The practical answer: Use sample KYC PDFs to test ingestion, document routing, field extraction and review paths. Compare readable and degraded versions against explicit expected values. Treat identity authenticity, liveness and compliance controls as separate evaluations that require their own data and acceptance criteria.
Define what the sandbox should prove
An onboarding workflow can accept a file, identify its document type, read populated fields and send uncertain inputs for review. Each of these steps has an observable output you can test.
A sample identity document is not valid identification. Reading its fields successfully does not establish that a person has been verified. Start by listing the document-workflow behaviors you want to check, and keep verification claims outside that list.
Choose the relevant document examples
The free US KYC sample pack includes specimen US passport, passport card, driver’s license and Social Security card documents, plus a utility proof of address, all for one consistent synthetic profile. Each has a readable and degraded version, for ten PDFs in total.
Check that each document type is actually supported by your workflow. The examples are useful inputs for testing routing and extraction, but they are not a comprehensive set of US issuers, document editions or demographic variations.
Compare fields without guessing
Establish reference values from the readable source and state the normalization rules. Preserve leading zeroes in identifiers when required by your output contract. Decide whether a missing field should be absent, null or explicitly flagged.
Use the degraded counterparts to test blurred, poorly photographed, photocopied, stained and cropped input. When the source does not support a value, the expected response should reflect that uncertainty. Do not count a plausible guess as a correct extraction.
Check consistency across the applicant’s documents
Onboarding rarely evaluates one document in isolation. A name, date of birth and address may need to agree across a passport, a driver’s license and a proof of address. Because the KYC pack keeps one synthetic profile consistent across all five document types, you can first confirm that your workflow matches those fields correctly.
Then introduce disagreement deliberately, one field at a time: a transposed date of birth, an abbreviated street name, a middle name that appears on only one document. Decide in advance which differences your workflow should accept, normalize or escalate. The same discipline applies to lending files. The free banking sample pack follows one fictional borrower across a checking statement, paystub, W-2, mortgage statement and utility bill, and our loan-document consistency guide explains how to break that consistency on purpose.
Handle incomplete applications
Test what happens when a required document is missing, when the same document is uploaded twice, or when a proof of address arrives without an identity document. These are workflow states, not extraction errors, and each needs its own expected outcome: a request for the missing document, a duplicate warning or a hold for review.
Exercise the full exception path
A useful sandbox test follows the document beyond the first model response. Check whether the application explains a failed upload, handles a multi-page onboarding file, displays the relevant source and routes unreadable fields for review.
Test the resubmission path as well. Replacing a degraded sample with the readable counterpart should produce a clear new evaluation, with a traceable link to the correct document version.
Expand coverage before drawing conclusions
Five document pairs are a starting point, not a statistically representative benchmark. Record the exact document conditions and failure types rather than presenting one sample score as general KYC accuracy.
If your evaluation includes authenticity checks, liveness, fraud detection or regulatory controls, define separate tests and appropriate data for those tasks. This PDF pack does not establish performance in those areas.
Where GritWorks fits
GritRender generates fictional KYC, onboarding and banking documents with consistent profiles and controlled variations, including deliberately inconsistent fields, each with known values. GritRedact removes sensitive personal identifiers from existing KYC records when an approved use needs real documents. GritTransform extracts the populated fields into JSON for your onboarding system.
Start with the free US KYC and banking sample packs. Need custom KYC testing data for your onboarding workflow? Talk to our team.
