fix(planning): the plan gate's default drops to 0.80, where reworded plans actually score (#4079)

Measured live after deploy: three plans reworded from existing active
milestones scored 0.83-0.87 against the milestone they restated, and the
nearest distinct plans 0.72-0.77. At the inherited 0.90 the semantic arm
matched nothing, so only an identical title was caught. 0.80 sits in the gap.
The Settings form's default moves with it (test_settings_defaults_agree).

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01821k5B3Ysecp9fNYs92Kuy

frontend/src/views/SettingsView.vue

@@ -105,7 +105,7 @@ const kbDupThresholdSnippet = ref("0.82");
 const kbDupThresholdNote = ref("0.93");
 const kbDupThresholdTask = ref("0.93");
 // PLAN_MATCH_DEFAULT_THRESHOLD in services/dedup.py.
-const kbPlanMatchThreshold = ref("0.90");
+const kbPlanMatchThreshold = ref("0.80");
 const savingKbInject = ref(false);
 const kbInjectSaved = ref(false);
 
@@ -159,7 +159,7 @@ async function saveKbInject() {
   const dupTask = Math.min(1, Math.max(0, Number(kbDupThresholdTask.value) || 0.93));
   // Same `|| default` guard: a floor of 0 would hand back an existing plan
   // for every new one, and no plan could be started without force.
-  const planT = Math.min(1, Math.max(0, Number(kbPlanMatchThreshold.value) || 0.9));
+  const planT = Math.min(1, Math.max(0, Number(kbPlanMatchThreshold.value) || 0.8));
   // Same `|| default` reasoning: falling back to 0 would surface every
   // snippet in the corpus on every edit, which is the failure this knob fixes.
   const wpT = Math.min(1, Math.max(0, Number(kbWritePathThreshold.value) || 0.68));

src/scribe/services/dedup.py

@@ -739,16 +739,22 @@ async def find_duplicate_rule(
 # project is exactly what adding steps to its existing plan needs, so a caller
 # who reaches this gate can act on whatever it returns.
 #
-# Its own threshold, as a setting (rule 25). The note gate's 0.90 is where it
+# Its own threshold, as a setting (rule 25), and lower than the note gate's
-# starts, because it asks the same question ("the same thing, reworded") with
+# 0.90 because plan documents are shaped differently: a milestone is embedded
-# the same embedder. Plan documents are shaped differently, though: a milestone
+# as title, description and plan (embeddings.milestone_document), while the
-# is embedded as title, description and plan (embeddings.milestone_document),
+# candidate usually has no description and carries its steps instead, so even
-# while the candidate usually has no description and carries its steps
+# a faithful rewording never scores like a copy.
-# instead. That difference has not been measured, and a gate that blocks on
+#
-# noise teaches sessions to pass force=true every time, which is worse than no
+# Measured after the first deploy (2026-09-15, bge-small-en-v1.5, #4079): three
-# gate. So the default is conservative and the operator can lower it.
+# plans reworded from existing active milestones scored 0.83, 0.85 and 0.87
+# against the milestone they restated; the nearest DIFFERENT plan for each, and
+# a distinct plan on a neighbouring topic, scored 0.72-0.77. At 0.90 the
+# semantic arm matched nothing, leaving only the title arm. 0.80 sits in the
+# gap. A gate that blocks on noise teaches sessions to pass force=true every
+# time, which is why it is not lower; four samples is thin, which is why the
+# operator can move it. Retune alongside the embedder, not the corpus.
 PLAN_MATCH_THRESHOLD_KEY = "kb_plan_match_threshold"
-PLAN_MATCH_DEFAULT_THRESHOLD = 0.90
+PLAN_MATCH_DEFAULT_THRESHOLD = 0.80
 
 
 async def get_plan_match_threshold(user_id: int) -> float: