From 43efa37c6a473af59ea94a0f18bd3e2ca9fe708f Mon Sep 17 00:00:00 2001
From: mdecimus <11444311+mdecimus@users.noreply.github.com>
Date: Thu, 18 Dec 2025 09:24:22 +0100
Subject: [PATCH] Update spam autolearn behaviour

---
 crates/common/src/config/spamfilter.rs   | 16 ++++++------
 crates/spam-filter/src/analysis/score.rs | 32 +++++++++++++-----------
 2 files changed, 25 insertions(+), 23 deletions(-)

diff --git a/crates/common/src/config/spamfilter.rs b/crates/common/src/config/spamfilter.rs
index 35688b39..2c228a05 100644
--- a/crates/common/src/config/spamfilter.rs
+++ b/crates/common/src/config/spamfilter.rs
@@ -87,8 +87,8 @@ pub struct ClassifierConfig {
     pub min_spam_samples: u64,
     pub auto_learn_reply_ham: bool,
     pub auto_learn_card_is_ham: bool,
-    pub auto_learn_spam_score: f32,
-    pub auto_learn_ham_score: f32,
+    pub auto_learn_spam_trap: bool,
+    pub auto_learn_spam_rbl_count: u32,
     pub hold_samples_for: u64,
     pub train_frequency: Option<u64>,
     pub log_scale: bool,
@@ -489,12 +489,12 @@ impl ClassifierConfig {
             auto_learn_reply_ham: config
                 .property_or_default("spam-filter.trusted-reply.learn", "true")
                 .unwrap_or(true),
-            auto_learn_spam_score: config
-                .property_or_default("spam-filter.classifier.auto-learn.spam-score", "8.0")
-                .unwrap_or(8.0),
-            auto_learn_ham_score: config
-                .property_or_default("spam-filter.classifier.auto-learn.ham-score", "-8.0")
-                .unwrap_or(-8.0),
+            auto_learn_spam_trap: config
+                .property_or_default("spam-filter.classifier.auto-learn.spam-trap", "true")
+                .unwrap_or(true),
+            auto_learn_spam_rbl_count: config
+                .property_or_default("spam-filter.classifier.auto-learn.spam-rbl-count", "2")
+                .unwrap_or(2),
             hold_samples_for: config
                 .property_or_default::<Duration>("spam-filter.classifier.samples.hold-for", "180d")
                 .unwrap_or(Duration::from_secs(180 * 24 * 60 * 60))
diff --git a/crates/spam-filter/src/analysis/score.rs b/crates/spam-filter/src/analysis/score.rs
index 8a96a40b..c1c1bcab 100644
--- a/crates/spam-filter/src/analysis/score.rs
+++ b/crates/spam-filter/src/analysis/score.rs
@@ -56,6 +56,8 @@ impl SpamFilterAnalyzeScore for Server {
         // Calculate final score
         let mut results = vec![];
         let mut header_len = 60;
+        let mut is_spam_trap = false;
+        let mut rbl_count = 0;
 
         for tag in &ctx.result.tags {
             let score = match self.core.spam.lists.scores.get(tag) {
@@ -68,6 +70,11 @@ impl SpamFilterAnalyzeScore for Server {
                 }
                 None | Some(SpamFilterAction::Disabled) => 0.0,
             };
+            if tag == "SPAM_TRAP" {
+                is_spam_trap = true;
+            } else if score > 1.0 && tag.starts_with("RBL_") {
+                rbl_count += 1;
+            }
             ctx.result.score += score;
             header_len += tag.len() + 10;
             if score != 0.0 || !tag.starts_with("X_") {
@@ -145,11 +152,8 @@ impl SpamFilterAnalyzeScore for Server {
                 let _ = write!(&mut headers, "X-Spam-LLM: {category} ({explanation})\r\n",);
             }
 
-            let class = if final_score >= self.core.spam.scores.spam_threshold {
-                "spam"
-            } else {
-                "ham"
-            };
+            let is_spam = final_score >= self.core.spam.scores.spam_threshold;
+            let class = if is_spam { "spam" } else { "ham" };
 
             if avg_confidence != 0.0 {
                 let _ = write!(
@@ -163,18 +167,16 @@ impl SpamFilterAnalyzeScore for Server {
                 );
             }
 
-            // Autolearn
+            // Autolearn SPAM
             let mut train_spam = None;
-            let config = self.core.spam.classifier.as_ref().unwrap();
-            if config.auto_learn_spam_score > 0.0 && final_score >= config.auto_learn_spam_score {
-                if !ctx.result.has_tag("PROB_SPAM_HIGH") {
-                    train_spam = Some(true);
-                }
-            } else if config.auto_learn_ham_score < 0.0
-                && final_score <= config.auto_learn_ham_score
-                && !ctx.result.has_tag("PROB_HAM_HIGH")
+            if is_spam
+                && self.core.spam.classifier.as_ref().is_some_and(|c| {
+                    (c.auto_learn_spam_trap && is_spam_trap)
+                        || (c.auto_learn_spam_rbl_count > 0
+                            && rbl_count >= c.auto_learn_spam_rbl_count)
+                })
             {
-                train_spam = Some(false);
+                train_spam = Some(true);
             }
 
             SpamFilterAction::Allow(SpamFilterScore {