Capture conditions
Include faint ink, blur, skew, clipped strokes and overlapping stamps where these occur in your documents.
BETASIGNATURE DETECTION + VERIFICATION
Evaluate the beta on your documents.
Bring the forms, agreements, mandates, delivery notes, claims documents or cheques your team processes. Plan a ChequeDB beta evaluation around finding signatures, returning positions, comparing authorised references and running the validations your workflow needs.
Forms · contracts · mandates · delivery notes · cheques
Document → signature evidenceILLUSTRATION
THE FOUR QUESTIONS
Two drawn signature candidates appear on this fictional agreement. Their positions do not identify the signers.
Synthetic drawings and example workflow. No upload, model inference or live validation is performed here.
SCOPE WORKSHEET
Choose a workflow and define a useful result.
Start with a bounded document workflow. Use this worksheet to agree the inputs, expected checks and acceptance criteria before evaluating results.
| Decision | What to bring or agree |
|---|---|
| Documents in scope | Document types, page layouts, capture conditions and required signature fields |
| Reference ownership | Authorised samples, their source and the intended signer association |
| Validation requirements | Presence, position, comparison and exception checks needed for this workflow |
| Expected labels | Known signature regions, relevant signer associations and documented comparison cases |
| Operational result | Who handles exceptions, what can proceed and what evidence a reviewer needs |
| Acceptance criteria | The error and review levels your team can accept for the agreed use case |
Include the documents that make the job difficult.
Evaluate complete pages as well as selected signature crops. A clean-crop comparison cannot tell you whether signatures are found and extracted correctly from real document backgrounds.
Include different layouts, natural reference variation and examples that should not be accepted as the required signature. Label the cases clearly so a missed signature, an extra region and a wrong comparison can be counted separately.
Document structure
Include blank required fields, multiple pages, several signature regions and marks outside the expected areas.
Comparison challenges
Include genuine variation and known non-matches. If forgery screening matters, agree appropriately labelled imitation cases as a distinct test category.
Keep tuning and evaluation separate.
Use one set to agree operating rules and a separate set to evaluate them. Avoid reusing a candidate image as its own reference. Where the evaluation tests generalisation to new people, include signers absent from the tuning set while supplying their eligible references.
Record which inputs, reference sets and operating rules were evaluated. Report results by the document and quality conditions that matter to your team. A result on one clean layout should not stand in for the whole workflow.
MEASUREMENT PLAN
Measure each stage before judging the whole flow.
These are proposed evaluation measures, not published ChequeDB performance results. Agree the definitions and denominators with the evaluation team.
Keep unusable-image and reference-gap cases visible in the report. Hiding them behind one accuracy number makes the operational workload difficult to judge. Include sample counts and the criteria used to label each case.
| Stage | Measure | Question it answers |
|---|---|---|
| Detection | Missed labelled signatures and spurious candidates | Were the required regions found, and which extra marks were detected? |
| Positioning | Region overlap and crop completeness | Is the signature located on the right page and preserved in the crop? |
| Verification | False matches among known non-match comparisons | How often does a comparison accept the wrong association? |
| Verification | False non-matches among genuine comparisons | How often does genuine variation create a rejected comparison? |
| Validation | Incorrect check outcomes and unresolved cases | Do the agreed requirements produce the intended handling? |
| Operations | Review volume and exception resolution | Can the team inspect and resolve the cases that need attention? |
Bring your document requirements to the beta discussion.
Tell us which documents you process, whether you need presence checks or reference comparison, and what should happen when a requirement is unresolved. Agree the current integration contract, supported inputs, data-handling arrangements and evaluation scope before sharing the evaluation set.
The existing demo enquiry is the starting point for a signature beta evaluation. Submit your workflow requirements so we can discuss the fit and next steps.
QUESTIONS & ANSWERS
Beta evaluation: practical questions.
Is the beta limited to cheques?
No. The beta covers signature detection and verification across scanned documents generally, including forms, agreements, mandates, delivery notes, claims documents and cheques. Agree the specific layouts and inputs for your evaluation.
Can we evaluate detection without reference samples?
Yes. Detection and positioning can be evaluated on labelled document regions. Reference-based verification needs authorised samples associated with the intended signer.
What accuracy should we expect?
There is no universal percentage promised here. Evaluate detection, positioning, comparison and review outcomes on a separate, representative set, with acceptance criteria agreed for your workflow.
How do we request an evaluation?
Use the demo enquiry and describe your signature workflow. The beta discussion establishes the scope, integration requirements and next steps; submitting an enquiry does not automatically enrol you or upload documents.
CHEQUEDB SIGNATURES · BETA
Start with your documents.
Define the checks that matter.
Discuss your document types, signature references and validation workflow with the ChequeDB team.