fix(search): let cancelled live reads stop instead of reporting partial coverage

apps/sim/lib/sim-search/live/application.test.ts

@@ -68,6 +68,7 @@ import {
   readLiveDocument,
   searchLiveKnowledge,
 } from '@/lib/sim-search/live/application'
+import { readDrive } from '@/lib/sim-search/live/google'
 import { NativeSearchError } from '@/lib/sim-search/live/http'
 import { createPolicyVerifier } from '@/lib/sim-search/live/policy'
 import { defaultLiveSearchPolicy } from '@/lib/sim-search/live/policy-schema'
@@ -405,6 +406,62 @@ describe('authorized live retrieval', () => {
     ).rejects.toThrow('Revoked')
     expect(mocks.read).not.toHaveBeenCalled()
   })
+  it.each([
+    ['a stop reason', 'user_stop:test'],
+    ['an AbortError', new DOMException('The operation was aborted.', 'AbortError')],
+    ['a TimeoutError', new DOMException('The operation timed out.', 'TimeoutError')],
+  ])(
+    'rejects a read cancelled with %s instead of returning partial discussion coverage',
+    async (_, reason) => {
+      const search = await searchLiveKnowledge.execute({ principal, input })
+      const controller = new AbortController()
+      mocks.read.mockImplementation((_provider, client) => readDrive(client, 'doc'))
+      mocks.json.mockImplementation(async (path: string) => {
+        if (!path.endsWith('/comments'))
+          return {
+            id: 'doc',
+            name: 'Launch',
+            mimeType: 'application/pdf',
+            webViewLink: document.url,
+          }
+        controller.abort(reason)
+        throw controller.signal.reason
+      })
+      await expect(
+        readLiveDocument.execute({
+          principal,
+          input: {
+            workspaceId: 'workspace',
+            documentId: search.results[0]!.documentId,
+            limit: 1,
+            resultSecretRegistry: new ResolvedSecretTraceRegistry([]),
+            signal: controller.signal,
+          },
+        })
+      ).rejects.toBe(reason)
+    }
+  )
+  it('rejects a read cancelled while its current scope was being verified', async () => {
+    const search = await searchLiveKnowledge.execute({ principal, input })
+    const controller = new AbortController()
+    mocks.service.mockResolvedValueOnce(undefined)
+    mocks.service.mockImplementationOnce(async () => {
+      controller.abort('user_stop:test')
+      return { policy: defaultLiveSearchPolicy(), verify: async () => true, partial: false }
+    })
+    await expect(
+      readLiveDocument.execute({
+        principal,
+        input: {
+          workspaceId: 'workspace',
+          documentId: search.results[0]!.documentId,
+          limit: 1,
+          resultSecretRegistry: new ResolvedSecretTraceRegistry([]),
+          signal: controller.signal,
+        },
+      })
+    ).rejects.toBe('user_stop:test')
+  })
   it('rejects cross-user document references before token resolution', async () => {
     const search = await searchLiveKnowledge.execute({ principal, input })
     const reference = decodeLiveReference(search.results[0].documentId)

apps/sim/lib/sim-search/live/application.ts

@@ -661,7 +661,12 @@ export const readLiveDocument = defineAuthorizedKnowledgeUseCase({
           'Document is outside your organization’s search scope'
         )
       document = await measureSearchStage('live.read', () => session.read(reference, input.filters))
-      if (!(await session.verifyCurrent(document)))
+      /** Readers degrade section failures to warnings, so the signal decides cancellation. */
+      signal.throwIfAborted()
+      const current = await session.verifyCurrent(document)
+      /** A verifier may report a check cut short by cancellation as a normal result. */
+      signal.throwIfAborted()
+      if (!current)
         throw new OrchestrationError(
           'not_found',
           'Document is outside your organization’s search scope'

apps/sim/lib/sim-search/live/discussion-reads.test.ts

@@ -258,6 +258,20 @@ describe('Drive discussion reads', () => {
     expect(content).toMatch(/rate limit/i)
   })
 
+  it('degrades a failed comment request to partial coverage', async () => {
+    const api: NativeClient = {
+      text,
+      async json(path) {
+        if (path.endsWith('/comments'))
+          throw Object.assign(new Error('socket hang up'), { code: 'ECONNRESET' })
+        return DRIVE_FILE
+      },
+    }
+    const { content } = await readDrive(api, 'doc')
+    expect(content).toContain('Original document text')
+    expect(content).toMatch(/incomplete[\s\S]*could not be fully retrieved/i)
+  })
+
   it('detects repeated comment cursors without silently claiming all comments were read', async () => {
     let pages = 0
     const api: NativeClient = {

apps/sim/lib/sim-search/live/discussion.ts

@@ -1,4 +1,4 @@
-import { truncate } from '@sim/utils/string'
+import { truncateAtCodePoint } from '@sim/utils/string'
 import { NativeSearchError } from '@/lib/sim-search/live/http'
 
 /** Leaves room for file content and authorization within one live read's request budget. */
@@ -13,6 +13,7 @@ interface DiscussionPage {
 /**
  * Collects a bounded discussion without representing provider failures or omitted pages as an
  * empty, complete history. The caller puts the warning before the document's first read window.
+ * Cancellation is judged by the caller's signal after the read, not by the error's shape.
  */
 export async function readDiscussionSection(
   label: string,
@@ -29,7 +30,7 @@ export async function readDiscussionSection(
       for (const entry of result.entries) {
         if (!entry) continue
         const remaining = Math.max(0, DISCUSSION_MAX_CHARACTERS - characters - 2)
-        entries.push(truncate(entry, remaining))
+        entries.push(truncateAtCodePoint(entry, remaining, ''))
         characters += Math.min(entry.length, remaining) + 2
         if (entry.length > remaining) {
           omitted = 'exceeded the discussion text limit'

apps/sim/lib/sim-search/live/linear.test.ts

@@ -134,6 +134,12 @@ describe('Linear live search boundary', () => {
   it('rejects malformed search responses and project references', async () => {
     const api = client(async () => ({ data: { searchIssues: {} } }))
     await expect(searchLinear(api, input)).rejects.toThrow('unsupported')
+    const ascending = client(async () => ({
+      data: { searchIssues: { nodes: [issue], pageInfo: { hasNextPage: false } } },
+    }))
+    await expect(
+      searchLinear(ascending, { ...input, filters: { sortBy: 'oldest' } })
+    ).rejects.toThrow('unsupported')
     await expect(
       searchLinear(api, {
         ...input,

apps/sim/lib/sim-search/live/linear.ts

@@ -138,7 +138,10 @@ export async function searchLinear(
     variables
   )
   const result = object(data[field])
-  if (!Array.isArray(result.nodes) || typeof object(result.pageInfo).hasNextPage !== 'boolean')
+  if (
+    !Array.isArray(result.nodes) ||
+    typeof object(result.pageInfo)[backwards ? 'hasPreviousPage' : 'hasNextPage'] !== 'boolean'
+  )
     throw new NativeSearchError(
       'unavailable',
       'Linear search returned an unsupported result format.'

apps/sim/lib/sim-search/live/meeting-content.ts

@@ -1,8 +1,8 @@
-import { truncate } from '@sim/utils/string'
+import { truncateAtCodePoint } from '@sim/utils/string'
 
 /** Coverage notices precede the text so the first paginated read cannot hide a provider cap. */
 export function boundedMeetingContent(content: string, limit = 200_000): string {
   if (content.length <= limit) return content
   const notice = '[Meeting content truncated. Open the original meeting for the remainder.]\n\n'
-  return notice + truncate(content, limit - notice.length, '')
+  return notice + truncateAtCodePoint(content, limit - notice.length, '')
 }

apps/sim/lib/sim-search/live/notion-mcp.test.ts

@@ -153,6 +153,16 @@ describe('Notion live MCP boundary', () => {
     expect(page.partial).toBe(true)
   })
 
+  it('keeps AI search coverage partial when the result omits its search type', async () => {
+    const client: ManagedSearchMcpClient = {
+      call: async (name) =>
+        name === 'notion-get-tool-access'
+          ? { current_tool_access: { ai_search: { status: 'available' } } }
+          : { results: [result] },
+    }
+    expect(await searchNotionMcp(client, input)).toMatchObject({ partial: true })
+  })
+
   it('uses keyword search when AI access needs an upgrade without inventing unknown timestamps', async () => {
     const client: ManagedSearchMcpClient = {
       call: async (name) => {

apps/sim/lib/sim-search/live/notion-mcp.ts

@@ -155,7 +155,9 @@ export async function searchNotionMcp(
   const next = string(result.next_cursor ?? result.nextCursor)
   const nextCursor = cursorKey && next && !clipped ? next : undefined
   const notices = array(result.notices).length > 0
-  const aiSearch = result.type === 'ai_search'
+  const aiSearch =
+    result.type === 'ai_search' ||
+    (tool === 'notion-ai-search' && result.type !== 'workspace_search')
   const hasMore = clipped || result.has_more === true || result.hasMore === true || Boolean(next)
   const cappedWithoutCoverage =
     rows.length >= limit && !next && result.has_more !== false && result.hasMore !== false