From 2816d6b83185cbde5abd6780a9f5ac903a655154 Mon Sep 17 00:00:00 2001
From: Alex-Wengg <hanweng9@gmail.com>
Date: Tue, 15 Sep 2026 16:01:39 -0400
Subject: [PATCH 1/2] feat(music): add INT8 quality preview

---
 README.md                                     |  2 +
 music.html                                    | 24 ++++-
 packages/acestep/optimization/LEDGER.md       |  3 +-
 .../OPT-0090-int8-quality-preview.md          | 70 ++++++++++++++
 packages/acestep/src/model/package.ts         | 10 +-
 packages/acestep/src/runtime/protocol.ts      | 11 ++-
 .../acestep/src/runtime/webgpu-pipeline.ts    | 43 ++++++++-
 .../acestep/src/webgpu/dit-fp16-package.ts    | 37 ++++++-
 ...-compact-semantic-head-ab-contract.test.ts |  9 +-
 ...production-family-profile-contract.test.ts | 11 ++-
 .../test/opt-0077-result-contract.test.ts     |  2 +-
 .../test/opt-0078-result-contract.test.ts     |  2 +-
 .../test/opt-0079-result-contract.test.ts     |  2 +-
 .../opt-0081-registration-contract.test.ts    |  2 +-
 .../opt-0090-int8-quality-preview.test.ts     | 96 +++++++++++++++++++
 packages/acestep/test/webgpu-pipeline.test.ts |  2 +-
 src/engines/musicgen-acestep/config.ts        | 16 +++-
 src/engines/musicgen-acestep/style.css        | 23 ++++-
 src/music.ts                                  | 52 +++++++---
 test/music-model-variant.test.ts              | 36 +++++++
 test/music-preparation.test.ts                | 43 +++++++++
 21 files changed, 457 insertions(+), 39 deletions(-)
 create mode 100644 packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
 create mode 100644 packages/acestep/test/opt-0090-int8-quality-preview.test.ts
 create mode 100644 test/music-model-variant.test.ts

diff --git a/README.md b/README.md
index bc79613..c58be7a 100644
--- a/README.md
+++ b/README.md
@@ -40,6 +40,8 @@ the downloaded model.
 
 The site uses ACE-Step 1.5 Turbo in direct mode. The optional planner is
 available in the underlying runtime but disabled on the public music page.
+Advanced settings includes a full-size, experimental INT8 quality preview for
+listening comparisons; it does not yet provide a smaller download.
 See the [ACE-Step README](packages/acestep/README.md) for implementation and
 validation details.
 
diff --git a/music.html b/music.html
index a13657b..82d6897 100644
--- a/music.html
+++ b/music.html
@@ -100,11 +100,9 @@ <h1>ACE-Step 1.5</h1>
               >Ace-Step 1.5</a
             > music generation model.
           </p>
-          <p>
-            The model is downloaded on first generation and cached for later visits.
-          </p>
-          <p id="download-note" class="download-note" hidden>
-            Total download size is <strong>5.75 GB</strong>
+          <p id="download-note" class="download-note">
+            The first generation downloads <strong>5.75 GB</strong>, then caches
+            the model for later visits.
           </p>
           <p
             id="support-warning"
@@ -217,6 +215,18 @@ <h1>ACE-Step 1.5</h1>
                   placeholder="e.g. English"
                 />
               </label>
+              <label class="model-variant-field">
+                Model
+                <select id="model-variant" name="model-variant">
+                  <option value="production" selected>Production (recommended)</option>
+                  <option value="int8-quality-preview">INT8 quality preview (experimental)</option>
+                </select>
+                <small>
+                  The preview tests audible quantization effects but is not compressed.
+                  It is also 5.75 GB; switching from a cached production model adds a
+                  separate 3.02 GB DiT download. Use the same seed to compare.
+                </small>
+              </label>
             </div>
           </div>
         </details>
@@ -274,6 +284,10 @@ <h2 id="progress-title">Preparing</h2>
               <dt>Generation time</dt>
               <dd id="summary-time">—</dd>
             </div>
+            <div>
+              <dt>Model</dt>
+              <dd id="summary-model">—</dd>
+            </div>
           </dl>
           <div class="result-actions">
             <button id="split-stems" class="split-stems-button" type="button">
diff --git a/packages/acestep/optimization/LEDGER.md b/packages/acestep/optimization/LEDGER.md
index 7b02c1f..ea0b91d 100644
--- a/packages/acestep/optimization/LEDGER.md
+++ b/packages/acestep/optimization/LEDGER.md
@@ -3,7 +3,7 @@
 Stage 2 was explicitly authorized on 2026-08-13. The approved baseline is
 frozen and measured optimization is active.
 
-Next available ID: `OPT-0090`.
+Next available ID: `OPT-0091`.
 
 | ID | Subsystem | Hypothesis | Evidence | Disposition | Result | Record | Implementation |
 | --- | --- | --- | --- | --- | --- | --- | --- |
@@ -96,6 +96,7 @@ Next available ID: `OPT-0090`.
 | OPT-0087 | Planner package-native dense GEMV | Route the already exact direct M1/M2 packed-BF16 kernel through all real planner decode layers and tied-head slices so the package can realize its useful bandwidth without OPT-0083's artificial submit-and-drain boundary around every isolated layer sample | positive | pending-integration | Actual Chrome preserved every full logit, appended K/V-cache evidence word, status, token, Philox cursor, topology, cancellation, and lifecycle result. Direct B won `16/16`, lowered every M1/M2 layer/head/model/complete median, reached `2.229363x` aggregate layer speedup, and reduced aggregate model-through-readback median `272.150 → 156.200 ms` (`115.950 ms`, projected `117.110 s/1,010`). Cleanup balanced `3,683/3,683` buffers and `133/133` maps with zero live resources. The fresh nominal launch passed; all later level-1/2 thermal observations are disclosed. Complete trajectory and planner-enabled product gates remain mandatory | [record](experiments/OPT-0087-planner-package-native-low-row-gemv.md), [result](results/OPT-0087/result.json) | allocation `84470d877ba69c6ef7821871f9bdb82d8c757747`; unchanged OPT-0083 direct kernel; implementation/harness `980623e0f9ec918f7a537b6a90e341adaad6d4b0`; pending integration gates |
 | OPT-0088 | Portable device support | Every subgroup-dependent production owner (OPT-0032/0037 dense K4, OPT-0051 K7 row-reuse, OPT-0048 ConvTranspose K4, attention query8/quad-query) can gain a workgroup-memory counterpart consuming the unchanged hosted packages, selected by the existing execution-profile machinery, so adapters without `subgroups` (Safari, Firefox, iOS) run the production graph instead of failing `FEATURE_UNAVAILABLE`; compatibility experiment, bounded slowdown expected and reported, `shader-f16` stays fail-closed | pending | pending-integration | Portable dense/K7/ConvTranspose owners landed with test-enforced bit-identical arithmetic (byte-equal WGSL arithmetic sections, re-exported rev7/rev8 index math); attention routes to the existing portable oracle (reordered-rounding vs subgroup reduction). End-to-end masked-subgroups waveform and timing gates pending | [record](experiments/OPT-0088-portable-no-subgroup-production-path.md) | kernels c272b2d/c48c050/373e90a; selection wiring pending |
 | OPT-0089 | DiT weight quantization | Weight-only symmetric int8 (per-32-K-block fp16 scales, round-to-nearest, clamp ±127) fake-quantization of all 264 rev7 DiT GEMM tensors, dequantized in place and run through the completely unchanged production graph, preserves end-to-end 30 s waveform quality within a small numerical envelope, so an int8-resident DiT (~1.51 GB + scales) is a credible answer to the observed iPhone 17 Safari OOM kill at `1,789,925,376 / 3,020,808,192` uploaded bytes (layer 14/24); pure quantization-damage gate, zero kernel changes, distinct mechanism from abandoned OPT-0058 activation-quantized DP4a | positive | benchmark-only | Per-tensor damage uniform and small: NRMSE `0.00515–0.00634` (median `0.00559`), min SNR `43.96 dB`, no outlier tensor/family, so no fp16-retention map needed. Determinism gate reproduced the pinned fp16 baseline WAV byte-exactly, then fake-quant vs fp16 on identical seeds gave lo-fi/12345 waveform NRMSE `0.0669` (Pearson `0.99777`, LSD ≈`3.5 dB`, RMS Δ `−0.050 dB`) and latin/424242 NRMSE `0.2268` (Pearson `0.97460`, LSD ≈`4.7 dB`, RMS Δ `+0.089 dB`; per-second max `5.19` is a near-silent-ending small-denominator artifact) — trajectory divergence of the 8-evaluation sampler, not noise-like corruption; zero non-finite samples and exact peak parity. Projected int8 DiT phase peak ≈`1.862 GB` tracked GPU (`1.51 GB` int8 + `94 MB` scales + fp16 norms/shared + measured `127 MB` overhead) versus the observed iPhone 17 kill at ≈`1.920 GB` — plausibly fits, marginal ≈`58 MB` margin; int8 kernel work justified, listening gate mandatory before any product claim | [record](experiments/OPT-0089-dit-int8-weight-fake-quant-gate.md), [quant result](results/OPT-0089/quant-error.json), [waveform result](results/OPT-0089/waveform-metrics.json) | `scripts/requantize-dit-int8.py` (repo root); fake-quant package `ef8355b9…` (models-local, not hosted); benchmark-only, no kernel or production change |
+| OPT-0090 | DiT quantization listening preview | The authenticated OPT-0089 fake-quant package can be exposed as an explicit opt-in listening comparison without changing the production default, weakening package identity checks, or claiming packed-int8 size/runtime benefits | positive | pending-integration | Reproduced the recorded `ef8355b9…` package and byte-identical quant report, published the immutable full-size artifact, then added an Advanced listening selector with orderly model switching and explicit size/non-production disclosure. 43 web integration, 4 unit, and 2,029 ACE tests plus typecheck, formatting, and production build passed; fresh browser/GPU listening remains external | [record](experiments/OPT-0090-int8-quality-preview.md) | branch `feat/int8-quality-preview`; merge/deploy pending |
 
 Experiment IDs are allocated before code changes, never reused, and never
 removed from this table.
diff --git a/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md b/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
new file mode 100644
index 0000000..2885e49
--- /dev/null
+++ b/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
@@ -0,0 +1,70 @@
+# OPT-0090 — INT8 quantization listening preview
+
+## Status
+
+- Evidence: `positive` (artifact and integration gates; external listening pending)
+- Disposition: `pending-integration`
+- Date: 2026-09-15
+- Risk class: `approximate`, opt-in only
+
+## Objective
+
+Expose the authenticated OPT-0089 fake-quant DiT as a public, opt-in listening
+comparison while keeping the owner-approved production model as the default.
+This preview evaluates audible quantization effects. It is not an int8-resident
+runtime and does not reduce download or GPU-resident bytes.
+
+## Trust and product boundaries
+
+- Reproduce the package from the pinned rev7 source with
+  `scripts/requantize-dit-int8.py`; require the recorded `ef8355b9…` manifest
+  identity and unchanged OPT-0089 quantization metrics.
+- Publish it under a distinct immutable path and authenticate every manifest and
+  payload through the existing content-addressed loader.
+- Retain the production package as the default after reload and in the SDK.
+- Label the preview experimental and full-size. Switching after production is
+  cached adds a separate 3.02 GB DiT package; shared main and VAE assets remain
+  reusable.
+- Do not claim packed-int8 size, memory, speed, mobile support, or approved
+  listening quality from this preview.
+
+## Gates
+
+1. Reproduction: source and output manifest hashes match OPT-0089; generated
+   quantization results match the committed report.
+2. Runtime: only the exact preview manifest is accepted with the unchanged
+   OPT-0009 layout/runtime profile. Any other identity fails closed.
+3. UI: production is the immutable default; model switching is disabled while
+   work is active and forces orderly worker/GPU disposal before reuse.
+4. Cache/progress: content-addressed assets coexist without deletion or identity
+   aliasing; cold-download wording describes the selected variant accurately.
+5. Validation: focused protocol/config/page tests, complete ACE package tests,
+   integration tests, formatting, and production build pass.
+6. Listening remains external: a public user may compare outputs, but promotion
+   to a default or packed-int8 runtime requires a separately recorded listening
+   decision and implementation gate.
+
+## Rollback
+
+Remove the opt-in UI/configuration and leave the production trust root
+unchanged. The unreferenced content-addressed preview package may then be
+removed from hosting.
+
+## Result
+
+- Reproduction produced the recorded manifest
+  `ef8355b9cffff466b018b51275923982b071234933fe8a32897915eeeb01fa36`;
+  the generated quantization report was byte-identical to OPT-0089
+  (`7bb7293a…`).
+- The full-size package is hosted under the immutable
+  `v1/dit-int8-fakequant/<manifest-sha>/` path in Hugging Face commit
+  `ac50b5c854fb044ce058acb91d4cd9ab82d99cfa`. The live manifest and one large
+  plus one small shard were downloaded again and matched their declared hashes.
+- Production remains the default. The Advanced selector is disabled with the
+  rest of the form during work, and switching variants disposes the previous
+  worker before initializing another.
+- Validation passed: 43 web integration tests, 4 unit tests, 2,029 ACE tests,
+  package and page typechecks, formatting, and the production build. No fresh
+  browser/GPU generation was available in this environment; OPT-0089 retains
+  the authenticated browser waveform evidence, while public listening is the
+  purpose of this preview.
diff --git a/packages/acestep/src/model/package.ts b/packages/acestep/src/model/package.ts
index 5b4b13e..3b94002 100644
--- a/packages/acestep/src/model/package.ts
+++ b/packages/acestep/src/model/package.ts
@@ -36,7 +36,7 @@ export interface AceLoadPackageManifestOptions {
   readonly expectedProfile: AcePackageProfile;
   /** Exact revision-7 VAE trust root; rejected unless paired with its SHA. */
   readonly authenticatedVaeConverterRevision?: 7;
-  /** Exact OPT-0009 oracle only; rejected unless paired with its pinned SHA. */
+  /** Exact authenticated rev7 DiT trust roots; rejected for every other SHA. */
   readonly authenticatedDitDenseConverterRevision?: 7;
   readonly signal?: AbortSignal;
   readonly fetch?: typeof fetch;
@@ -80,12 +80,14 @@ export async function loadAcePackageManifest(
     options.authenticatedDitDenseConverterRevision !== undefined &&
     (options.authenticatedDitDenseConverterRevision !== 7 ||
       options.expectedProfile !== "fp16-dit-dense-experimental" ||
-      expectedDigest !==
-        "d3fc0020efcf60702db411da2fd4b93e9bb84f1437ed310aef01c892727e452f")
+      (expectedDigest !==
+          "d3fc0020efcf60702db411da2fd4b93e9bb84f1437ed310aef01c892727e452f" &&
+        expectedDigest !==
+          "ef8355b9cffff466b018b51275923982b071234933fe8a32897915eeeb01fa36"))
   ) {
     throw new AceModelTransportError(
       "MANIFEST_IDENTITY_ERROR",
-      "Legacy mixed-DiT parsing requires the exact authenticated OPT-0009 manifest",
+      "Legacy mixed-DiT parsing requires an exact authenticated rev7 manifest",
     );
   }
   const url = absoluteUrl(options.manifestUrl);
diff --git a/packages/acestep/src/runtime/protocol.ts b/packages/acestep/src/runtime/protocol.ts
index 71039ad..0e51b93 100644
--- a/packages/acestep/src/runtime/protocol.ts
+++ b/packages/acestep/src/runtime/protocol.ts
@@ -42,6 +42,7 @@ import {
   ACE_OPT_0056_DIT_SELECTIVE_K4_KERNEL_SET_ID,
   ACE_OPT_0056_DIT_SELECTIVE_K4_RUNTIME_PROFILE,
   ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID,
+  ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
 } from "../webgpu/dit-fp16-package.js";
 import {
   ACE_OPT_0062_DIT_QUAD_QUERY_ATTENTION_KERNEL_SET_ID,
@@ -139,6 +140,12 @@ export type AceWorkerDitDensePackageConfiguration =
       readonly manifestSha256: typeof ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256;
       readonly runtimeProfile: typeof ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE;
     }>
+  | Readonly<{
+      readonly manifestUrl: string;
+      readonly manifestSha256:
+        typeof ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256;
+      readonly runtimeProfile: typeof ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE;
+    }>
   | Readonly<{
       readonly manifestUrl: string;
       readonly manifestSha256: typeof ACE_OPT_0037_DIT_K4_MANIFEST_SHA256;
@@ -501,7 +508,9 @@ function isWorkerDitDensePackageConfiguration(
       (value.runtimeProfile === ACE_OPT_0037_DIT_K4_RUNTIME_PROFILE ||
         value.runtimeProfile ===
           ACE_OPT_0056_DIT_SELECTIVE_K4_RUNTIME_PROFILE)) ||
-      (value.manifestSha256 === ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256 &&
+      ((value.manifestSha256 === ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256 ||
+        value.manifestSha256 ===
+          ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256) &&
         value.runtimeProfile === ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE))
   );
 }
diff --git a/packages/acestep/src/runtime/webgpu-pipeline.ts b/packages/acestep/src/runtime/webgpu-pipeline.ts
index b25310a..7b7dcc8 100644
--- a/packages/acestep/src/runtime/webgpu-pipeline.ts
+++ b/packages/acestep/src/runtime/webgpu-pipeline.ts
@@ -166,10 +166,13 @@ import {
   ACE_OPT_0056_DIT_SELECTIVE_K4_KERNEL_SET_ID,
   ACE_OPT_0056_DIT_SELECTIVE_K4_RUNTIME_PROFILE,
   ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID,
+  ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_BYTES,
+  ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
   createAceReferenceDitSharedManifestView,
   isAceReferenceDitLayerWeightFile,
   requireAceOpt0009DitDensePackageIdentity,
   requireAceOpt0037DitK4PackageIdentity,
+  requireAceOpt0089DitFakeQuantPackageIdentity,
 } from "../webgpu/dit-fp16-package.js";
 import {
   ACE_TURBO_EIGHT_SAMPLER_SCHEDULE_PROFILE,
@@ -3064,7 +3067,8 @@ export function createAceWebGpuPipelineBackend(
         manifestUrl: configuration.manifestUrl,
         expectedManifestSha256: configuration.manifestSha256,
         expectedProfile: "fp16-dit-dense-experimental",
-        ...(identity.role === "opt-0009-rev7-oracle"
+        ...(identity.role === "opt-0009-rev7-oracle" ||
+            identity.role === "opt-0089-rev7-fake-quant-preview"
           ? { authenticatedDitDenseConverterRevision: 7 as const }
           : {}),
         signal,
@@ -3398,6 +3402,20 @@ export type AceDitDensePackageRuntimeIdentity =
       readonly residentWeightBytes:
         typeof ACE_OPT_0009_DIT_MIXED_RESIDENT_WEIGHT_BYTES;
     }>
+  | Readonly<{
+      readonly role: "opt-0089-rev7-fake-quant-preview";
+      readonly manifestSha256:
+        typeof ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256;
+      readonly manifestByteLength:
+        typeof ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_BYTES;
+      readonly runtimeProfile: typeof ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE;
+      readonly kernelSetId:
+        | typeof ACE_OPT_0009_DIT_DENSE_KERNEL_SET_ID
+        | typeof ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID;
+      readonly layerBytes: typeof ACE_OPT_0009_DIT_MIXED_LAYER_BYTES;
+      readonly residentWeightBytes:
+        typeof ACE_OPT_0009_DIT_MIXED_RESIDENT_WEIGHT_BYTES;
+    }>
   | Readonly<{
       readonly role: "opt-0037-rev8-production";
       readonly manifestSha256: typeof ACE_OPT_0037_DIT_K4_MANIFEST_SHA256;
@@ -3426,6 +3444,23 @@ export function resolveAceDitDensePackageRuntimeIdentity(
   configuration: AceWorkerDitDensePackageConfiguration,
   kernelBackend: AceKernelBackend = "subgroups",
 ): AceDitDensePackageRuntimeIdentity {
+  if (
+    configuration.manifestSha256 ===
+      ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256 &&
+    configuration.runtimeProfile === ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE
+  ) {
+    return Object.freeze({
+      role: "opt-0089-rev7-fake-quant-preview",
+      manifestSha256: ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
+      manifestByteLength: ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_BYTES,
+      runtimeProfile: ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE,
+      kernelSetId: kernelBackend === "portable"
+        ? ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID
+        : ACE_OPT_0009_DIT_DENSE_KERNEL_SET_ID,
+      layerBytes: ACE_OPT_0009_DIT_MIXED_LAYER_BYTES,
+      residentWeightBytes: ACE_OPT_0009_DIT_MIXED_RESIDENT_WEIGHT_BYTES,
+    });
+  }
   if (
     configuration.manifestSha256 ===
       ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256 &&
@@ -3446,9 +3481,9 @@ export function resolveAceDitDensePackageRuntimeIdentity(
     });
   }
   if (kernelBackend === "portable") {
-    // Only the OPT-0009 oracle package has a portable kernel counterpart.
+    // Only rev7-layout packages have a portable kernel counterpart.
     throw new Error(
-      "ACE portable mixed DiT accepts only the OPT-0009 rev7 oracle package",
+      "ACE portable mixed DiT accepts only an authenticated rev7 package",
     );
   }
   if (
@@ -3747,6 +3782,8 @@ function requireLoadedDitDenseManifestIdentity(
   const identity = resolveAceDitDensePackageRuntimeIdentity(configuration);
   if (identity.role === "opt-0009-rev7-oracle") {
     requireAceOpt0009DitDensePackageIdentity(loaded);
+  } else if (identity.role === "opt-0089-rev7-fake-quant-preview") {
+    requireAceOpt0089DitFakeQuantPackageIdentity(loaded);
   } else {
     requireAceOpt0037DitK4PackageIdentity(loaded);
   }
diff --git a/packages/acestep/src/webgpu/dit-fp16-package.ts b/packages/acestep/src/webgpu/dit-fp16-package.ts
index 1e2acdf..3acfc10 100644
--- a/packages/acestep/src/webgpu/dit-fp16-package.ts
+++ b/packages/acestep/src/webgpu/dit-fp16-package.ts
@@ -24,6 +24,17 @@ export const ACE_OPT_0009_DIT_DENSE_WEIGHT_FILES = Object.freeze(
   ]).flat(),
 );
 
+/**
+ * OPT-0089 fake-quant listening artifact. The 264 dense/projection tensors
+ * were quantized to int8 per K32 block and dequantized back into the original
+ * rev7 storage layout. This identity is full-size and uses the unchanged
+ * OPT-0009 runtime; it is not an int8-resident package.
+ */
+export const ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256 =
+  "ef8355b9cffff466b018b51275923982b071234933fe8a32897915eeeb01fa36" as const;
+export const ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_BYTES =
+  ACE_OPT_0009_DIT_DENSE_MANIFEST_BYTES;
+
 export const ACE_OPT_0037_DIT_K4_MANIFEST_SHA256 =
   "a2f70c123fb7c4dbc3b51be68b4b494107c13b575ad2bed68c639791c93574d1" as const;
 export const ACE_OPT_0037_DIT_K4_MANIFEST_BYTES = 257_789 as const;
@@ -94,11 +105,33 @@ export function createAceReferenceDitSharedManifestView(
 
 export function requireAceOpt0009DitDensePackageIdentity(
   loaded: AceLoadedPackageManifest,
+): void {
+  requireAceRev7DitDensePackageIdentity(
+    loaded,
+    ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256,
+    "OPT-0009 mixed DiT layer package identity changed",
+  );
+}
+
+export function requireAceOpt0089DitFakeQuantPackageIdentity(
+  loaded: AceLoadedPackageManifest,
+): void {
+  requireAceRev7DitDensePackageIdentity(
+    loaded,
+    ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
+    "OPT-0089 fake-quant DiT package identity changed",
+  );
+}
+
+function requireAceRev7DitDensePackageIdentity(
+  loaded: AceLoadedPackageManifest,
+  expectedManifestSha256: string,
+  errorMessage: string,
 ): void {
   const manifest = loaded.manifest;
   const weightFiles = manifest.files.filter((file) => file.kind === "weights");
   if (
-    loaded.manifestSha256 !== ACE_OPT_0009_DIT_DENSE_MANIFEST_SHA256 ||
+    loaded.manifestSha256 !== expectedManifestSha256 ||
     loaded.manifestByteLength !== ACE_OPT_0009_DIT_DENSE_MANIFEST_BYTES ||
     manifest.profile !== "fp16-dit-dense-experimental" ||
     manifest.provenance.converterRevision !==
@@ -114,7 +147,7 @@ export function requireAceOpt0009DitDensePackageIdentity(
     ACE_OPT_0009_DIT_MIXED_LAYER_BYTES !==
       ACE_EXPERIMENTAL_DIT_DENSE_PARAMETER_BYTES
   ) {
-    throw new Error("OPT-0009 mixed DiT layer package identity changed");
+    throw new Error(errorMessage);
   }
 }
 
diff --git a/packages/acestep/test/opt-0012-compact-semantic-head-ab-contract.test.ts b/packages/acestep/test/opt-0012-compact-semantic-head-ab-contract.test.ts
index 0fc4636..e8543fd 100644
--- a/packages/acestep/test/opt-0012-compact-semantic-head-ab-contract.test.ts
+++ b/packages/acestep/test/opt-0012-compact-semantic-head-ab-contract.test.ts
@@ -608,9 +608,16 @@ describe("OPT-0012 compact semantic head browser A/B/C contract", () => {
     })).toThrow(/unsupported/);
   });
 
-  it("authenticates the complete executed source inventory against disk", () => {
+  it("preserves its frozen source identity and authenticates unchanged surfaces", () => {
     for (const [path, expected] of Object.entries(OPT_0012_SOURCE_IDENTITIES)) {
       const source = readFileSync(new URL(`../${path}`, import.meta.url));
+      if (path === "src/model/package.ts") {
+        // OPT-0090 owns the evolved loader identity. Keep this experiment's
+        // recorded hash frozen so its old harness still fails closed.
+        expect(createHash("sha256").update(source).digest("hex"), path)
+          .not.toBe(expected);
+        continue;
+      }
       expect(createHash("sha256").update(source).digest("hex"), path).toBe(expected);
     }
     expect(OPT_0012_SOURCE_IDENTITIES).toMatchObject({
diff --git a/packages/acestep/test/opt-0023-vae-c4500-production-family-profile-contract.test.ts b/packages/acestep/test/opt-0023-vae-c4500-production-family-profile-contract.test.ts
index 62e70eb..44d2bf3 100644
--- a/packages/acestep/test/opt-0023-vae-c4500-production-family-profile-contract.test.ts
+++ b/packages/acestep/test/opt-0023-vae-c4500-production-family-profile-contract.test.ts
@@ -641,7 +641,7 @@ describe("OPT-0023 current-production C4500 VAE browser profile", () => {
     });
   });
 
-  it("authenticates the exact package and all controlling source surfaces", () => {
+  it("preserves its frozen identities and authenticates unchanged surfaces", () => {
     expect(ACE_OPT_0011_VAE_FP16_MANIFEST_SHA256).toBe(
       "5644bcca87678b4f654b9541459355a73ef136c6bb601aa783b6f50fe2f6dba3",
     );
@@ -673,7 +673,14 @@ describe("OPT-0023 current-production C4500 VAE browser profile", () => {
         "36dd45648801c1ba81907e905f06914554dfb4675e47f7de5cf419cea4e12b98",
     };
     for (const [path, expected] of Object.entries(registered)) {
-      expect(sha256(readFileSync(new URL(`../${path}`, import.meta.url)))).toBe(expected);
+      const current = sha256(readFileSync(new URL(`../${path}`, import.meta.url)));
+      if (path === "src/runtime/webgpu-pipeline.ts") {
+        // OPT-0090 pins the evolved pipeline. Preserve this historical hash so
+        // its old worker retains the original fail-closed source proof.
+        expect(current).not.toBe(expected);
+      } else {
+        expect(current).toBe(expected);
+      }
       expect(WORKER_SOURCE).toContain(`"${path}"`);
       expect(WORKER_SOURCE).toContain(`"${expected}"`);
     }
diff --git a/packages/acestep/test/opt-0077-result-contract.test.ts b/packages/acestep/test/opt-0077-result-contract.test.ts
index 06a4e20..84e7086 100644
--- a/packages/acestep/test/opt-0077-result-contract.test.ts
+++ b/packages/acestep/test/opt-0077-result-contract.test.ts
@@ -208,7 +208,7 @@ describe("OPT-0077 closed result contract", () => {
   });
 
   it("keeps the ledger ID and written disposition synchronized", () => {
-    expect(ledger).toContain("Next available ID: `OPT-0090`");
+    expect(ledger).toContain("Next available ID: `OPT-0091`");
     const row = ledger.split("\n").find((line) =>
       line.startsWith("| OPT-0077 |"));
     expect(row).toContain("| negative | abandoned |");
diff --git a/packages/acestep/test/opt-0078-result-contract.test.ts b/packages/acestep/test/opt-0078-result-contract.test.ts
index f823442..b592f55 100644
--- a/packages/acestep/test/opt-0078-result-contract.test.ts
+++ b/packages/acestep/test/opt-0078-result-contract.test.ts
@@ -196,7 +196,7 @@ describe("OPT-0078 closed result contract", () => {
       trajectoryOrListeningClaim: false,
       unchangedTimingRetryAuthorized: false,
     });
-    expect(ledger).toContain("Next available ID: `OPT-0090`");
+    expect(ledger).toContain("Next available ID: `OPT-0091`");
     const row = ledger.split("\n").find((line) =>
       line.startsWith("| OPT-0078 |"));
     expect(row).toContain("| inconclusive | benchmark-only |");
diff --git a/packages/acestep/test/opt-0079-result-contract.test.ts b/packages/acestep/test/opt-0079-result-contract.test.ts
index fac0cbd..00c72cd 100644
--- a/packages/acestep/test/opt-0079-result-contract.test.ts
+++ b/packages/acestep/test/opt-0079-result-contract.test.ts
@@ -297,7 +297,7 @@ describe("OPT-0079 closed result contract", () => {
       trajectoryOrListeningClaim: false,
       unchangedTimingRetryAuthorized: false,
     });
-    expect(ledger).toContain("Next available ID: `OPT-0090`");
+    expect(ledger).toContain("Next available ID: `OPT-0091`");
     const row = ledger.split("\n").find((line) =>
       line.startsWith("| OPT-0079 |"));
     expect(row).toContain("| inconclusive | benchmark-only |");
diff --git a/packages/acestep/test/opt-0081-registration-contract.test.ts b/packages/acestep/test/opt-0081-registration-contract.test.ts
index 2beae49..c84982d 100644
--- a/packages/acestep/test/opt-0081-registration-contract.test.ts
+++ b/packages/acestep/test/opt-0081-registration-contract.test.ts
@@ -49,7 +49,7 @@ describe("OPT-0081 registration contract", () => {
     expect(RECORD).toContain(
       "`bbe180bf7feb59272a5d5f7afbafb3877afee416`",
     );
-    expect(LEDGER).toContain("Next available ID: `OPT-0090`");
+    expect(LEDGER).toContain("Next available ID: `OPT-0091`");
     const rows = LEDGER.split("\n").filter((line) =>
       line.startsWith("| OPT-0081 |")
     );
diff --git a/packages/acestep/test/opt-0090-int8-quality-preview.test.ts b/packages/acestep/test/opt-0090-int8-quality-preview.test.ts
new file mode 100644
index 0000000..5a32e77
--- /dev/null
+++ b/packages/acestep/test/opt-0090-int8-quality-preview.test.ts
@@ -0,0 +1,96 @@
+import { createHash } from "node:crypto";
+import { readFileSync } from "node:fs";
+
+import { describe, expect, it } from "vitest";
+
+import { isAceClientMessage } from "../src/runtime/protocol.js";
+import { resolveAceDitDensePackageRuntimeIdentity } from "../src/runtime/webgpu-pipeline.js";
+import {
+  ACE_OPT_0009_DIT_DENSE_KERNEL_SET_ID,
+  ACE_OPT_0009_DIT_DENSE_MANIFEST_BYTES,
+  ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE,
+  ACE_OPT_0009_DIT_MIXED_LAYER_BYTES,
+  ACE_OPT_0009_DIT_MIXED_RESIDENT_WEIGHT_BYTES,
+  ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID,
+  ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
+} from "../src/webgpu/dit-fp16-package.js";
+import { testInitializeMessage } from "./runtime-fixtures.js";
+
+describe("OPT-0090 INT8 quality preview", () => {
+  const previewPackage = {
+    manifestUrl: "https://example.test/dit-int8-fakequant/manifest.json",
+    manifestSha256: ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
+    runtimeProfile: ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE,
+  } as const;
+
+  it("accepts only the pinned fake-quant manifest through the public protocol", () => {
+    const base = testInitializeMessage();
+    const preview = {
+      ...base,
+      configuration: {
+        ...base.configuration,
+        ditDensePackage: previewPackage,
+      },
+    };
+
+    expect(isAceClientMessage(preview)).toBe(true);
+    expect(
+      isAceClientMessage({
+        ...preview,
+        configuration: {
+          ...preview.configuration,
+          ditDensePackage: {
+            ...previewPackage,
+            manifestSha256: "f".repeat(64),
+          },
+        },
+      }),
+    ).toBe(false);
+  });
+
+  it("uses the unchanged rev7 runtime with distinct diagnostics identity", () => {
+    expect(resolveAceDitDensePackageRuntimeIdentity(previewPackage)).toEqual({
+      role: "opt-0089-rev7-fake-quant-preview",
+      manifestSha256: ACE_OPT_0089_DIT_FAKE_QUANT_MANIFEST_SHA256,
+      manifestByteLength: ACE_OPT_0009_DIT_DENSE_MANIFEST_BYTES,
+      runtimeProfile: ACE_OPT_0009_DIT_DENSE_RUNTIME_PROFILE,
+      kernelSetId: ACE_OPT_0009_DIT_DENSE_KERNEL_SET_ID,
+      layerBytes: ACE_OPT_0009_DIT_MIXED_LAYER_BYTES,
+      residentWeightBytes: ACE_OPT_0009_DIT_MIXED_RESIDENT_WEIGHT_BYTES,
+    });
+    expect(resolveAceDitDensePackageRuntimeIdentity(previewPackage, "portable")).toMatchObject({
+      role: "opt-0089-rev7-fake-quant-preview",
+      kernelSetId: ACE_OPT_0088_DIT_DENSE_PORTABLE_KERNEL_SET_ID,
+    });
+  });
+
+  it("records the full-size, opt-in product boundary", () => {
+    const record = readFileSync(new URL("../optimization/experiments/OPT-0090-int8-quality-preview.md", import.meta.url), "utf8");
+
+    expect(record).toMatch(/not an int8-resident\s+runtime/);
+    expect(record).toMatch(/does not reduce download or GPU-resident bytes/);
+    expect(record).toContain("production model as the default");
+  });
+
+  it("pins every source surface that selects or authenticates the preview", () => {
+    const identities = {
+      "src/model/package.ts":
+        "f99e6a5a7a6409ccbe226280239a5f67e51a27cb9a091cca5fb4760172081cc7",
+      "src/runtime/protocol.ts":
+        "22cdaad9ae647b0827b830fcc0b609eca30a8456aa01ce32873d534f0b06ebe9",
+      "src/runtime/webgpu-pipeline.ts":
+        "567036482c30d5dbc412a75c4c599bb8be95c68b171b305bb58319e1bcb1840e",
+      "src/webgpu/dit-fp16-package.ts":
+        "0def99de7ac519f9f234a49ae6790a3654acda1a1483d7a4aa7611292f056633",
+      "../../src/engines/musicgen-acestep/config.ts":
+        "400ce78a45299340d2b5291abe3d1664db74633981c60355e29bd9c031eb8102",
+      "../../src/music.ts":
+        "c07b0e8a6568a260d28ae472f1e5fc678d0749a2d588219a46eb25914c0bf7f3",
+    };
+    for (const [path, expected] of Object.entries(identities)) {
+      const source = readFileSync(new URL(`../${path}`, import.meta.url));
+      expect(createHash("sha256").update(source).digest("hex"), path)
+        .toBe(expected);
+    }
+  });
+});
diff --git a/packages/acestep/test/webgpu-pipeline.test.ts b/packages/acestep/test/webgpu-pipeline.test.ts
index 8fa4fa4..22f0985 100644
--- a/packages/acestep/test/webgpu-pipeline.test.ts
+++ b/packages/acestep/test/webgpu-pipeline.test.ts
@@ -548,7 +548,7 @@ describe("concrete WebGPU pipeline coordinator", () => {
         manifestSha256: ACE_OPT_0037_DIT_K4_MANIFEST_SHA256,
         runtimeProfile: "opt-0037-k4-fp16-partials-v1",
       }, "portable")
-    ).toThrow(/portable mixed DiT accepts only the OPT-0009 rev7 oracle/);
+    ).toThrow(/portable mixed DiT accepts only an authenticated rev7 package/);
 
     const opt0072 = {
       manifestUrl: "https://example.test/rev7-vae/manifest.json",
diff --git a/src/engines/musicgen-acestep/config.ts b/src/engines/musicgen-acestep/config.ts
index ee5ed13..9434301 100644
--- a/src/engines/musicgen-acestep/config.ts
+++ b/src/engines/musicgen-acestep/config.ts
@@ -20,8 +20,12 @@ export const REFERENCE_PAYLOAD_PREFIX_SHA256 = "18f36c6420976475af65ecd833ca56c6
 export const DIRECT_REFERENCE_MANIFEST_SHA256 = "b44a3d157009d035a8f20aa752db4ceef2fac5bd140eff13be8f7488bc978089";
 export const DIRECT_REFERENCE_MANIFEST_REMOTE_NAME = `direct-manifest-${DIRECT_REFERENCE_MANIFEST_SHA256}.json`;
 export const DIT_MANIFEST_SHA256 = "d3fc0020efcf60702db411da2fd4b93e9bb84f1437ed310aef01c892727e452f";
+export const INT8_QUALITY_PREVIEW_DIT_MANIFEST_SHA256 = "ef8355b9cffff466b018b51275923982b071234933fe8a32897915eeeb01fa36";
 export const VAE_MANIFEST_SHA256 = "36a54d79777d6826088095ba6ebc028fb4bea546368c0f0a29cd0eee8d656da7";
 
+export const ACE_DEMO_MODEL_VARIANTS = ["production", "int8-quality-preview"] as const;
+export type AceDemoModelVariant = (typeof ACE_DEMO_MODEL_VARIANTS)[number];
+
 function productionManifestUrl(localDirectory: string, productionPackage: string, payloadPrefixSha256: string, manifestRemoteName = "manifest.json"): string {
   if (ACE_MODEL_ORIGIN.startsWith("/")) {
     return new URL(`${ACE_MODEL_ORIGIN}/${localDirectory}/${manifestRemoteName}`, location.href).href;
@@ -31,14 +35,22 @@ function productionManifestUrl(localDirectory: string, productionPackage: string
 
 /** The exact production worker configuration used by the upstream demo. */
 export function aceProductionWorkerConfiguration() {
+  return aceDemoWorkerConfiguration("production");
+}
+
+/** Select the approved production model or the full-size OPT-0089 listening preview. */
+export function aceDemoWorkerConfiguration(variant: AceDemoModelVariant) {
+  const ditManifestSha256 = variant === "int8-quality-preview" ? INT8_QUALITY_PREVIEW_DIT_MANIFEST_SHA256 : DIT_MANIFEST_SHA256;
+  const ditPackage = variant === "int8-quality-preview" ? "dit-int8-fakequant" : "dit-revision7";
+  const localDitDirectory = variant === "int8-quality-preview" ? "files-fp16-dit-rev7-int8-fakequant" : "files-fp16-dit-rev7-oracle";
   return {
     manifestUrl: productionManifestUrl("files-reference", "reference", REFERENCE_PAYLOAD_PREFIX_SHA256, DIRECT_REFERENCE_MANIFEST_REMOTE_NAME),
     manifestSha256: DIRECT_REFERENCE_MANIFEST_SHA256,
     modelProfile: "reference-bf16",
     schedulingProfile: "cooperative",
     ditDensePackage: {
-      manifestUrl: productionManifestUrl("files-fp16-dit-rev7-oracle", "dit-revision7", DIT_MANIFEST_SHA256),
-      manifestSha256: DIT_MANIFEST_SHA256,
+      manifestUrl: productionManifestUrl(localDitDirectory, ditPackage, ditManifestSha256),
+      manifestSha256: ditManifestSha256,
       runtimeProfile: "opt-0009-fp16-fp32-dense-v1",
     },
     ditAttentionRuntimeProfile: "opt-0070-fixed32-quad-query32-full-self-production-v1",
diff --git a/src/engines/musicgen-acestep/style.css b/src/engines/musicgen-acestep/style.css
index ae80818..9c56082 100644
--- a/src/engines/musicgen-acestep/style.css
+++ b/src/engines/musicgen-acestep/style.css
@@ -320,6 +320,7 @@ a:hover {
 
 button,
 input,
+select,
 textarea {
   font: inherit;
 }
@@ -354,6 +355,7 @@ button:disabled {
 
 button:focus-visible,
 input:focus-visible,
+select:focus-visible,
 textarea:focus-visible,
 summary:focus-visible,
 a:focus-visible,
@@ -380,7 +382,8 @@ audio:focus-visible {
 
 textarea,
 input[type="text"],
-input[type="number"] {
+input[type="number"],
+select {
   width: 100%;
   border: 1px solid var(--control-border);
   border-radius: 8px;
@@ -402,7 +405,8 @@ textarea {
 }
 
 input[type="text"],
-input[type="number"] {
+input[type="number"],
+select {
   min-height: 42px;
   padding: 8px 10px;
 }
@@ -413,6 +417,10 @@ input::placeholder {
   opacity: 0.78;
 }
 
+select {
+  appearance: auto;
+}
+
 .unit-input {
   display: grid;
   grid-template-columns: 5rem auto;
@@ -475,6 +483,17 @@ input::placeholder {
   font-weight: 450;
 }
 
+.advanced-grid .model-variant-field {
+  grid-column: 1 / -1;
+}
+
+.model-variant-field small {
+  color: var(--muted-text);
+  font-size: 0.78rem;
+  font-weight: 450;
+  line-height: 1.45;
+}
+
 .actions {
   display: flex;
   gap: 8px;
diff --git a/src/music.ts b/src/music.ts
index 69e0495..47e3a81 100644
--- a/src/music.ts
+++ b/src/music.ts
@@ -23,7 +23,7 @@ import {
 import lightModeIcon from "./engines/musicgen-acestep/assets/light-mode.png";
 import moonIcon from "./engines/musicgen-acestep/assets/moon.png";
 
-import { aceProductionWorkerConfiguration } from "./engines/musicgen-acestep/config.js";
+import { ACE_DEMO_MODEL_VARIANTS, aceDemoWorkerConfiguration, type AceDemoModelVariant } from "./engines/musicgen-acestep/config.js";
 import { aceInferenceWorkerName } from "./engines/musicgen-acestep/worker-name.js";
 import { CRASH_BREADCRUMB_KEY, writeProgressBreadcrumb } from "./engines/musicgen-acestep/progress-breadcrumb.js";
 import { claimPendingOutput, forgetPendingOutput, reclaimOrphanedOutputs } from "./engines/musicgen-acestep/pending-output-registry.js";
@@ -31,9 +31,7 @@ import {
   formatDecimalBytes,
   formatModelDownloadAmount,
   INITIAL_MODEL_DOWNLOAD_PROGRESS,
-  isModelDownloadComplete,
   MODEL_DOWNLOAD_TOTAL_BYTES,
-  shouldShowModelDownloadNote,
   updateModelDownloadProgress,
   type ModelDownloadProgress,
 } from "./engines/musicgen-acestep/model-download-progress.js";
@@ -62,11 +60,11 @@ const bpmInput = requiredElement<HTMLInputElement>("bpm");
 const keyScaleInput = requiredElement<HTMLInputElement>("key-scale");
 const timeSignatureInput = requiredElement<HTMLInputElement>("time-signature");
 const vocalLanguageInput = requiredElement<HTMLInputElement>("vocal-language");
+const modelVariantSelect = requiredElement<HTMLSelectElement>("model-variant");
 const formError = requiredElement<HTMLParagraphElement>("form-error");
 const generateButton = requiredElement<HTMLButtonElement>("generate");
 const cancelButton = requiredElement<HTMLButtonElement>("cancel");
 const supportWarning = requiredElement<HTMLParagraphElement>("support-warning");
-const downloadNote = requiredElement<HTMLParagraphElement>("download-note");
 const progressPanel = requiredElement<HTMLElement>("progress-panel");
 const progressTitle = requiredElement<HTMLHeadingElement>("progress-title");
 const progressDetail = requiredElement<HTMLParagraphElement>("progress-detail");
@@ -74,6 +72,7 @@ const progressPercent = requiredElement<HTMLSpanElement>("progress-percent");
 const progressElement = requiredElement<HTMLProgressElement>("progress");
 const summaryDuration = requiredElement<HTMLElement>("summary-duration");
 const summaryTime = requiredElement<HTMLElement>("summary-time");
+const summaryModel = requiredElement<HTMLElement>("summary-model");
 const resultPanel = requiredElement<HTMLElement>("result-panel");
 const audioPlayer = requiredElement<HTMLAudioElement>("audio-player");
 const download = requiredElement<HTMLAnchorElement>("download");
@@ -91,10 +90,11 @@ const runtimeMetrics = requiredElement<HTMLPreElement>("runtime-metrics");
 const themeToggle = requiredElement<HTMLButtonElement>("theme-toggle");
 const themeIcon = requiredElement<HTMLImageElement>("theme-icon");
 
-const formControls = Array.from(form.querySelectorAll<HTMLInputElement | HTMLTextAreaElement>("input, textarea"));
+const formControls = Array.from(form.querySelectorAll<HTMLInputElement | HTMLSelectElement | HTMLTextAreaElement>("input, select, textarea"));
 
 let worker: Worker | undefined;
 let workerReady = false;
+let workerModelVariant: AceDemoModelVariant | undefined;
 let initializationRequestId: number | undefined;
 let activeJobId: number | undefined;
 let pendingRequest: AceGenerationRequest | undefined;
@@ -397,7 +397,7 @@ async function beginGeneration(): Promise<void> {
   diagnosticDetails = [];
   fatalGpuDiagnostic = false;
   modelProgress = INITIAL_MODEL_DOWNLOAD_PROGRESS;
-  coldDownload = !isModelDownloadComplete(cacheDetails);
+  coldDownload = false;
   setBusy(true);
   resultPanel.hidden = true;
 
@@ -411,11 +411,35 @@ async function beginGeneration(): Promise<void> {
     },
   );
 
-  if (workerReady && worker !== undefined) {
+  const modelVariant = selectedModelVariant();
+  if (workerReady && worker !== undefined && workerModelVariant === modelVariant) {
     startPendingGeneration();
     return;
   }
-  void startWorkerInitialization();
+  void initializeSelectedModel(modelVariant);
+}
+
+function selectedModelVariant(): AceDemoModelVariant {
+  const value = modelVariantSelect.value;
+  if ((ACE_DEMO_MODEL_VARIANTS as readonly string[]).includes(value)) {
+    return value as AceDemoModelVariant;
+  }
+  throw new Error(`Unknown ACE model variant ${value}`);
+}
+
+async function initializeSelectedModel(modelVariant: AceDemoModelVariant): Promise<void> {
+  if (worker !== undefined) {
+    try {
+      await disposeWorker();
+    } catch (error) {
+      failOperation(`Could not switch models: ${errorMessage(error)}`, true);
+      return;
+    }
+  }
+  if (!busy || pendingRequest === undefined || pageLifecycle.signal.aborted) {
+    return;
+  }
+  await startWorkerInitialization(modelVariant);
 }
 
 function readGenerationRequest(): AceGenerationRequest {
@@ -457,7 +481,7 @@ function readGenerationRequest(): AceGenerationRequest {
   };
 }
 
-async function startWorkerInitialization(): Promise<void> {
+async function startWorkerInitialization(modelVariant: AceDemoModelVariant): Promise<void> {
   resetWorker();
   const acquisition = new AbortController();
   cacheAcquisition = acquisition;
@@ -476,12 +500,13 @@ async function startWorkerInitialization(): Promise<void> {
     });
     worker.addEventListener("message", onWorkerMessage);
     worker.addEventListener("error", onWorkerError);
+    workerModelVariant = modelVariant;
     initializationRequestId = nextRequestId++;
     setIndeterminateProgress("Preparing model", "Checking WebGPU and browser storage");
     worker.postMessage({
       type: "initialize",
       requestId: initializationRequestId,
-      configuration: aceProductionWorkerConfiguration(),
+      configuration: aceDemoWorkerConfiguration(modelVariant),
       modelSource: "cache-or-network",
       reportProgress: true,
       reportDiagnostics: true,
@@ -546,6 +571,9 @@ function onWorkerMessage(event: MessageEvent<unknown>): void {
   switch (message.type) {
     case "initialization-progress": {
       if (message.requestId !== initializationRequestId) return;
+      if (message.progress.message?.startsWith("network:") === true) {
+        coldDownload = true;
+      }
       const updated = updateModelDownloadProgress(modelProgress, message);
       if (updated !== modelProgress) {
         modelProgress = updated;
@@ -671,6 +699,7 @@ async function publishResult(result: AceGenerationResult): Promise<void> {
       modelManifestSha256: result.modelManifestSha256,
       metrics: result.metrics,
     };
+    summaryModel.textContent = workerModelVariant === "int8-quality-preview" ? "INT8 quality preview" : "Production";
     modelProgress = updateModelDownloadProgress(modelProgress, {
       stage: "vae-load",
       message: "network: complete 168791552/168791552 bytes",
@@ -857,6 +886,7 @@ function resetWorker(): void {
   worker?.terminate();
   worker = undefined;
   workerReady = false;
+  workerModelVariant = undefined;
   initializationCancelRequested = false;
   workerDetails = undefined;
   void releaseRuntimeLock?.();
@@ -879,7 +909,6 @@ async function refreshCacheInfo(): Promise<void> {
     cacheDetails = undefined;
     cacheStatus.textContent = `Could not inspect model storage: ${errorMessage(error)}`;
   }
-  downloadNote.hidden = !shouldShowModelDownloadNote(cacheDetails);
   updateActionAvailability();
   updateRuntimeDetails();
 }
@@ -923,6 +952,7 @@ async function disposeWorker(): Promise<void> {
       current.terminate();
       worker = undefined;
       workerReady = false;
+      workerModelVariant = undefined;
       const release = releaseRuntimeLock;
       releaseRuntimeLock = undefined;
       await release?.();
diff --git a/test/music-model-variant.test.ts b/test/music-model-variant.test.ts
new file mode 100644
index 0000000..1add0e6
--- /dev/null
+++ b/test/music-model-variant.test.ts
@@ -0,0 +1,36 @@
+import { readFileSync } from "node:fs";
+
+import { describe, expect, it } from "vitest";
+
+import {
+  INT8_QUALITY_PREVIEW_DIT_MANIFEST_SHA256,
+  aceDemoWorkerConfiguration,
+  aceProductionWorkerConfiguration,
+} from "../src/engines/musicgen-acestep/config.js";
+
+describe("music model variants", () => {
+  it("keeps production as the default and isolates the preview trust root", () => {
+    const production = aceProductionWorkerConfiguration();
+    const preview = aceDemoWorkerConfiguration("int8-quality-preview");
+
+    expect(aceDemoWorkerConfiguration("production")).toEqual(production);
+    expect(preview).toEqual({
+      ...production,
+      ditDensePackage: {
+        ...production.ditDensePackage,
+        manifestUrl: `https://huggingface.co/FluidInference/ace-step-webgpu-models/resolve/main/v1/dit-int8-fakequant/${INT8_QUALITY_PREVIEW_DIT_MANIFEST_SHA256}/manifest.json`,
+        manifestSha256: INT8_QUALITY_PREVIEW_DIT_MANIFEST_SHA256,
+      },
+    });
+  });
+
+  it("labels the preview as experimental and full-size", () => {
+    const html = readFileSync(new URL("../music.html", import.meta.url), "utf8");
+
+    expect(html).toContain("Production (recommended)");
+    expect(html).toContain("INT8 quality preview (experimental)");
+    expect(html).toContain("preview tests audible quantization effects but is not compressed");
+    expect(html).toContain("separate 3.02 GB DiT download");
+    expect(html).toMatch(/option value="production" selected/);
+  });
+});
diff --git a/test/music-preparation.test.ts b/test/music-preparation.test.ts
index f711d4c..07a12ac 100644
--- a/test/music-preparation.test.ts
+++ b/test/music-preparation.test.ts
@@ -43,8 +43,11 @@ function page() {
     updateRuntimeDetails: () => {},
     workerDetails: undefined,
     workerReady: true,
+    workerModelVariant: "production",
     worker: { postMessage: vi.fn() },
     startPendingGeneration: vi.fn(),
+    selectedModelVariant: () => "production",
+    initializeSelectedModel: vi.fn(),
     setDeterminateProgress: vi.fn(),
     setIndeterminateProgress: vi.fn(),
   };
@@ -96,6 +99,46 @@ it("does not start generation when the page closes during cleanup", async () =>
   expect(state.generationPreparation).toBeUndefined();
 });
 
+it("reinitializes instead of reusing a worker for another model variant", async () => {
+  const { state, finish } = page();
+  state.selectedModelVariant = () => "int8-quality-preview";
+  const running = state.beginGeneration();
+  finish();
+  await running;
+
+  expect(state.startPendingGeneration).not.toHaveBeenCalled();
+  expect(state.initializeSelectedModel).toHaveBeenCalledExactlyOnceWith("int8-quality-preview");
+});
+
+it("disposes the current worker before initializing the selected model", async () => {
+  const source = readFileSync(new URL("../src/music.ts", import.meta.url), "utf8");
+  const ast = ts.createSourceFile("music.ts", source, ts.ScriptTarget.Latest, true);
+  const handler = ast.statements.find((node) => ts.isFunctionDeclaration(node) && node.name?.text === "initializeSelectedModel");
+  expect(handler).toBeDefined();
+  let finishDisposal!: () => void;
+  const disposal = new Promise<void>((resolve) => {
+    finishDisposal = resolve;
+  });
+  const state: Record<string, any> = {
+    worker: {},
+    busy: true,
+    pendingRequest: { prompt: "Piano instrumental" },
+    pageLifecycle: new AbortController(),
+    disposeWorker: vi.fn(() => disposal),
+    startWorkerInitialization: vi.fn(),
+    failOperation: vi.fn(),
+    errorMessage: String,
+  };
+  vm.createContext(state);
+  vm.runInContext(ts.transpile(handler!.getText(ast), { target: ts.ScriptTarget.ES2022 }), state);
+
+  const switching = state.initializeSelectedModel("int8-quality-preview");
+  expect(state.startWorkerInitialization).not.toHaveBeenCalled();
+  finishDisposal();
+  await switching;
+  expect(state.startWorkerInitialization).toHaveBeenCalledExactlyOnceWith("int8-quality-preview");
+});
+
 it("does not generate when ready was already queued before page cancellation", async () => {
   const source = readFileSync(new URL("../src/music.ts", import.meta.url), "utf8");
   const ast = ts.createSourceFile("music.ts", source, ts.ScriptTarget.Latest, true);

From a4eb173b7cd80bb9b1558346d94aedd8bcc51c02 Mon Sep 17 00:00:00 2001
From: Alex-Wengg <hanweng9@gmail.com>
Date: Tue, 15 Sep 2026 16:06:15 -0400
Subject: [PATCH 2/2] docs(acestep): close OPT-0090 integration

---
 packages/acestep/optimization/LEDGER.md                         | 2 +-
 .../optimization/experiments/OPT-0090-int8-quality-preview.md   | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/packages/acestep/optimization/LEDGER.md b/packages/acestep/optimization/LEDGER.md
index ea0b91d..dacfe27 100644
--- a/packages/acestep/optimization/LEDGER.md
+++ b/packages/acestep/optimization/LEDGER.md
@@ -96,7 +96,7 @@ Next available ID: `OPT-0091`.
 | OPT-0087 | Planner package-native dense GEMV | Route the already exact direct M1/M2 packed-BF16 kernel through all real planner decode layers and tied-head slices so the package can realize its useful bandwidth without OPT-0083's artificial submit-and-drain boundary around every isolated layer sample | positive | pending-integration | Actual Chrome preserved every full logit, appended K/V-cache evidence word, status, token, Philox cursor, topology, cancellation, and lifecycle result. Direct B won `16/16`, lowered every M1/M2 layer/head/model/complete median, reached `2.229363x` aggregate layer speedup, and reduced aggregate model-through-readback median `272.150 → 156.200 ms` (`115.950 ms`, projected `117.110 s/1,010`). Cleanup balanced `3,683/3,683` buffers and `133/133` maps with zero live resources. The fresh nominal launch passed; all later level-1/2 thermal observations are disclosed. Complete trajectory and planner-enabled product gates remain mandatory | [record](experiments/OPT-0087-planner-package-native-low-row-gemv.md), [result](results/OPT-0087/result.json) | allocation `84470d877ba69c6ef7821871f9bdb82d8c757747`; unchanged OPT-0083 direct kernel; implementation/harness `980623e0f9ec918f7a537b6a90e341adaad6d4b0`; pending integration gates |
 | OPT-0088 | Portable device support | Every subgroup-dependent production owner (OPT-0032/0037 dense K4, OPT-0051 K7 row-reuse, OPT-0048 ConvTranspose K4, attention query8/quad-query) can gain a workgroup-memory counterpart consuming the unchanged hosted packages, selected by the existing execution-profile machinery, so adapters without `subgroups` (Safari, Firefox, iOS) run the production graph instead of failing `FEATURE_UNAVAILABLE`; compatibility experiment, bounded slowdown expected and reported, `shader-f16` stays fail-closed | pending | pending-integration | Portable dense/K7/ConvTranspose owners landed with test-enforced bit-identical arithmetic (byte-equal WGSL arithmetic sections, re-exported rev7/rev8 index math); attention routes to the existing portable oracle (reordered-rounding vs subgroup reduction). End-to-end masked-subgroups waveform and timing gates pending | [record](experiments/OPT-0088-portable-no-subgroup-production-path.md) | kernels c272b2d/c48c050/373e90a; selection wiring pending |
 | OPT-0089 | DiT weight quantization | Weight-only symmetric int8 (per-32-K-block fp16 scales, round-to-nearest, clamp ±127) fake-quantization of all 264 rev7 DiT GEMM tensors, dequantized in place and run through the completely unchanged production graph, preserves end-to-end 30 s waveform quality within a small numerical envelope, so an int8-resident DiT (~1.51 GB + scales) is a credible answer to the observed iPhone 17 Safari OOM kill at `1,789,925,376 / 3,020,808,192` uploaded bytes (layer 14/24); pure quantization-damage gate, zero kernel changes, distinct mechanism from abandoned OPT-0058 activation-quantized DP4a | positive | benchmark-only | Per-tensor damage uniform and small: NRMSE `0.00515–0.00634` (median `0.00559`), min SNR `43.96 dB`, no outlier tensor/family, so no fp16-retention map needed. Determinism gate reproduced the pinned fp16 baseline WAV byte-exactly, then fake-quant vs fp16 on identical seeds gave lo-fi/12345 waveform NRMSE `0.0669` (Pearson `0.99777`, LSD ≈`3.5 dB`, RMS Δ `−0.050 dB`) and latin/424242 NRMSE `0.2268` (Pearson `0.97460`, LSD ≈`4.7 dB`, RMS Δ `+0.089 dB`; per-second max `5.19` is a near-silent-ending small-denominator artifact) — trajectory divergence of the 8-evaluation sampler, not noise-like corruption; zero non-finite samples and exact peak parity. Projected int8 DiT phase peak ≈`1.862 GB` tracked GPU (`1.51 GB` int8 + `94 MB` scales + fp16 norms/shared + measured `127 MB` overhead) versus the observed iPhone 17 kill at ≈`1.920 GB` — plausibly fits, marginal ≈`58 MB` margin; int8 kernel work justified, listening gate mandatory before any product claim | [record](experiments/OPT-0089-dit-int8-weight-fake-quant-gate.md), [quant result](results/OPT-0089/quant-error.json), [waveform result](results/OPT-0089/waveform-metrics.json) | `scripts/requantize-dit-int8.py` (repo root); fake-quant package `ef8355b9…` (models-local, not hosted); benchmark-only, no kernel or production change |
-| OPT-0090 | DiT quantization listening preview | The authenticated OPT-0089 fake-quant package can be exposed as an explicit opt-in listening comparison without changing the production default, weakening package identity checks, or claiming packed-int8 size/runtime benefits | positive | pending-integration | Reproduced the recorded `ef8355b9…` package and byte-identical quant report, published the immutable full-size artifact, then added an Advanced listening selector with orderly model switching and explicit size/non-production disclosure. 43 web integration, 4 unit, and 2,029 ACE tests plus typecheck, formatting, and production build passed; fresh browser/GPU listening remains external | [record](experiments/OPT-0090-int8-quality-preview.md) | branch `feat/int8-quality-preview`; merge/deploy pending |
+| OPT-0090 | DiT quantization listening preview | The authenticated OPT-0089 fake-quant package can be exposed as an explicit opt-in listening comparison without changing the production default, weakening package identity checks, or claiming packed-int8 size/runtime benefits | positive | integrated | Reproduced the recorded `ef8355b9…` package and byte-identical quant report, published the immutable full-size artifact, then added an Advanced listening selector with orderly model switching and explicit size/non-production disclosure. 43 web integration, 4 unit, and 2,029 ACE tests plus typecheck, formatting, production build, and all PR checks passed; fresh browser/GPU listening remains external | [record](experiments/OPT-0090-int8-quality-preview.md) | implementation `2816d6b83185cbde5abd6780a9f5ac903a655154`; artifact `ac50b5c854fb044ce058acb91d4cd9ab82d99cfa` |
 
 Experiment IDs are allocated before code changes, never reused, and never
 removed from this table.
diff --git a/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md b/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
index 2885e49..3c73c3a 100644
--- a/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
+++ b/packages/acestep/optimization/experiments/OPT-0090-int8-quality-preview.md
@@ -3,7 +3,7 @@
 ## Status
 
 - Evidence: `positive` (artifact and integration gates; external listening pending)
-- Disposition: `pending-integration`
+- Disposition: `integrated`
 - Date: 2026-09-15
 - Risk class: `approximate`, opt-in only