diff --git a/src/mind/mod.rs b/src/mind/mod.rs
index 81bcb09..c2cb365 100644
--- a/src/mind/mod.rs
+++ b/src/mind/mod.rs
@@ -320,7 +320,7 @@ impl MindState {
 /// Background task completion events.
 enum BgEvent {
     ScoringDone,
-    FinetuneCandidates(Vec<learn::FinetuneCandidate>),
+    FinetuneCandidate(learn::FinetuneCandidate),
 }
 
 // --- Mind: cognitive state machine ---
@@ -656,7 +656,12 @@ impl Mind {
     /// once this runs continuously, we'll just train whatever lands at full
     /// context without filtering.
     pub fn start_finetune_scoring(&self) {
-        let threshold = self.shared.lock().unwrap().learn_threshold;
+        let threshold = {
+            let mut s = self.shared.lock().unwrap();
+            // Clear the previous run's candidates so this run's stream in fresh.
+            s.finetune_candidates.clear();
+            s.learn_threshold
+        };
 
         let agent = self.agent.clone();
         let bg_tx = self.bg_tx.clone();
@@ -678,12 +683,12 @@ impl Mind {
 
             activity.update(format!("finetune: scoring {} responses...", responses_considered)).await;
 
+            let bg_tx_cb = bg_tx.clone();
             let stats = match learn::score_finetune_candidates(
-                &context, score_count, &client, threshold,
+                &context, score_count, &client, threshold, &activity,
+                |c| { let _ = bg_tx_cb.send(BgEvent::FinetuneCandidate(c)); },
             ).await {
-                Ok((candidates, max_div)) => {
-                    let above_threshold = candidates.len();
-                    let _ = bg_tx.send(BgEvent::FinetuneCandidates(candidates));
+                Ok((above_threshold, max_div)) => {
                     FinetuneScoringStats {
                         responses_considered,
                         above_threshold,
@@ -801,8 +806,8 @@ impl Mind {
                         BgEvent::ScoringDone => {
                             self.shared.lock().unwrap().scoring_in_flight = false;
                         }
-                        BgEvent::FinetuneCandidates(candidates) => {
-                            self.shared.lock().unwrap().finetune_candidates = candidates;
+                        BgEvent::FinetuneCandidate(c) => {
+                            self.shared.lock().unwrap().finetune_candidates.push(c);
                         }
                     }
                 }
diff --git a/src/subconscious/learn.rs b/src/subconscious/learn.rs
index 3c12efc..2424fa5 100644
--- a/src/subconscious/learn.rs
+++ b/src/subconscious/learn.rs
@@ -492,22 +492,28 @@ pub struct FinetuneCandidate {
 
 /// Score and enrich finetune candidates with full context.
 ///
-/// Returns (candidates, max_divergence) - candidates ready for review with
-/// context/continuation token IDs, and the highest divergence seen.
+/// Candidates are delivered via `on_candidate` one-at-a-time as they become
+/// ready: scoring happens once (one /score call), then for each candidate
+/// that passes the threshold we optionally generate an alternate response
+/// and then emit it. The activity status is updated during the alternate
+/// phase so the UI doesn't look stuck.
+///
+/// Returns (count_above_threshold, max_divergence).
 pub async fn score_finetune_candidates(
     context: &ContextState,
     count: usize,
     client: &ApiClient,
     min_divergence: f64,
-) -> anyhow::Result<(Vec<FinetuneCandidate>, f64)> {
+    activity: &crate::agent::ActivityGuard,
+    mut on_candidate: impl FnMut(FinetuneCandidate),
+) -> anyhow::Result<(usize, f64)> {
     let scores = score_finetune(context, count, client).await?;
 
     let max_divergence = scores.iter().map(|(_, d)| *d).fold(0.0f64, f64::max);
 
     let entries = context.conversation();
-    let mut candidates = Vec::new();
-
     let trained = load_trained();
+    let mut candidates: Vec<FinetuneCandidate> = Vec::new();
 
     for (entry_idx, divergence) in scores {
         if divergence < min_divergence {
@@ -522,7 +528,7 @@ pub async fn score_finetune_candidates(
             continue;
         }
 
-        // Extract response text
+        // Extract response text.
         let response_text = match node {
             AstNode::Branch { children, .. } => {
                 children.iter()
@@ -536,7 +542,7 @@ pub async fn score_finetune_candidates(
             _ => continue,
         };
 
-        // Build token IDs: context = everything before response, continuation = response
+        // Build token IDs: context = everything before response, continuation = response.
         let (context_ids, _) = build_token_ids(context, 0..entry_idx, Filter::None);
         let continuation_ids: Vec<u32> = node.token_ids().into_iter().collect();
 
@@ -551,17 +557,23 @@ pub async fn score_finetune_candidates(
         });
     }
 
-    // Generate alternates if enabled
-    if alternates_enabled() && !candidates.is_empty() {
-        for candidate in &mut candidates {
+    let total = candidates.len();
+    let gen_alternates = alternates_enabled() && total > 0;
+
+    for (i, mut candidate) in candidates.into_iter().enumerate() {
+        if gen_alternates {
+            activity.update(
+                format!("finetune: generating alternate {}/{}", i + 1, total)
+            ).await;
             match generate_alternate(context, candidate.entry_idx, client).await {
                 Ok(text) => candidate.alternate_text = Some(text),
                 Err(e) => dbglog!("[finetune] alternate generation failed: {:#}", e),
             }
         }
+        on_candidate(candidate);
     }
 
-    Ok((candidates, max_divergence))
+    Ok((total, max_divergence))
 }
 
 /// Generate what the model would say without memories for a given entry.