You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: src/mail_sovereignty/classifier.py
+15-3Lines changed: 15 additions & 3 deletions
Original file line number
Diff line number
Diff line change
@@ -84,6 +84,18 @@ def _rule_confidence(
84
84
85
85
Rule chain (ordered by specificity, first match wins):
86
86
87
+
```
88
+
- MX only 80% (Most likely unfinished config)
89
+
- MX + SPF 90% (Managed setup)
90
+
- AD + SPF 90% (Security GW upfront)
91
+
- MX + SPF + AD 95% (Cloud Setup)
92
+
93
+
AD Cases sind natürlich nur MS-relevant. Mit den restlichen Indikatoren könntest du bspw. dann die restlichen 5-10% auffüllen. Die %-te kann man natürlich auch anders aufteilen.
94
+
```
95
+
96
+
// messen wie viel Rule gebraucht wird, evtl R5-R8 nicht unbeding relevant
97
+
98
+
87
99
==== ========================== ====
88
100
Rule Condition Base
89
101
==== ========================== ====
@@ -93,7 +105,7 @@ def _rule_confidence(
93
105
R4 MX ∧ TENANT 0.85
94
106
R5 SPF ∧ TENANT 0.80
95
107
R6 SPF ∧ GW 0.70
96
-
R7 MX 0.60
108
+
R7 MX 0.60 // eher auf 80%
97
109
R8 SPF 0.50
98
110
R9 else 0.40
99
111
==== ========================== ====
@@ -143,7 +155,7 @@ def _independent_confidence(
143
155
144
156
* MX + SPF present → 0.90
145
157
* MX only → 0.60
146
-
* Any evidence → 0.50
158
+
* No MX, secondary evidence only → 0.20
147
159
* Nothing → 0.0
148
160
149
161
After the base is selected, each distinct signal kind in *evidence*
0 commit comments