improvement(knowledge): score the page on the original vectors, resolve a search's context once, and read identities off the walk (#8100)

* improvement(knowledge): score the vector page on the original vectors and pin the clock in the PDF chunk test

* improvement(knowledge): score both pages on the original vectors and say so where the walk is scored

* improvement(knowledge): resolve a search's context once and carry candidate identities off the walk

* improvement(knowledge): one scoped search shape, page and pool cleanups, admission ahead of the embedding beside the scope reads

* improvement(knowledge): embed only after every prerequisite holds, refuse a contradicting owner, count the shared fill read

* improvement(knowledge): reset the projection-fill memo per integration iteration

Author: waleedlatif1

apps/sim/lib/billing/core/billing-attribution.ts

@@ -742,13 +742,16 @@ export async function resolveBillingAttribution({
 
 /** The organization payer is independent of the person making the request. */
 export async function resolveOrganizationBillingPayer(organizationId: string) {
-  const [owner] = await db
-    .select({ userId: member.userId })
-    .from(member)
-    .where(and(eq(member.organizationId, organizationId), eq(member.role, 'owner')))
-    .limit(1)
+  /** The owner and the subscription are independent reads; neither waits on the other. */
+  const [[owner], payerSubscription] = await Promise.all([
+    db
+      .select({ userId: member.userId })
+      .from(member)
+      .where(and(eq(member.organizationId, organizationId), eq(member.role, 'owner')))
+      .limit(1),
+    getOrganizationSubscription(organizationId, { onError: 'throw' }),
+  ])
   if (!owner) throw new Error('Organization billing owner is unavailable')
-  const payerSubscription = await getOrganizationSubscription(organizationId, { onError: 'throw' })
   if (payerSubscription && payerSubscription.referenceId !== organizationId)
     throw new Error('Organization subscription belongs to a different payer')
   return { organizationId, billedAccountUserId: owner.userId, payerSubscription }

apps/sim/lib/knowledge/__integration__/kb-block-search.integration.ts

@@ -11,6 +11,7 @@ import {
 } from '@/lib/knowledge/__integration__/seed-source-access-fixture'
 import { createKnowledgeAccessProvider } from '@/lib/knowledge/access/scope'
 import {
+  forgetProjectionFilled,
   resolvePermittedDocuments,
   retrieveKnowledgeSearch,
   VECTOR_PROBE_DOCUMENT_LIMIT,
@@ -92,6 +93,8 @@ describe('API-key KB block fan-out', () => {
   it.each([false, true])(
     'completes 18 concurrent KB searches with access checks intact (tag filter: %s)',
     async (withTags) => {
+      /** The projection-fill memo outlives an iteration; each one must read it once, like a cold process. */
+      forgetProjectionFilled()
       const previousDebug = db.$client.options.debug
       const statements: string[] = []
       db.$client.options.debug = (_connection, query) => {
@@ -132,18 +135,20 @@ describe('API-key KB block fan-out', () => {
           statements.filter((query) => query.includes(fragment))
         /**
          * Every statement runs under the leg's deadline: the candidate search applies it with the
-         * scan settings in one statement, and the probe, the exact ranking, the rerank and
-         * hydration each open with one of their own.
+         * scan settings in one statement, and the probe, the exact ranking and hydration each
+         * open with one of their own. The projection-fill read is shared by the searches that
+         * miss its memo together, so it appears once.
          */
-        expect(matching('statement_timeout')).toHaveLength(bases.length * 5)
+        expect(matching('statement_timeout')).toHaveLength(bases.length * 4 + 1)
         /**
          * A scope this small leaves the bounded traversal short of its candidate limit, so every
          * search probes once and rescues once — never a widening retry loop.
          */
         expect(matching('hnsw.iterative_scan')).toHaveLength(bases.length)
         expect(matching('AS visible')).toHaveLength(bases.length)
         expect(matching(') + 0 LIMIT')).toHaveLength(bases.length)
-        expect(matching('"embedding_search"."id" = ANY(')).toHaveLength(bases.length)
+        /** The walk carries each candidate's identities, so a filled projection reads no page. */
+        expect(matching('"embedding_search"."id" = ANY(')).toHaveLength(0)
         /** The probe enumerates visible documents and reports saturation; it never ranks them. */
         expect(
           statements.filter(

apps/sim/lib/knowledge/access/availability.ts

@@ -64,7 +64,10 @@ export async function resolveKnowledgeAccessAvailability(
     throw new Error('Knowledge access requires one resource owner')
   /** A caller that brings its own billing snapshot is answered from that snapshot, uncached. */
   if (context.ownerBilling) return readKnowledgeAccessAvailability(context)
-  const key = `${context.organizationId ?? ''}|${context.workspaceId ?? ''}|${context.userId ?? ''}`
+  /** An organization's answer depends on the organization alone; a workspace's on its viewer too. */
+  const key = context.organizationId
+    ? `${context.organizationId}||`
+    : `|${context.workspaceId ?? ''}|${context.userId ?? ''}`
   const availability = await availabilityCache.fetch(key, { context })
   if (!availability) throw new Error('Knowledge access availability could not be resolved')
   return availability
@@ -89,14 +92,14 @@ async function readKnowledgeAccessAvailability(
     return { sourceMirrored: false, memberScoped: false }
   }
   if (context.organizationId) {
-    return {
-      sourceMirrored:
-        !isHosted || (await isOrganizationOnEnterprisePlan(context.organizationId, 'throw')),
-      memberScoped: await isScopedCredentialGroupsAvailable({
+    const [enterprise, memberScoped] = await Promise.all([
+      isHosted ? isOrganizationOnEnterprisePlan(context.organizationId, 'throw') : true,
+      isScopedCredentialGroupsAvailable({
         kind: 'organization',
         organizationId: context.organizationId,
       }),
-    }
+    ])
+    return { sourceMirrored: enterprise, memberScoped }
   }
   if (!context.workspaceId) throw new Error('Knowledge access requires a resource owner')
   const ownerBilling =

apps/sim/lib/knowledge/application/contexts.ts

@@ -18,15 +18,12 @@ import {
 } from '@/lib/knowledge/connectors/service'
 import type { ActiveKnowledgeDocument } from '@/lib/knowledge/documents/service'
 import { getKnowledgeDocument, getKnowledgeDocumentById } from '@/lib/knowledge/documents/service'
+import type { ActiveKnowledgeBaseReference } from '@/lib/knowledge/knowledge-base-reference'
 import {
   getRestorableKnowledgeBase,
   type RestorableKnowledgeBase,
 } from '@/lib/knowledge/orchestration/restore'
-import {
-  type ActiveKnowledgeBaseReference,
-  getActiveKnowledgeBaseReference,
-  getKnowledgeBaseById,
-} from '@/lib/knowledge/service'
+import { getActiveKnowledgeBaseReference, getKnowledgeBaseById } from '@/lib/knowledge/service'
 import { getTagDefinitionById } from '@/lib/knowledge/tags/service'
 import type { DocumentTagDefinition } from '@/lib/knowledge/tags/types'
 import type { KnowledgeBaseWithCounts } from '@/lib/knowledge/types'