Define the task
Identify the question, output type and operating conditions the evaluation actually addresses.
Forensic research & validation
Forensic research
& method validation.
Understand the task, reference population, error measures and input conditions behind a forensic analysis result.
THE BASIS OF A RESULT
01Question
02Reference population
03Measured result
04Documented scope
Four connected parts of a meaningful evaluation.How to read an evaluation
More than a headline figure.
A connected view of the evidence.
Know the population
Consider the writers, scripts, cameras, devices and media conditions represented in the reference material.
Read the measures
Look at errors, inconclusive outcomes and failure modes alongside the headline result.
Keep the scope visible
Carry the tested conditions and known boundaries into the technical record.
Research and evaluation notes
Specific tests.
Clearly defined scope.
These supplied test notes have not been independently reproduced for this website. A result from a particular dataset is not a universal error rate or case-specific probability.
9 research notes
Multi-reference signature comparison
Supplied unseen-scan notes report approximately one in five genuine samples rejected.
Test scope mattersCombined signature evidence assessment
Supplied unseen-scan notes report approximately one in seven genuine samples rejected.
Test scope mattersExperimental signature verifier
Supplied unseen-writer notes report roughly one in three skilled forgeries still accepted.
Practice onlySignature & chop extraction
Supplied screening notes report localisation recall 0.31 and automated subtype recall 0.00.
Screening toolCamera sensor-pattern comparison
Supplied validation separates two synthetic sensors; real-camera validation is not established.
PreliminaryAudio edit-point screening
Supplied validation uses synthetic tests rather than a representative casework corpus.
Screening onlyRegistered tools without descriptions
Some registered entries lack a verified input, output or implementation description. Consult the current module inventory and the latest coverage update.
Scope pendingWithdrawn two-signature comparator
The latest supplied platform review reports acceptance of 93% of different-writer pairs. The comparator and its legacy display are withdrawn from casework. This is separate from the practice-only verifier.
WithdrawnImplementation coverage update
The active synthetic-image detector and the separate handwriting tool need implementation documentation. The replacement page detector is in development. Specialist video and recovery entries require individual verification.
Coverage reviewExplore the capability in context.
Bring the evidence question and input conditions into your platform discussion.