Frequently Asked Questions
Answers to the most common questions about Review Matrix.
General
What is Review Matrix?
Review Matrix is a document extraction feature that lets you ask a set of questions across multiple documents and receive structured, citation-backed answers. It is designed for workflows like M&A due diligence, lease portfolio review, loan agreement analysis, and any scenario where you need to extract the same data points from many documents at once.
How is Review Matrix different from Document Review?
Document Review analyzes a single document in depth, identifying risks, issues, and key clauses. Review Matrix extracts specific answers from multiple documents against a shared set of questions. Use Document Review when you need a comprehensive analysis of one document. Use Review Matrix when you need to compare specific data points across many documents.
Can I run Review Matrix on a single document?
Yes. While Review Matrix is designed for multi-document extraction, you can run it on a single document. This can be useful when you have a structured set of questions you want answered from one contract and prefer the matrix format over a narrative review.
Documents
How many documents can I include in one extraction?
You can include multiple documents in a single extraction. The practical limit depends on your available credits, since each document-question pair consumes credits.
What file types are supported?
Review Matrix supports the following file types:
- PDF (including scanned PDFs, which are processed with OCR)
- DOCX (Microsoft Word)
- DOC (Legacy Microsoft Word)
- TXT (Plain text)
Can I add more documents to an existing extraction?
Not directly. If you need to include additional documents, create a new extraction with the expanded document set. You can use the same question template to ensure consistency. Previous extraction results are preserved separately.
What happens with scanned or image-based PDFs?
Judicio automatically applies OCR (optical character recognition) to scanned PDFs before extraction. The quality of extraction depends on the scan quality. Clean, high-resolution scans produce the best results.
For the best results with scanned documents, ensure the originals are scanned at 300 DPI or higher, are not skewed, and have good contrast. Handwritten text is not reliably extracted.
Questions
How many questions can I include?
There is no hard limit on the number of questions. However, more questions increase both the credit cost and the extraction time. For most workflows, 10 to 30 questions cover the key data points comprehensively.
Can I mix question types in a single extraction?
Yes. Each question has its own type (Text, Yes/No, Date, Number, List, or Clause Extract), and you can use different types for different questions within the same extraction.
What if a question does not apply to some of my documents?
If a document does not contain information relevant to a question, the answer will show as "Not found" with a low or very low confidence score. This is expected behavior -- not every question applies to every document in a mixed set.
Can I edit questions after starting an extraction?
No. Once an extraction has started, the question set is locked. If you need to change questions, wait for the current extraction to complete (or cancel it), then start a new extraction with the revised questions.
Results and accuracy
How accurate are the extracted answers?
Accuracy depends on several factors: the clarity of your questions, the quality of the source documents, and how directly the documents address each question. For well-structured documents with clear language, accuracy is typically very high (90%+ of answers are correct or substantially correct). Ambiguous contract language or poor scan quality can reduce accuracy.
What do the confidence scores mean?
Confidence scores indicate how certain Judicio is about each answer:
| Score | Meaning |
|---|---|
| 90--100 (High) | Clear, unambiguous answer found. |
| 70--89 (Medium) | Likely answer, may involve some interpretation. |
| 50--69 (Low) | Potentially relevant text found, but uncertain. |
| Below 50 (Very Low) | Could not find a confident answer. |
See Understanding Results for a full explanation.
Can I correct an answer that the AI got wrong?
Yes. Select any answer to open the edit panel, make your corrections, and save. The original AI-generated answer is preserved and you can revert at any time. See Editing Answers for details.
Are results saved automatically?
Yes. Results are saved to your project as the extraction runs. You can close the browser and return later to view completed results. In-progress extractions continue running in the background.
Credits and pricing
How are credits calculated for Review Matrix?
Credit cost is based on the number of documents multiplied by the number of questions, with adjustments for document length. The exact cost is displayed on the summary screen before you start the extraction. You will not be charged more than the displayed estimate.
Credit estimate
The credit estimate shown before extraction is a maximum. Actual usage may be slightly lower if some documents are short or do not contain relevant information for certain questions.
Can I see the credit cost before running an extraction?
Yes. The summary screen displayed before you select Start Extraction shows the estimated credit cost. Review it carefully before proceeding.
What if I run out of credits mid-extraction?
If your credits are exhausted during an extraction, the extraction pauses. You can purchase additional credits and resume the extraction from where it stopped. Results already extracted are preserved.
Exporting
What export formats are available?
Excel (.xlsx), CSV (.csv), JSON (.json), PDF (.pdf), and DOCX (.docx). See Exporting Data for details on what each format includes.
Do exports include my manual edits and annotations?
Yes. All exports reflect the current state of the matrix, including any edits, annotations, and confidence score overrides. JSON, PDF, and DOCX formats also include edit history.
Can I export a subset of the results?
Yes. Apply filters in Table view to narrow the results, then choose Export Filtered Results from the export dialog. The exported file contains only the filtered subset.
Troubleshooting
My extraction is taking a long time. Is something wrong?
Extraction time depends on the number of documents, the number of questions, and document length. A large extraction (100+ documents with 20+ questions) can take several minutes. The real-time progress indicator shows which document is currently being processed. If the progress indicator has stopped updating for more than five minutes, try refreshing the page.
Some answers show "Not found" -- what should I do?
"Not found" means the document does not appear to contain information relevant to that question. This is normal in mixed document sets. If you believe the information is present, try:
- Rephrasing the question to use terminology that matches the document.
- Checking whether the information is in an attachment or schedule that was not included in the upload.
- Verifying that the document was fully processed (check for OCR issues with scanned PDFs).
Can I cancel a running extraction?
Yes. Select Cancel Extraction from the actions menu while the extraction is running. Results already extracted are preserved. You are only charged credits for the portions that were completed.
I accidentally deleted an extraction. Can I recover it?
Deleted extractions are moved to the trash and can be recovered within 30 days. Navigate to your project's trash to restore a deleted extraction.
Still have questions?
If your question is not answered here, contact Judicio support at support@judicio.ai.