BEX Workstation

Decision confirmation

Match appears top left, No match bottom right, Inconclusive in the centre. Each can be recalled until its window closes.

Subject impression in the comparison model — A working model of the comparison screen.

Candidate impression in the comparison model — A working model of the comparison screen.

Modalities

Ten printRolled, plain and palms
LatentAnalysis before candidates
FaceProbe beside each candidate
IrisLeft and right, segmented

Fourteen impressions

Ten rolled impressions, two four-finger slaps and two thumbs. The workstation can compare any two of them.

impressions per ten print record

The fourteen impressions of a ten print record
Rolled
Right thumb	1
Right index	1
Right middle	1
Right ring	1
Right little	1
Left thumb	1
Left index	1
Left middle	1
Left ring	1
Left little	1
Plain
Right four-finger slap	1
Left four-finger slap	1
Right thumb	1
Left thumb	1
Impressions	14

Latent examination

Candidates stay sealed until the examiner has marked the latent and committed the analysis, so no candidate can influence the features. Sealing is an agency policy setting, and each examination records whether it was sealed. Ask for a walkthrough.

A partial latent fingerprint lifted from a surface, grey ridges on a mottled ground — A working model of the latent screen. Mark features on the latent, then commit the analysis to see the candidates.

A working model of the latent screen. Mark features on the latent, then commit the analysis to see the candidates.

How a case moves

A case is submitted.

curl -X POST "$BEX/api/v1/intake/cases" \
  -H "Authorization: Bearer $BEX_SERVICE_TOKEN" \
  -F modality=tenprint \
  -F external_id=DEMO-702FD3531F \
  -F queues="Ten print · Standard" \
  -F source="Demo corpus (demo-prints)" \
  -F images='[{"role":"subject","position":1,"ppi":500}]' \
  -F files=@r1-thumb.png

The next available examiner gets the case. Select a finger to compare it.

An examiner decides.

Match

DEMO-702FD3531F · rendered 15:30:45Z

15 Recall window open. Publishes to the system of record at zero.

The decision stays pending until the recall window closes. Agency policy sets the window, up to 25 seconds.

The record replays it.

Audit replay

Subject impression as it was at this step — Each step is re-rendered from the stored operations rather than from screenshots.

Candidate impression as it was at this step — Each step is re-rendered from the stored operations rather than from screenshots.

Limits and safeguards

The numbers and limits the workstation is built and tested to
Area	What	How it is held	Figure
Speed	Next case on screen	Measured on every draw, and 401 ms at p95 in load testing.	< 5 s p95
	Concurrent examiners	Work is leased in Postgres, without a separate queue service.	100
	Queue depth without slowdown	Each queue is read through its own index.	300,000

Record	Maximum recall window	Enforced by a database constraint.	25 s
	Audit trail edits	The audit table is append-only and hash-chained.	None
	Stored screenshots	The record stores operations and replays them.	None

Imaging	Lossy re-encoding	Tiles are PNG, and lossy input is marked LOSSY.	None
	Built-in matcher	Matching stays with the search source.	None

Boundary	Unconfigured outbound requests	Every asset is served locally, and redirects are refused.	None
	Services to run	API, imaging worker, web, Postgres.	4

Book a demo or get access

Request a demo or Request access

Both open a pre-filled email to Capital Technology Group.