Tips for creating effective attributes in Analyzer in Sign

Creating effective attributes is an iterative process. Start with a simple definition, review how Analyzer interprets it across real documents, and add detail only when the results show that clarification is needed.

Start with a simple definition

Begin with a direct description of the information you want to extract.

Avoid adding constraints, conditions, or examples until you understand how the initial definition performs. Starting simple makes it easier to identify which wording improves or reduces accuracy.

For example:

    Extract the retainer fee from the document.

Write clear, plain-language definitions

Describe the concept as if you were explaining it to someone who is unfamiliar with the document.

Focus on meaning rather than where the information appears.

Avoid relying on:

• Formatting.
• Section numbers.
• Page position.
• Other layout-dependent instructions.

For example:

    A rate card lists the roles or positions and the hourly rate for each.

Refine an attribute

Refinement typically takes more than one evaluation cycle.

Diagnose unexpected results

Use the pattern in the results to decide how to change the definition.

Results are too broad

If the attribute extracts unrelated content:

• Make the definition more specific.
• Specify the expected answer format when useful.
• Choose an appropriate value type, such as Number for numeric results.

For example:

    Return the answer as a percentage only.

Results are too narrow or missing

If expected values are not returned:

• Remove unnecessary constraints.
• Add positive and negative examples to clarify the intended result.
• Confirm that the target information actually appears in the documents being evaluated.

Results vary across documents

If a definition works for some documents but not others:

• Validate it first against documents you know contain the target information.
• Expand testing to a broader mix of documents only after the initial results are consistent.
• Manually verify the source document when you are unsure whether the expected value is present.

Use examples deliberately

Examples can clarify what the definition should include or exclude.

• Use positive examples to show values that should be extracted.
• Use negative examples to show values that should not be extracted.
• Limit examples to three to five when possible.

Examples should refine a clear definition, not replace one.

Use multi-step logic only when needed

Some attributes depend on context or a sequence of decisions. In those cases, describe the logic explicitly.

For example

Extract the retainer fee by following these steps:

1. Locate the fees section of the document.
2. Identify retainer fees expressed as percentages at the start of the term.
3. Treat X% as standard; all other percentages indicate negotiated rates.
4. Return the percentage and whether it is standard or negotiated.

Use this approach only when a simpler definition does not produce reliable results.

Use evaluation feedback

Review extracted values during testing and mark results as correct or incorrect.

When a result is incorrect, provide the correct value when possible. This helps you identify false positives, missed matches, and patterns that indicate where the definition needs refinement.

Validate before scaling

Before applying an attribute broadly, validate it against a representative sample of documents.

Validation helps you:

• Identify edge cases.
• Avoid applying an inaccurate definition across a large document set.
• Build confidence that the attribute behaves consistently.

If results still appear incorrect, verify the source documents before refining the definition further. Common causes include:

• Poor text quality in scanned documents.
• Password-protected or locked files.
• Corrupted source files.
• Missing target information.

Confirm that the expected information exists in the document before changing a definition that may already be correct.