@@ -345,6 +345,74 @@ describe('SearchModal', () => {
     }
   })
 
+  it('keeps a block above its same-name trigger for the exact-name query', async () => {
+    const Icon = () => null
+    const original = { ...mockSearchState.data }
+    mockSearchState.data = {
+      ...mockSearchState.data,
+      tools: [
+        {
+          id: 'gmail',
+          name: 'Gmail',
+          icon: Icon,
+          bgColor: '#E8453C',
+          type: 'gmail',
+          searchValue: 'gmail gmail',
+        },
+      ],
+      triggers: [{ id: 'gmail', name: 'Gmail', icon: Icon, bgColor: '#E8453C', type: 'gmail' }],
+    }
+
+    try {
+      await act(async () => {
+        root.render(<SearchModal open onOpenChange={vi.fn()} pageContext='workflow' />)
+      })
+
+      await enterSearchQuery('gmail')
+      const rows = Array.from(document.querySelectorAll<HTMLElement>('[cmdk-item]')).map(
+        (el) => el.textContent ?? ''
+      )
+      expect(rows[0]).toContain('Gmail')
+      expect(rows[0]).not.toContain('Gmail Trigger')
+      expect(rows[1]).toContain('Gmail Trigger')
+    } finally {
+      mockSearchState.data = original
+    }
+  })
+
+  it('ranks prefix-matched rows above actions that only contain the letter mid-word', async () => {
+    const Icon = () => null
+    const original = { ...mockSearchState.data }
+    mockSearchState.data = {
+      ...mockSearchState.data,
+      tools: [
+        {
+          id: 'hex',
+          name: 'Hex',
+          icon: Icon,
+          bgColor: '#111',
+          type: 'hex',
+          searchValue: 'hex hex',
+        },
+      ],
+    }
+
+    try {
+      await act(async () => {
+        root.render(<SearchModal open onOpenChange={vi.fn()} pageContext='workflow' />)
+      })
+
+      await enterSearchQuery('h')
+      const rows = Array.from(document.querySelectorAll<HTMLElement>('[cmdk-item]')).map(
+        (el) => el.textContent ?? ''
+      )
+      expect(rows[0]).toContain('Hex')
+      expect(rows.findIndex((row) => row.includes('New chat'))).toBeGreaterThan(0)
+    } finally {
+      mockSearchState.data = original
+    }
+  })
+
   it('puts the workflow verb actions first for their bare-verb queries', async () => {
     const Icon = () => null
     const original = { ...mockSearchState.data }
 
@@ -970,6 +970,9 @@ function SearchModalContent({
       ...(pageContext ? rankActionGroup(actionsByGroup.page, 'Actions') : []),
       ...rankActionGroup(actionsByGroup.sim, 'Sim'),
     ]
+    const blockNames = new Set(
+      [...availableBlocks, ...availableTools].map((item) => item.name.toLowerCase())
+    )
 
     return {
       actions: rankedActions.map(({ item, score }) => ({ section: 'actions', item, score })),
@@ -988,8 +991,14 @@ function SearchModalContent({
         section: 'triggers',
         item,
         /* The display rename ("Start" → "Start Trigger") costs the exact-name
-           bonus, so a query that IS the trigger's name ranks it like a page row. */
-        score: item.baseName.toLowerCase() === query.toLowerCase() ? PAGE_MATCH_TIER : score,
+           bonus, so a query that IS the trigger's name ranks it like a page row
+           — unless a block shares that name (Gmail, Slack). Then the query names
+           the block first, and the lift would leapfrog its exact-name match. */
+        score:
+          item.baseName.toLowerCase() === query.toLowerCase() &&
+          !blockNames.has(item.baseName.toLowerCase())
+            ? PAGE_MATCH_TIER
+            : score,
       })),
       tools: rank(
         'tools',
 
@@ -100,6 +100,34 @@ describe('getGlobalSearchResults', () => {
     ).toEqual(['new-chat-action', 'new-chat-result'])
   })
 
+  it('keeps a mid-word-matched action below word-start entity matches', () => {
+    const action = {
+      id: 'create-folder',
+      name: 'Create folder',
+      icon: () => null,
+      context: 'global' as const,
+      run: () => {},
+    }
+    const [actionMatch] = scoreActions([action], 'a')
+    const [blockMatch] = scoreAndSort([{ name: 'Airtable' }], (item) => item.name, 'a')
+
+    expect(actionMatch.score).toBeLessThan(blockMatch.score)
+  })
+
+  it('still biases a word-start action match above entity name matches', () => {
+    const action = {
+      id: 'create-workflow',
+      name: 'Create workflow',
+      icon: () => null,
+      context: 'global' as const,
+      run: () => {},
+    }
+    const [actionMatch] = scoreActions([action], 'w')
+    const [blockMatch] = scoreAndSort([{ name: 'Webhook' }], (item) => item.name, 'w')
+
+    expect(actionMatch.score).toBeGreaterThan(blockMatch.score)
+  })
+
   it('breaks identical visible-name matches by the original section order', () => {
     const workflow = { id: 'new-chat-workflow', name: 'New chat', href: '/new-chat-workflow' }
     const chat = { id: 'new-chat-result', name: 'New chat', href: '/new-chat-result' }
 
@@ -578,19 +578,32 @@ export function scoreSectionItems<T>(
 }
 
 /**
- * Rank offset added to every matched action. Actions are the palette's few
+ * Rank offset added to a matched action. Actions are the palette's few
  * runnable verbs, so a matched action outranks entity rows of the same match
  * quality — a name-matched action beats name-matched entities, a
  * keyword-matched action beats other secondary-text matches — while the
  * half-tier offset deliberately cannot bridge into the next tier up
- * ({@link SECTION_MATCH_TIER}, {@link PAGE_MATCH_TIER}).
+ * ({@link SECTION_MATCH_TIER}, {@link PAGE_MATCH_TIER}). A name hit that
+ * starts mid-word ("h" in "New chat") is NOT the same quality as the
+ * word-start matches the offset would leapfrog, so it forgoes the bias.
  */
 export const ACTION_MATCH_BIAS = 500_000
 
+/**
+ * Whether a match begins where a word begins — the string start, right after a
+ * separator, or at a camelCase hump. The empty query (no positions) counts as
+ * a word start.
+ */
+function isWordStartMatch(text: string, positions: readonly number[]): boolean {
+  if (positions.length === 0) return true
+  return isHardBoundary(text.toLowerCase(), positions[0]) || isCamelBoundary(text, positions[0])
+}
+
 /**
  * Scores actions by visible name before falling back to their keywords.
- * Every match is lifted by {@link ACTION_MATCH_BIAS}; a query listed in the
- * action's `exactQueries` ranks it like a page row instead.
+ * Word-start matches are lifted by {@link ACTION_MATCH_BIAS}; a mid-word name
+ * hit keeps its honest score so word-start entity matches outrank it; a query
+ * listed in the action's `exactQueries` ranks it like a page row instead.
  */
 export function scoreActions(
   actions: ActionItem[],
@@ -606,10 +619,15 @@ export function scoreActions(
     search,
     (action) => `${toSearchToken(action.name)} ${action.keywords ?? ''}`,
     maxResults
-  ).map(({ item, score }) => ({
-    item,
-    score: item.exactQueries?.includes(query) ? PAGE_MATCH_TIER : score + ACTION_MATCH_BIAS,
-  }))
+  ).map(({ item, score }) => {
+    if (item.exactQueries?.includes(query)) return { item, score: PAGE_MATCH_TIER }
+    /* Section-lifted rows (the query IS the group label) keep the bias
+       wholesale — only plain name-tier scores are quality-checked. */
+    const byName = fuzzyMatch(item.name, query)
+    const midWordNameMatch =
+      score < SECTION_MATCH_TIER && byName.matched && !isWordStartMatch(item.name, byName.positions)
+    return { item, score: midWordNameMatch ? score : score + ACTION_MATCH_BIAS }
+  })
 }
 
 /**