the item.` Compare where the box sits against where the value prints on the *failing* document.
   
   **What you should see:** the box covering the value with a little margin on every side. A box sitting on whitespace, or clipping half the value, is your answer — redraw it.

4. Check the rule's scope too: a **Document Layout** selector of `Any` means the rule runs for every layout, and page targeting (**First Page**, **Last Page**, **Specific Page**) decides which page it reads. A rule aimed at the wrong layout or page extracts nothing from the documents you care about.

5. If this vendor now sends two designs (old and new stationery), one box can't serve both — retrain properly instead of nudging the box back and forth. See [Train CapturePoint templates that keep working](/capturepoint/templates/train-capturepoint-templates).

> **IMPORTANT:** The dialog notes that there must be relevant text on the template image at this location for the rule to function. If the trained sample has nothing at that spot, the rule can never fire — retrain from a better sample.

## Layer 3: Can OCR actually read what's in the box?

1. Zoom into the failing page image at the box location. Faint or dot-matrix print, decorative fonts, stamps over the value, and skewed scans all defeat recognition even when a human reads them fine.

2. If the value is handwritten, turn on **Text is Handwriting** on the rule and configure the **Handwriting Recognition** section — printed-text recognition won't read handwriting.

3. In the **Index** view, look at the field's confidence badge — a percentage such as `87%` whose color scales with the score. A low score confirms a weak read; no badge means no score was recorded for that value.

4. If the image itself is the problem, fix it at the source — rescan straighter and cleaner — rather than loosening every rule to tolerate a bad scan.

> **NOTE:** For fields that matter, open **Confidence Scoring** in Rule Properties and turn on **Enable Confidence Scoring** with an **Alert Threshold**, so weak reads stop for verification instead of exporting quietly.

## Layer 4: Is validation rejecting a correctly read value?

A rule can read the value perfectly and still output blank because a pattern refuses it.

1. In **Rule Properties** > **Identification**, check the **Regular Expression Type**. A preset like `Date with 4-Digit Year (12/25/2011)` rejects `25-Dec-2026` outright — right read, wrong pattern.

2. Type the exact value from the failing page into **Test Value** and read the **Test Result:** line.
   
   **What you should see:** `Success: True, Matches: 1`. `Success: False` means this pattern is your blank field; `Invalid Regular Expression` means the pattern itself is broken.

3. Check **Output Translation** the same way — its **Input Regular Expression** and **Replacement Expression** rewrite the value after extraction, and a bad pair can empty or mangle it. It has its own **Test Value** box.

4. Check the field itself: in **Field Properties** > **Formatting**, a **Format**, **Custom Mask**, or **Decimal Places** setting can reshape what the rule delivered.

## Layer 5: Did the export drop it?

If the value looks correct on the field in the **Index** view but arrives blank or missing at the destination, extraction is fine — review the **Export** view's destination configuration and confirm the field is actually mapped to a destination field.

> **IMPORTANT:** An export error complaining about a page or image it can't read usually traces back to a damaged capture, not to your layout or mappings. Don't rebuild templates to work around it — contact support with the failing document.

## Re-test on several documents, not just the one that failed

1. Make one change at the failing layer, then on the **Field Rules** screen click **RECOGNIZE CURRENT** to re-run the rules on the open document.

2. When it passes, click **RECOGNIZE ALL**. If asked `Some documents have already been classified.` / `How would you like to proceed?`, choose the option to reprocess all documents.
   
   **What you should see:** `Reprocessing Index Rules...` with a pages-remaining count. Processing continues in the background if you close the message.

3. Judge the fix across the whole batch — a change that helps one document and breaks five others is a miss:

- [ ] The document that failed now extracts the right value
- [ ] Several other recent documents from the same layout still extract correctly
- [ ] Documents from other layouts didn't change

**Success check:** the once-blank field shows the expected value on every representative document, with no new blanks anywhere else.

## Still stuck?

Contact support with this compact evidence:

- The failing page image, plus one that extracts correctly for contrast
- The Document Layout name, and the layout shown next to **Classified using Document Layout** for the failing document
- The field name, its rule type (for example **Text Extraction**), and expected vs. actual value
- The rule's **Location & Position** values and its **Regular Expression Type** and pattern
- Whether **RECOGNIZE CURRENT** reproduces the problem every time

## What's next

- [Train CapturePoint templates that keep working](https://help.ademero.com/capturepoint/templates/train-capturepoint-templates)
- [Tell CapturePoint where documents start and end with QCards](https://help.ademero.com/capturepoint/templates/qcards-where-documents-start-and-end)