diff --git a/docs/STORAGE_BACKEND_MATRIX.md b/docs/STORAGE_BACKEND_MATRIX.md
new file mode 100644
index 0000000..30870ec
--- /dev/null
+++ b/docs/STORAGE_BACKEND_MATRIX.md
@@ -0,0 +1,63 @@
+# Storage & DB backend matrix
+
+A selection of storage / DB backends and their properties — the estate's own backends first, then the
+external field for context. The dimensions are **the SocioProphet storage-standard properties**, not a
+generic feature list: content-addressing, provenance, integrity-at-rest, fault tolerance, and
+sovereignty are what the standard load-bears on. `(✓)` = supported via an adapter / optional dependency.
+
+> Measure it or don't claim it. Perf numbers for any row belong in a reproducible **EvaluationRecord**
+> (SocioProphet/socioprophet-standards-storage `evaluation-record-standard.v1`), produced by
+> `ts/src/storage-bench.ts` — never inline in this table.
+
+## Estate backends
+
+| Backend | Model | Content-addressed | Provenance-stamped | Integrity-at-rest | Fault tolerance | Sovereign / BYOS | In-memory | Distributed | Fail-closed verify |
+|---|---|---|---|---|---|---|---|---|---|
+| `InMemoryObjectBackend` | blob KV | ✓ (sha256 key) | (via store) | (via codex) | – | – | ✓ | – | ✓ (bench) |
+| `CanonicalObjectStore` | content-addressed object store | ✓ | ✓ (`ObjectProvenance`) | ✓ (**codex-seal** at ingest) | (backend) | (backend) | (backend) | (backend) | ✓ (`verify → Syndrome`) |
+| `S3ObjectBackend` (BYOS) | object store on S3/MinIO | ✓ | (via store) | (via codex) | ✓ (S3/MinIO) | ✓ (customer holds bytes) | – | ✓ | ✓ (bench) |
+| `RocksDBBackend` | embedded LSM KV (AtomSpace WAL) | ✓ (atom handle) | ✓ (log entry) | ✓ (replay) | ✓ (WAL) | ✓ (local-first) | (cache) | – | ✓ |
+| `HypercoreBackend` | signed append-only log | ✓ | ✓ (signed) | ✓ (merkle) | ✓ (replicated) | ✓ (per-writer keys) | – | ✓ (Autobase) | ✓ |
+| `AtomSpace` (metagraph) | typed hypergraph, PLN truth | ✓ (structural hash) | ✓ (Values) | ✓ (codex sealer) | (backend) | ✓ | ✓ | (super-peer) | ✓ |
+| `FederatedAtomSpace` | Autobase-merged sovereign logs | ✓ | ✓ (per-op writer+seq) | ✓ (causal-cut proof) | ✓ (multi-writer) | ✓ (federation) | – | ✓ | ✓ (proof withholds) |
+
+## External field (context)
+
+| Backend | Model | Content-addressed | Provenance | Fault tolerance | In-memory | Distributed | Language |
+|---|---|---|---|---|---|---|---|
+| RocksDB | embedded LSM KV | – | – | ✓ (WAL) | (cache) | – | C++ |
+| Neo4j | property graph | – | – | ✓ | (cache) | ✓ | Java |
+| Kùzu | embedded graph | – | – | ✓ | ✓ | – | C++ |
+| Redis | in-memory KV | – | – | (✓) | ✓ | ✓ | C |
+| S3 / MinIO | object store | (by key) | – | ✓ | – | ✓ | Go / — |
+| Postgres | relational | – | – | ✓ | (cache) | (✓) | C |
+| IPFS | content-addressed | ✓ | – | ✓ | – | ✓ | Go |
+
+## What the estate adds over the field
+
+The standard-aligned columns are exactly where the estate backends differ from the external field:
+**content-addressing + provenance + integrity-at-rest (codex-seal) + causal-cut proof** are first-class,
+not bolt-ons. A backend that returns the wrong bytes fast is a *failure* here — see the fail-closed
+integrity check in `storage-bench.ts` (a corrupting/losing backend fails the run, proven both ways in
+`storage-bench.test.ts`).
+
+## Running the benchmark
+
+```ts
+import { InMemoryObjectBackend } from './object-store.js'
+import { benchObjectBackend, toEvaluationRecord } from './storage-bench.js'
+
+const result = await benchObjectBackend(new InMemoryObjectBackend(),
+  { backend_id: 'in-memory', n_ops: 10_000, payload_bytes: 4096, seed: 42 })
+
+// store as evidence — a number is only a claim if it is backed by a reproducible record.
+const record = toEvaluationRecord(result, {
+  subject_ref: 'hellgraph:CanonicalObjectStore',
+  evaluation_track_ref: 'track:storage-backend-perf.v1',
+})
+```
+
+The `EvaluationRecord` conforms to SocioProphet/socioprophet-standards-storage
+`standards/evaluation-record-standard.v1` (`subject_type: platform_capability`,
+`attempt_mode: benchmark_run`), carrying the raw measurements as evidence so results are storable,
+reviewable, and regression-checkable across epochs.
diff --git a/ts/dist/index.d.mts b/ts/dist/index.d.mts
index 3daa451..73307aa 100644
Binary files a/ts/dist/index.d.mts and b/ts/dist/index.d.mts differ
diff --git a/ts/dist/index.d.ts b/ts/dist/index.d.ts
index 3daa451..73307aa 100644
Binary files a/ts/dist/index.d.ts and b/ts/dist/index.d.ts differ
diff --git a/ts/dist/index.js b/ts/dist/index.js
index e258c09..b97460c 100644
Binary files a/ts/dist/index.js and b/ts/dist/index.js differ
diff --git a/ts/dist/index.mjs b/ts/dist/index.mjs
index c6fb5c2..8b13c56 100644
Binary files a/ts/dist/index.mjs and b/ts/dist/index.mjs differ
diff --git a/ts/src/index.ts b/ts/src/index.ts
index 388e1c8..19639b7 100644
--- a/ts/src/index.ts
+++ b/ts/src/index.ts
@@ -77,3 +77,4 @@ export * from './nlq'
 export * from './effect-request-data'
 export * from './vendor-graph'
 export * from './agent-graph-ingest'
+export * from './storage-bench'
diff --git a/ts/src/storage-bench.test.ts b/ts/src/storage-bench.test.ts
new file mode 100644
index 0000000..335ea80
--- /dev/null
+++ b/ts/src/storage-bench.test.ts
@@ -0,0 +1,66 @@
+import { test } from 'node:test'
+import assert from 'node:assert/strict'
+import { InMemoryObjectBackend, type ObjectBackend } from './object-store.js'
+import { benchObjectBackend, toEvaluationRecord } from './storage-bench.js'
+
+const CFG = { backend_id: 'in-memory', n_ops: 64, payload_bytes: 256, seed: 42 }
+
+test('a correct backend passes with full integrity + populated latency/throughput', async () => {
+  const r = await benchObjectBackend(new InMemoryObjectBackend(), CFG)
+  assert.equal(r.result, 'pass')
+  assert.equal(r.integrity_verified, CFG.n_ops)
+  assert.equal(r.integrity_failures, 0)
+  const put = r.ops.find((o) => o.op === 'put')!
+  assert.equal(put.count, CFG.n_ops)
+  assert.ok(put.latency.p95_ms >= put.latency.p50_ms)      // percentiles ordered
+  assert.ok(put.throughput_ops_s > 0 && put.throughput_mb_s > 0)
+})
+
+test('the run is deterministic for a fixed seed (reproducible — the standard requires it)', async () => {
+  const a = await benchObjectBackend(new InMemoryObjectBackend(), CFG)
+  const b = await benchObjectBackend(new InMemoryObjectBackend(), CFG)
+  assert.equal(a.integrity_verified, b.integrity_verified)
+  assert.equal(a.result, b.result)
+})
+
+// a fast liar: stores nothing and returns wrong bytes. It MUST fail the bench.
+class CorruptingBackend implements ObjectBackend {
+  async put(): Promise<void> {}
+  async get(): Promise<Buffer> { return Buffer.from('not the bytes you stored') }
+}
+
+test('a backend that returns the wrong bytes FAILS (integrity is first-class, teeth both ways)', async () => {
+  const r = await benchObjectBackend(new CorruptingBackend(), CFG)
+  assert.equal(r.result, 'fail')
+  assert.ok(r.integrity_failures > 0)
+  assert.equal(r.integrity_verified, 0)
+})
+
+// a backend that loses data (get returns undefined) is also a failure, counted as a get failure.
+class LosingBackend implements ObjectBackend {
+  async put(): Promise<void> {}
+  async get(): Promise<Buffer | undefined> { return undefined }
+}
+
+test('a backend that loses data FAILS (missing blob is not a silent pass)', async () => {
+  const r = await benchObjectBackend(new LosingBackend(), CFG)
+  assert.equal(r.result, 'fail')
+  const get = r.ops.find((o) => o.op === 'get')!
+  assert.equal(get.failures, CFG.n_ops)
+})
+
+test('toEvaluationRecord conforms to the storage standard shape', async () => {
+  const r = await benchObjectBackend(new InMemoryObjectBackend(), CFG)
+  const rec = toEvaluationRecord(r, {
+    subject_ref: 'hellgraph:InMemoryObjectBackend',
+    evaluation_track_ref: 'track:storage-backend-perf.v1',
+  })
+  assert.equal(rec.subject_type, 'platform_capability')
+  assert.equal(rec.attempt_refs[0]!.attempt_mode, 'benchmark_run')
+  assert.ok(['pass', 'fail'].includes(rec.result))
+  assert.equal(rec.review_state, 'draft')
+  for (const k of ['id', 'evaluation_track_ref', 'subject_ref', 'evidence_bundle_ref', 'created_at', 'updated_at'])
+    assert.ok((rec as Record<string, unknown>)[k], `missing required field ${k}`)
+  // the raw, reproducible measurements are carried as evidence (no bare numbers).
+  assert.equal(rec.attempt_refs[0]!.measurements.integrity_verified, CFG.n_ops)
+})
diff --git a/ts/src/storage-bench.ts b/ts/src/storage-bench.ts
new file mode 100644
index 0000000..e31c317
--- /dev/null
+++ b/ts/src/storage-bench.ts
@@ -0,0 +1,195 @@
+/**
+ * storage-bench — a performance + integrity benchmark harness for storage / DB backends,
+ * aligned to the SocioProphet storage standard (evidence-first: a number is only a claim if it is
+ * backed by a reproducible record).
+ *
+ * It measures any `ObjectBackend` (InMemory / S3-BYOS / RocksDB / …) on the operations a
+ * content-addressed store actually performs — put, get, and verify (get + re-hash) — reporting
+ * latency percentiles and throughput. Integrity is a FIRST-CLASS result, not a footnote: if a
+ * round-tripped blob does not byte-match, or its content hash does not re-derive, the run is a
+ * FAILURE. A fast backend that returns the wrong bytes fails here; that is the point.
+ *
+ * `toEvaluationRecord` emits a record conforming to
+ * SocioProphet/socioprophet-standards-storage `standards/evaluation-record-standard.v1.md`
+ * (subject_type=platform_capability, attempt_mode=benchmark_run) so a result is storable,
+ * reviewable, and regression-checkable — never a bare number.
+ */
+import { createHash } from 'node:crypto'
+import { performance } from 'node:perf_hooks'
+import type { ObjectBackend } from './object-store.js'
+
+export interface StorageBenchConfig {
+  backend_id: string
+  n_ops: number
+  payload_bytes: number
+  /** deterministic seed so a run is reproducible (the storage standard requires it) */
+  seed?: number
+}
+
+export interface LatencyStats {
+  p50_ms: number
+  p95_ms: number
+  p99_ms: number
+  max_ms: number
+  mean_ms: number
+}
+
+export interface OpResult {
+  op: 'put' | 'get' | 'verify'
+  count: number
+  failures: number
+  latency: LatencyStats
+  throughput_ops_s: number
+  throughput_mb_s: number
+}
+
+export interface StorageBenchResult {
+  backend_id: string
+  config: StorageBenchConfig
+  ops: OpResult[]
+  integrity_verified: number
+  integrity_failures: number
+  /** fail-closed: any integrity failure (or op failure) makes the whole run a failure */
+  result: 'pass' | 'fail'
+  started_at: string
+  finished_at: string
+}
+
+const sha256 = (b: Buffer): string => createHash('sha256').update(b).digest('hex')
+
+/** Deterministic pseudo-random bytes from a seed, so a run reproduces exactly. */
+function seededPayload(seed: number, i: number, size: number): Buffer {
+  const out = Buffer.allocUnsafe(size)
+  let x = (seed ^ (i * 2654435761)) >>> 0
+  for (let j = 0; j < size; j++) {
+    x ^= x << 13; x ^= x >>> 17; x ^= x << 5; x >>>= 0
+    out[j] = x & 0xff
+  }
+  return out
+}
+
+function stats(samples: number[]): LatencyStats {
+  if (!samples.length) return { p50_ms: 0, p95_ms: 0, p99_ms: 0, max_ms: 0, mean_ms: 0 }
+  const s = [...samples].sort((a, b) => a - b)
+  const q = (p: number) => s[Math.min(s.length - 1, Math.floor(p * (s.length - 1)))]
+  return {
+    p50_ms: q(0.5), p95_ms: q(0.95), p99_ms: q(0.99), max_ms: s[s.length - 1],
+    mean_ms: samples.reduce((a, b) => a + b, 0) / samples.length,
+  }
+}
+
+function opResult(op: OpResult['op'], lat: number[], failures: number, bytesEach: number): OpResult {
+  const totalMs = lat.reduce((a, b) => a + b, 0) || 1
+  return {
+    op,
+    count: lat.length,
+    failures,
+    latency: stats(lat),
+    throughput_ops_s: (lat.length / totalMs) * 1000,
+    throughput_mb_s: ((lat.length * bytesEach) / (1024 * 1024) / totalMs) * 1000,
+  }
+}
+
+/**
+ * Benchmark an ObjectBackend over `n_ops` content-addressed blobs: put, then get, then verify
+ * (re-hash the retrieved bytes). Deterministic given `seed`. Fail-closed on any integrity mismatch.
+ */
+export async function benchObjectBackend(
+  backend: ObjectBackend,
+  config: StorageBenchConfig,
+): Promise<StorageBenchResult> {
+  const seed = config.seed ?? 0x5010
+  const started_at = new Date().toISOString()
+  const hashes: string[] = []
+  const putLat: number[] = []
+  const getLat: number[] = []
+  const verLat: number[] = []
+  let putFail = 0, getFail = 0
+  let integrity_verified = 0, integrity_failures = 0
+
+  for (let i = 0; i < config.n_ops; i++) {
+    const bytes = seededPayload(seed, i, config.payload_bytes)
+    const hash = sha256(bytes)
+    hashes.push(hash)
+    const t0 = performance.now()
+    try { await backend.put(hash, bytes) } catch { putFail++ }
+    putLat.push(performance.now() - t0)
+  }
+
+  for (let i = 0; i < hashes.length; i++) {
+    const expected = seededPayload(seed, i, config.payload_bytes)
+    const t0 = performance.now()
+    let got: Buffer | undefined
+    try { got = await backend.get(hashes[i]!) } catch { got = undefined }
+    getLat.push(performance.now() - t0)
+    if (!got) { getFail++; continue }
+
+    // verify: bytes round-trip AND the content hash re-derives (content-addressing holds).
+    const tv = performance.now()
+    const rehash = sha256(got)
+    verLat.push(performance.now() - tv)
+    if (rehash === hashes[i] && got.equals(expected)) integrity_verified++
+    else integrity_failures++
+  }
+
+  const opFailures = putFail + getFail
+  const finished_at = new Date().toISOString()
+  return {
+    backend_id: config.backend_id,
+    config: { ...config, seed },
+    ops: [
+      opResult('put', putLat, putFail, config.payload_bytes),
+      opResult('get', getLat, getFail, config.payload_bytes),
+      opResult('verify', verLat, 0, config.payload_bytes),
+    ],
+    integrity_verified,
+    integrity_failures,
+    result: integrity_failures === 0 && opFailures === 0 && integrity_verified === config.n_ops ? 'pass' : 'fail',
+    started_at,
+    finished_at,
+  }
+}
+
+// ── SocioProphet storage-standard EvaluationRecord (evaluation-record-standard.v1) ──────────────
+
+export interface EvaluationRecord {
+  id: string
+  evaluation_track_ref: string
+  subject_ref: string
+  subject_type: 'platform_capability'
+  attempt_refs: { id: string; attempt_mode: 'benchmark_run'; measurements: StorageBenchResult }[]
+  metric_refs: string[]
+  result_summary: string
+  result: 'pass' | 'pass_with_findings' | 'remediation_required' | 'fail' | 'blocked' | 'unknown'
+  evidence_bundle_ref: string
+  review_state: 'draft'
+  created_at: string
+  updated_at: string
+}
+
+/** Wrap a bench result as a storage-standard EvaluationRecord (evidence-first, regression-checkable). */
+export function toEvaluationRecord(
+  r: StorageBenchResult,
+  opts: { subject_ref: string; evaluation_track_ref: string },
+): EvaluationRecord {
+  const put = r.ops.find((o) => o.op === 'put')!
+  const get = r.ops.find((o) => o.op === 'get')!
+  const now = new Date().toISOString()
+  return {
+    id: `evalrec:storage-bench:${r.backend_id}:${r.started_at}`,
+    evaluation_track_ref: opts.evaluation_track_ref,
+    subject_ref: opts.subject_ref,
+    subject_type: 'platform_capability',
+    attempt_refs: [{ id: `attempt:${r.backend_id}:${r.started_at}`, attempt_mode: 'benchmark_run', measurements: r }],
+    metric_refs: ['put.p95_ms', 'get.p95_ms', 'put.throughput_mb_s', 'integrity_verified'],
+    result_summary:
+      `${r.backend_id}: ${r.config.n_ops}×${r.config.payload_bytes}B — ` +
+      `put p95 ${put.latency.p95_ms.toFixed(3)}ms, get p95 ${get.latency.p95_ms.toFixed(3)}ms, ` +
+      `integrity ${r.integrity_verified}/${r.config.n_ops}`,
+    result: r.result === 'pass' ? 'pass' : 'fail',
+    evidence_bundle_ref: `evidence:storage-bench:${r.backend_id}:${r.started_at}`,
+    review_state: 'draft',
+    created_at: r.started_at,
+    updated_at: now,
+  }
+}