Sanitized Inputs seems a little ignored...

[er1c]

import com.nulabinc.zxcvbn.{Feedback, Strength, Zxcvbn}
import scala.collection.JavaConverters._

val zxcvbn: Zxcvbn = new Zxcvbn()
val measurement = zxcvbn.measure("demo_formule2", Seq("demo", "demoformule1", "demo_formule1", "demoformule", "formule", "formule1").asJava)

scala> measurement.getScore
res2: Int = 3

scala> measurement.getFeedback.getSuggestions
res5: java.util.List[String] = []

scala> measurement.getFeedback.getWarning
res6: String = ""

I don't have a comparison to the official dropbox version of Zxcvbn, but I expected the score to get calculated less than "good" based upon the current username (demo_formule1) password being passed in as the sanitizedInputs.

Is this expected behavior, or is there some other tweak I should be making?

[er1c]

I think I want something like the "Levenshtein Distance" (https://en.wikipedia.org/wiki/Levenshtein_distance) score between the dictionary values. I can compute that in a separate code path.

[Tostino]

Not to be a jerk and promote another library in this one's issues... but Nbvcxz (https://github.com/GoSimpleLLC/nbvcxz) does implement LD calculations for dictionary values. It's been one of the most worth while enhancements i've made as far as improving the seemingly common misspellings which do not get caught by a straight dictionary lookup.

If you don't feel like changing libraries, feel free to look at how I hooked the LD calculation into Nbvcxz for inspiration on how to implement it in zxcvbn4j (I just used the Apache Commons algorithm).