In src/java.base/share/classes/sun/security/util/CurveDB.java, we have

        add("sect163k1 [NIST K-163]", "1.3.132.0.1", BD,...
        add("sect163r2 [NIST B-163]", "1.3.132.0.15", BD,...
        add("sect233k1 [NIST K-233]", "1.3.132.0.26", BD,...
        add("sect233r1 [NIST B-233]", "1.3.132.0.27", B,...
        add("sect283k1 [NIST K-283]", "1.3.132.0.16", BD,...
        add("sect283r1 [NIST B-283]", "1.3.132.0.17", B,...
        add("sect409k1 [NIST K-409]", "1.3.132.0.36", BD,...
        add("sect409r1 [NIST B-409]", "1.3.132.0.37", B,...
        add("sect571k1 [NIST K-571]", "1.3.132.0.38", BD,...
        add("sect571r1 [NIST B-571]", "1.3.132.0.39", B,...

So here for the same field size, NIST K-* always has a BD flag and the NIST B-* one does not have it, with only one exception -- NIST B-163 has BD. The result is that when KeyPairGenerator::init(fieldSize) is called, NIST B-163 is chosen for field size 163, and for the other field sizes where both NIST B-* and NIST K-* are available, the NIST K-* one is chosen.

I wonder if this is a typo. There should only be one curve with the BD flag, otherwise, the final length->curve map will depend on the order of add() calls, which makes the flag useless.