fix(reads): the read gate is per-family; a write carrying unchanged data never re-embeds

Two cures from the pair's first production adoption, both measured live.

THE READ GATE IS PER-FAMILY. The report-driven gate refused on ANY
provider's not-ready verdict at every read choke point — so a pure
metadata find({ where }) was refused because the VECTOR leg was not
serving, and a deployment's badge reads returned errors for a verdict that
had nothing to do with them. A read may only be refused by the family it
actually consults: metadata reads by the metadata leg (plus graph for a
`connected` filter), vector search by the vector leg, traversal by the
graph leg. Callers name what they need; the existing narration-once-per-
generation and typed-refusal laws are unchanged within a family.

NO RE-EMBED ON UNCHANGED DATA. update() — and its transact() planner —
treated any write that carried `data` as a data change: with
deferEmbedding it queued a landing, and the worker re-embedded and re-landed
a vector for content that had not changed. A host heartbeat re-writing an
unchanged row every few seconds therefore fed a live index-row loop on a
production store. A write carrying the row's current data (structural
compare, key order normalized) is now not a data change: no re-embed, no
deferred landing, no vector rewrite; the metadata write itself still
commits. A real change re-embeds exactly as before.

Pinned in tests/integration/read-gate-scope-and-no-reembed.test.ts — both
pins red-proved against the unfixed code with the production shapes
verbatim. Two health-gate pins that encoded the old brain-global scope are
re-pointed to the family their reads consult.

src/brainy.ts

@@ -3598,7 +3598,15 @@ export class Brainy<T = any> implements BrainyInterface<T> {
       // re-embed below — a stale vector left behind with no path to ever
       // correct itself (a quiet loss, not the deferred-but-eventually-
       // correct flicker the deferEmbedding contract promises).
-      const hasNewData = params.data !== undefined && params.data !== null
+      const rawHasNewData = params.data !== undefined && params.data !== null
+      // NO RE-EMBED ON UNCHANGED DATA: a write carrying the row's CURRENT data
+      // is not a data change — no re-embed, no deferred landing, no vector
+      // rewrite. A host heartbeat re-writing an unchanged row every few
+      // seconds fed a live index-row loop on a production store (each
+      // "change" landed a vector); the amplifier dies here regardless of how
+      // often the host writes.
+      const dataUnchanged = rawHasNewData && Brainy.sameEntityData(params.data, existing.data)
+      const hasNewData = rawHasNewData && !dataUnchanged
       // MT5 deferred re-embedding: the OLD vector keeps serving semantic
       // search — stale-but-present, never absent (the flicker law) — until
       // the background worker embeds the new data and swaps it atomically.
@@ -4209,7 +4217,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     // writes while every non-find() read served empty from a not-ready
     // provider for 15 minutes. A CHECK only — it never builds; throws a typed
     // NotReady error if a provider's health report says it isn't serving.
-    this.ensureIndexesLoaded()
+    this.ensureIndexesLoaded(['graph'])
     const entityInt = this.graphEntityInt(uuid)
     if (entityInt === undefined) return []
     const neighborInts = await this.graphIndex.getNeighbors(entityInt, options)
@@ -6803,7 +6811,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     // READ-SURFACE READINESS GATE (see filterIdsBelted): a CHECK only — it
     // never builds. open() already brought every provider to serving before
     // init() returned; this throws a typed NotReady error if one isn't.
-    this.ensureIndexesLoaded()
+    this.ensureIndexesLoaded(['metadata'])
 
     // Loudly flag a degraded derived index (failed init rebuild, or an
     // adopt-forward degraded commit) so a partial result is never mistaken for
@@ -6815,6 +6823,13 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     let params: FindParams<T> =
       typeof query === 'string' ? await this.parseNaturalQuery(query) : query
 
+    // The vector and graph legs gate only the finds that consult them.
+    const consultsVector = Boolean(
+      (params.query && params.query.trim() !== '') || params.vector || params.near
+    )
+    if (consultsVector) this.ensureIndexesLoaded(['vector'])
+    if (params.connected) this.ensureIndexesLoaded(['graph'])
+
     // Id normalization (8.0): resolve the graph-traversal anchor id(s) so a
     // caller may constrain by natural key. Each maps to the canonical UUID
     // add() stored; real UUIDs pass through. Done once here so every downstream
@@ -10663,7 +10678,10 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     // content (see the identical hasNewData in update()); a plain truthy
     // check would silently skip re-embedding an emptied value and leave a
     // stale vector with no path to ever correct itself.
-    const hasNewData = params.data !== undefined && params.data !== null
+    const rawHasNewData = params.data !== undefined && params.data !== null
+    // No re-embed on unchanged data — the transact() mirror of update()'s rule.
+    const dataUnchanged = rawHasNewData && Brainy.sameEntityData(params.data, existing.data)
+    const hasNewData = rawHasNewData && !dataUnchanged
     let vector = existing.vector
     if (params.vector) {
       if (this.dimensions && params.vector.length !== this.dimensions) {
@@ -12158,7 +12176,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     // writes while every non-find() read served empty from a not-ready
     // provider for 15 minutes. A CHECK only — it never builds; throws a typed
     // NotReady error if a provider's health report says it isn't serving.
-    this.ensureIndexesLoaded()
+    this.ensureIndexesLoaded(['metadata'])
     try {
       return await this.metadataIndex.getIdsForFilter(filter, opts)
     } catch (err) {
@@ -14654,7 +14672,7 @@ export class Brainy<T = any> implements BrainyInterface<T> {
     // writes while every non-find() read served empty from a not-ready
     // provider for 15 minutes. A CHECK only — it never builds; throws a typed
     // NotReady error if a provider's health report says it isn't serving.
-    this.ensureIndexesLoaded()
+    this.ensureIndexesLoaded(['graph'])
     // 8.0 BigInt boundary: unmapped node → no relations.
     const nodeInt = this.graphEntityInt(nodeId)
     if (nodeInt === undefined) return []
@@ -16543,12 +16561,47 @@ export class Brainy<T = any> implements BrainyInterface<T> {
    * `prodLog.warn` ONCE per (provider, `report.generation`) — never once per
    * read — before any throw decision is made.
    */
-  private ensureIndexesLoaded(): void {
+  /**
-    const providers: ReadonlyArray<readonly [string, unknown, new (message: string, originalError?: Error) => BrainyError]> = [
+   * @description Whether two entity `data` payloads are the same content —
+   * the "no re-embed on unchanged data" comparison. Primitives compare by
+   * value; objects compare structurally with key order normalized.
+   * @param a - The incoming data.
+   * @param b - The stored data.
+   * @returns `true` when the content is identical.
+   */
+  private static sameEntityData(a: unknown, b: unknown): boolean {
+    if (a === b) return true
+    if (a === null || b === null || typeof a !== typeof b) return false
+    if (typeof a !== 'object') return false
+    const stable = (v: unknown): string =>
+      JSON.stringify(v, (_k, val) =>
+        val && typeof val === 'object' && !Array.isArray(val)
+          ? Object.keys(val as Record<string, unknown>).sort().reduce((o, k) => {
+              ;(o as Record<string, unknown>)[k] = (val as Record<string, unknown>)[k]
+              return o
+            }, {} as Record<string, unknown>)
+          : val
+      )
+    try { return stable(a) === stable(b) } catch { return false }
+  }
+
+  private ensureIndexesLoaded(
+    families: ReadonlyArray<'vector' | 'metadata' | 'graph'> = ['vector', 'metadata', 'graph']
+  ): void {
+    // PER-FAMILY SCOPE. This gate used to refuse on ANY provider's not-ready
+    // verdict at every read choke point — so a pure metadata find({where})
+    // was refused because the VECTOR leg was not serving; a production
+    // deployment's badge reads returned 500s for exactly that reason on the
+    // pair's first adoption. A read may only be refused by the family it
+    // actually consults: metadata reads by the metadata leg (+ graph for a
+    // `connected` filter), vector search by the vector leg, traversal by the
+    // graph leg. Callers name what they need.
+    const all: ReadonlyArray<readonly ['vector' | 'metadata' | 'graph', unknown, new (message: string, originalError?: Error) => BrainyError]> = [
       ['vector', this.index, VectorIndexNotReadyError],
       ['metadata', this.metadataIndex, MetadataIndexNotReadyError],
       ['graph', this.graphIndex, GraphIndexNotReadyError]
     ]
+    const providers = all.filter(([name]) => families.includes(name))
 
     for (const [name, provider, ErrorClass] of providers) {
       // Migration LOCK (#18) deference: a migrating provider owns its own

tests/integration/health-gate.test.ts

@@ -186,6 +186,8 @@ describe('health gate (b) — unledgered is unknown: never blocks a serving prov
 })
 
 describe('health gate (c) — degraded-but-serving narrates once per generation', () => {
+  // PER-FAMILY LAW (10.4.1): a metadata find() consults the METADATA leg only — the
+  // degraded report lives on the family the read actually consults.
   it('a heal:"repair" failure serves; narrates once per generation, twice across a generation bump', async () => {
     const brain = new Brainy(createTestConfig({ silent: true }))
     await brain.init()
@@ -195,7 +197,7 @@ describe('health gate (c) — degraded-but-serving narrates once per generation'
 
     const internals = internalsOf(brain)
     let generation = 1
-    internals.index.healthReport = () =>
+    internals.metadataIndex.healthReport = () =>
       healthReport({
         provider: 'vector',
         serving: true,
@@ -216,7 +218,7 @@ describe('health gate (c) — degraded-but-serving narrates once per generation'
     await expect(brain.find({ where: { team: 'atlas' } })).resolves.toHaveLength(1)
     expect(countNarrations()).toBe(2) // generation bumped — a second narration
 
-    delete internals.index.healthReport
+    delete internals.metadataIndex.healthReport
   })
 })
 
@@ -332,6 +334,7 @@ describe('health gate (f) — the ceremony door: explicit rebuild bypasses invar
 })
 
 describe('health gate (g) — a throwing healthReport() is a contract violation, never read as healthy', () => {
+  // PER-FAMILY LAW (10.4.1): the throwing report sits on the family the read consults.
   it('healthReport() that throws refuses loudly with the typed NotReady error naming the throw', async () => {
     const brain = new Brainy(createTestConfig({ silent: true }))
     await brain.init()
@@ -340,13 +343,13 @@ describe('health gate (g) — a throwing healthReport() is a contract violation,
     await brain.flush()
 
     const internals = internalsOf(brain)
-    internals.index.healthReport = () => {
+    internals.metadataIndex.healthReport = () => {
       throw new Error('accelerator: mmap window busy')
     }
 
-    await expect(brain.find({ where: { team: 'atlas' } })).rejects.toBeInstanceOf(VectorIndexNotReadyError)
+    await expect(brain.find({ where: { team: 'atlas' } })).rejects.toBeInstanceOf(MetadataIndexNotReadyError)
     await expect(brain.find({ where: { team: 'atlas' } })).rejects.toThrow(/mmap window busy/)
 
-    delete internals.index.healthReport
+    delete internals.metadataIndex.healthReport
   })
 })

tests/integration/read-gate-scope-and-no-reembed.test.ts

@@ -0,0 +1,83 @@
+/**
+ * @module tests/integration/read-gate-scope-and-no-reembed
+ * @description Two cures from the pair's first production adoption:
+ * (1) THE READ GATE IS PER-FAMILY — a not-serving VECTOR leg refuses vector
+ *     search only; a pure metadata find({ where }) and graph traversal keep
+ *     serving. The brain-global gate refused a deployment's badge reads for a
+ *     vector-leg verdict that had nothing to do with them.
+ * (2) NO RE-EMBED ON UNCHANGED DATA — an update() carrying the row's current
+ *     data lands no vector, defers no embed, rewrites nothing. A host
+ *     heartbeat re-writing an unchanged row fed a live index-row loop.
+ */
+import { describe, it, expect, beforeEach, afterEach } from 'vitest'
+import * as fs from 'node:fs'
+import * as os from 'node:os'
+import * as path from 'node:path'
+import { Brainy, VectorIndexNotReadyError } from '../../src/index.js'
+
+describe('read gate scope + no re-embed on unchanged data', () => {
+  let dir: string
+  let brain: any
+
+  beforeEach(async () => {
+    process.env.BRAINY_DETERMINISTIC_EMBEDDINGS = 'true'
+    dir = fs.mkdtempSync(path.join(os.tmpdir(), 'brainy-gate-scope-'))
+    brain = new Brainy({ requireSubtype: false, storage: { type: 'filesystem', path: dir }, silent: true, dimensions: 384 })
+    await brain.init()
+  })
+  afterEach(async () => {
+    await brain.close?.().catch(() => {})
+    fs.rmSync(dir, { recursive: true, force: true })
+  })
+
+  it('a not-serving VECTOR leg refuses vector search only — metadata and graph reads keep serving', async () => {
+    const a = await brain.add({ data: 'employee alpha', type: 'person', metadata: { status: 'active' } })
+    const b = await brain.add({ data: 'employee beta', type: 'person', metadata: { status: 'active' } })
+    await brain.relate({ from: a, to: b, type: 'relatedTo' })
+    await brain.flush()
+
+    // The vector provider says it is NOT serving (a rebuild-class failure).
+    brain.index.healthReport = () => ({
+      provider: 'vector', healthy: false, serving: false, generation: 7, unledgered: [],
+      invariants: [{ name: 'node-coverage', holds: false, heal: 'rebuild', detail: 'posted 0 < canonical 2', source: 'ledger' }],
+      checkedAt: 1, durationMs: 1
+    })
+    try {
+      const byStatus = await brain.find({ where: { status: 'active' } })
+      expect(byStatus.map((r: any) => r.id).sort(), 'metadata find serves').toEqual([a, b].sort())
+      const rel = await brain.related(a)
+      expect(rel.length, 'graph traversal serves').toBe(1)
+      await expect(brain.find({ query: 'employee' }), 'vector search refuses typed').rejects.toBeInstanceOf(VectorIndexNotReadyError)
+    } finally {
+      delete brain.index.healthReport
+    }
+  })
+
+  it('update() with the row\'s current data re-embeds nothing; a real change re-embeds', async () => {
+    const id = await brain.add({ data: 'invoice 1042 pending', type: 'document', metadata: { n: 1 } })
+    await brain.flush()
+    const before = (await brain.get(id, { includeVectors: true })).vector
+    const ledgerBefore = await brain.storage.getCanonicalCounts()
+    const logBefore = (await brain.transactionLog({ limit: 50 })).length
+
+    // The heartbeat shape: same data, re-written, deferred.
+    for (let i = 0; i < 3; i++) {
+      await brain.update({ id, data: 'invoice 1042 pending', metadata: { n: 1, tick: i }, deferEmbedding: true })
+    }
+    await brain.flush()
+    const after = (await brain.get(id, { includeVectors: true })).vector
+    const ledgerAfter = await brain.storage.getCanonicalCounts()
+    const log = await brain.transactionLog({ limit: 50 })
+    expect(after, 'vector untouched by unchanged-data writes').toEqual(before)
+    expect(ledgerAfter.vectors.all, 'vectored ledger untouched').toBe(ledgerBefore.vectors.all)
+    expect(log.filter((e: any) => e.origin === 'system:embed-landing').length, 'no landing commit for unchanged data').toBe(0)
+    expect(log.length - logBefore, 'the metadata writes themselves still commit').toBe(3)
+
+    // A REAL change re-embeds (deferred → the worker lands it).
+    await brain.update({ id, data: 'invoice 1042 PAID', deferEmbedding: true })
+    await brain.flush()
+    const changed = (await brain.get(id, { includeVectors: true })).vector
+    expect(changed, 'a real data change re-embeds').not.toEqual(before)
+    expect((await brain.storage.getCanonicalCounts()).vectors.all, 'a re-embed of a vectored row never double-counts').toBe(ledgerBefore.vectors.all)
+  })
+})