March 2023 Updates

[Gregory Fiumara]

Hello ELFT Stakeholders,

Several changes to the ELFT test have been made. We have updated report cards for the most current submission from each participant to reflect the changes.

Mixed-impression Distractors
Previously we ran two sets of searches: one against a set of plain impression distractors and one against a set of rolled impression distractors, both from identical subjects. We have combined the two distractor datasets into a single mixed-impression dataset.

Removal of "Hand" Results
Reporting results by region, hand, and subject was perhaps unnecessary. With the removal of a dedicated plain impression distractor dataset, hand-based accuracy has also been removed. Certain pieces of analysis will continue to display only region-based results for clarity.

Timing Dataset Reduction
The timing dataset has been reduced from 100 subjects to 25 subjects. This is to help turn results around faster while still providing a reliable measure of search speed. Search timing is now also only run with the single mixed-impression distractor dataset. Extraction timing was always based on individual impressions and is unchanged.

• Note/warning: several participants show dramatically increased search times with the mixed-impression distractor dataset, some to the point of exceeding the evaluation's threshold. The search time threshold is based on a duration per-subject, not per-impression. Durations will be enforced upon new submissions—as always, with discretion.

Size of Enrollment Templates
The overall size of enrollment templates on disk before incorporation into a database (e.g., the size of ELFT::TemplateArchive::archive) is now reported.

Similarity Score Thresholds
We now report the similarity score observed for certain FPIR values.

Coming Soon

• Feature-only probes: Probe template generation will not provide an image. This is possible for a reasonable percentage of the FBI Laboratory and FBI-Provided Solved Dataset #1 probe datasets.
• Other search scenarios: possible scenarios are described in the test plan. We previously stated we would limit search scenarios to single latent image to get the test up and running. Many participants have completed the test successfully and this limitation is no longer warranted.
• Distractor set size changes: we plan to both increase and decrease the size of the distractor set. You are hereby warned not to tune your search algorithm to 1.6m ten-finger subjects.
• New operational datasets, hopefully with new EFS features.
• Website and reporting improvements.

As always, your comments and questions are welcomed. If there's certain analysis you'd like to see, please let us know. We'd also encourage you to participate in the optional API methods extractTemplateData and extractCorrespondence. Implementing even a portion of these callbacks—such as exposing an internal quality metric—will likely be useful for future forensic science research and standards development.

Thanks,
-Greg