Test a threshold of 0

Make sure we get near-perfect recall (ATM-light) and bad precision.

javascript/ql/experimental/adaptivethreatmodeling/lib/experimental/adaptivethreatmodeling/ATMConfig.qll

@@ -105,8 +105,8 @@ abstract class ATMConfig extends string {
    *
    * Specifies the default cut-off value that controls how many alerts are produced.
    * The cut-off value must be in the range [0,1].
-   * A cut-off value of 0 only produces alerts that are likely true-positives.
-   * A cut-off value of 1 produces all alerts including those that are likely false-positives.
+   * A cut-off value of >~0.5 only produces alerts that are likely true-positives.
+   * A cut-off value of 0 produces all alerts including those that are likely false-positives.
    */
-  float getScoreCutoff() { result = 0.0 }
+  float getScoreCutoff() { result = 0. }
 }

javascript/ql/experimental/adaptivethreatmodeling/lib/experimental/adaptivethreatmodeling/EndpointScoring.qll

@@ -205,14 +205,8 @@ class EndpointScoringResults extends ScoringResults {
         exists(float sinkScore |
           ModelScoring::endpointScores(sink, getCfg().getASinkEndpointType().getEncoding(),
             sinkScore) and
-          // Include the endpoint if (a) the query endpoint type scores higher than all other
-          // endpoint types, or (b) the query endpoint type scores at least
-          // 0.5 - (getCfg().getScoreCutoff() / 2).
-          sinkScore >=
-            [
-              max(float s | ModelScoring::endpointScores(sink, _, s)),
-              0.5 - getCfg().getScoreCutoff() / 2
-            ]
+          // Include the endpoint if the query endpoint type scores higher than a fixed threshold
+          sinkScore >= getCfg().getScoreCutoff()
         )
       )
   }