Seal Register Bench

Enrol a seal face, age and handle it, read it again from an arbitrary view, then try to substitute it. The verdict comes with the detection probability and false-alarm rate that the register's size and the inspection frequency allow.

Verdict and stated error

MATCH

Detection probability·

False-alarm rate·

Wrong IDs a year·

Where the error budget comes from

Step	Basis	Value

How large a register this rig supports

The inspector's register

Seals in the register

Reads per seal per year

Wrong IDs allowed a year

Genuine seals sent to recheck

Between inspections

Surface lost since last enrolment

Handling scratches

Contamination on the face

Feature depth ÷ width (κ)

Pattern shared by the whole batch

κ sets how deep a surface pattern persists. The bench holds from κ 1.0 down to 0.03; drag it to the worst case.

Enrolled face and this read

Remove the shared batch pattern

Register log

Surface model ported from pebbleid/surface.py: depth-correlated microstructure, three-light slope recovery, batch pattern subtraction, band correlation, registration by shift search. Seal faces simulated to machined-metal surface statistics. Probabilities beyond the 30 register entries scored per read use the normal tail.