AlbertXXuu · AlbertXXuu · Aug 31, 2026 · Aug 31, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,6 +2,29 @@
 
 All notable changes to this project will be documented in this file.
 
+## [1.1.2] - 2026-08-31
+
+### Fixed
+
+- Remove framework-added padding and button margins from the Studio header navigation so the
+  packaged interface retains the canonical AlvenX header height across the audited desktop widths.
+
+### Clarified
+
+- Define the current CUDA `peak_gpu_memory_mb` value as PyTorch allocator memory rather than total
+  process or device VRAM, and require alternative measurement boundaries to remain separate.
+- Require an observed low-memory or latency problem before adding a quantized or alternative
+  runtime comparison, with matched revisions, tasks, prompts, generation settings, and metrics.
+- Record that useful boundaries from two early documentation-only project remnants were retained
+  without merging their abandoned scope into OpenMultimodalLab.
+
+The immutable `v1.0.0` research tag, datasets, raw results, report bundle, and published
+measurements remain unchanged. Source validation is recorded in
+`docs/reports/v1.1.2-patch-validation.md`; final remote provenance belongs to the AlvenX
+post-closure audit and GitHub Release.
+
+Compare: [`v1.1.1...v1.1.2`](https://github.com/AlbertXXuu/OpenMultimodalLab/compare/v1.1.1...v1.1.2)
+
 ## [1.1.1] - 2026-08-31
 
 ### Fixed

diff --git a/README.md b/README.md
@@ -12,10 +12,10 @@
 
 [简体中文](README.zh-CN.md)
 
-> Version status: current software is the `v1.1.1` closure maintenance release. The research and
-> evidence baseline remains the immutable `v1.0.0` public release. The closure line adds
-> presentation and maintenance work, not a new benchmark claim; `v1.1.1` corrects the documented
-> current-software clone target. See the
+> Version status: current software is the `v1.1.2` maintenance patch. The research and evidence
+> baseline remains the immutable `v1.0.0` public release. This patch aligns the packaged Studio
+> header with the canonical cross-product geometry and clarifies measurement and future-runtime
+> boundaries; it does not add benchmark evidence or change published results. See the
 > [maintenance policy](docs/MAINTENANCE.md) and [portfolio evidence](PORTFOLIO.md).
 
 A local-first, reproducible benchmark toolkit for answering a practical
@@ -153,7 +153,7 @@ The core path does not download a model and works on Python 3.11, 3.12, or
 3.13. The primary examples use Windows PowerShell:
 
 ```powershell
-git clone --branch v1.1.1 --depth 1 https://github.com/AlbertXXuu/OpenMultimodalLab.git
+git clone --branch v1.1.2 --depth 1 https://github.com/AlbertXXuu/OpenMultimodalLab.git
 cd OpenMultimodalLab
 
 py -3.11 -m venv .venv

diff --git a/README.zh-CN.md b/README.zh-CN.md
@@ -12,9 +12,9 @@
 
 [English](README.md)
 
-> 版本状态：当前软件是 `v1.1.1` 收尾维护版；研究与证据基线仍是不可变的 `v1.0.0`
-> 公开正式版。收尾版本只处理展示与维护，不产生新的基准结论；`v1.1.1` 修正文档中的
-> 当前软件克隆目标。详见
+> 版本状态：当前软件是 `v1.1.2` 维护修补版。研究与证据基线仍是不可变的 `v1.0.0`
+> 公开正式版。本次修补让打包后的 Studio Header 回到跨产品统一几何，并澄清测量与
+> 未来运行时边界；它不新增基准证据，也不改变已发布结果。详见
 > [维护政策](docs/MAINTENANCE.md)与[作品证据页](PORTFOLIO.md)。
 
 一个本地优先、强调可复现证据的多模态模型评测工具。它要回答的不是“哪个
@@ -132,7 +132,7 @@ flowchart LR
 核心路径不会下载模型，可在 Python 3.11、3.12 或 3.13 运行：
 
 ```powershell
-git clone --branch v1.1.1 --depth 1 https://github.com/AlbertXXuu/OpenMultimodalLab.git
+git clone --branch v1.1.2 --depth 1 https://github.com/AlbertXXuu/OpenMultimodalLab.git
 cd OpenMultimodalLab
 
 py -3.11 -m venv .venv

diff --git a/TASKS.md b/TASKS.md
@@ -1,8 +1,8 @@
 # 当前任务清单
 
-> `v1.0.0` 研究与证据基线已于 2026-08-10 正式发布，`v1.1.1` 展示与维护收尾版
-> 已于 2026-08-31 发布。后续工作不改写 v1.0.0 的标签、原始结果或证据；当前进入
-> 维护模式，不产生新的研究主张。
+> `v1.0.0` 研究与证据基线已于 2026-08-10 正式发布；当前公开维护软件是 `v1.1.2`。
+> 该展示与维护修补版不改写 v1.0.0 的标签、原始结果或证据；项目保持维护模式，
+> 不产生新的研究主张。
 
 ## Now：发布后可靠性审查
 

diff --git a/docs/08-risks-and-decisions.md b/docs/08-risks-and-decisions.md
@@ -11,7 +11,8 @@
 - v1.0 正式实验已在该 8GB GPU 上完成；可用磁盘空间属于运行时状态，不在
   文档中长期固定。
 - 维护目标是每周 2～3 次有证据的更新；具体投入时间按学业安排调整。
-- 现有工作区内已有 LocalModelLab 和 AgentReliabilityLab，本项目不覆盖它们。
+- 早期 LocalModelLab 与 AgentReliabilityLab 文档中的有效方法边界已分别提炼到本项目路线图、
+  评测协议和 AlvenX incubator；旧的无代码研究残余已清理，不再构成并行项目。
 
 ## 2. 当前假设
 
@@ -44,6 +45,9 @@
 
 原因：三者研究问题不同，独立仓库更容易形成清楚的 README、路线图和面试故事。
 
+这是创建仓库时的历史决定。后续整理仅提炼仍有效的方法约束并清理未实现旧文档，未把旧路线图或
+通用 Agent 平台范围并入 OpenMultimodalLab。
+
 ### D-002：先做评测核心
 
 第一阶段使用确定性 `mock` 后端，不立即安装大型模型。

diff --git a/docs/MAINTENANCE.md b/docs/MAINTENANCE.md
@@ -1,6 +1,6 @@
 # Maintenance policy
 
-Current public release: `v1.1.1`
+Current public and maintained release: `v1.1.2`
 Research/evidence baseline: immutable `v1.0.0`
 Development mode: maintenance; no new research claim
 

diff --git a/docs/evaluation-protocol.md b/docs/evaluation-protocol.md
@@ -233,6 +233,13 @@ The native Transformers visual-text adapters record the same fields:
 - `latency_ms`: adapter invocation plus deterministic evaluation in the runner;
 - `peak_gpu_memory_mb`: maximum CUDA memory allocated during generation.
 
+For the current PyTorch CUDA adapters, `peak_gpu_memory_mb` means allocator
+memory, not total process VRAM or the peak for the whole device. An alternative
+runtime must declare its measurement boundary. An NVML measurement must retain
+the idle baseline, sampling interval, and whether another process contaminated
+the observation. Results from different memory boundaries must not share one
+memory ranking.
+
 `output_tokens_per_second` counts generated token IDs, including terminal
 special tokens, over `generation_ms`. `decode_tokens_per_second` excludes the
 first generated token and divides by `generation_ms - ttft_ms`.

diff --git a/docs/post-v1-roadmap.md b/docs/post-v1-roadmap.md
@@ -61,8 +61,13 @@ An adapter that only makes another model import successfully is not enough.
 
 ### Later: conditional scope
 
-- Quantized or alternative runtimes when they enable a measured hardware or
-  latency use case.
+- Quantized or alternative runtimes only after a concrete low-memory or latency
+  problem is observed. Compare quality, TTFT, decode speed, memory, OOM behavior,
+  and effective context under the same immutable model revision, task set,
+  prompt/chat template, and generation protocol. Differences across GGUF, AWQ,
+  GPTQ, NF4, or different backends cannot be attributed to bit width alone.
+  Long-context/OOM gradients and 50-run stability checks begin only after a
+  candidate signal exists; they are not a standing benchmark matrix.
 - Additional task families when their media can be regenerated and reviewed.
 - A UI or remote execution layer only when CLI user evidence shows it removes a
   real barrier without weakening provenance.

diff --git a/docs/reports/v1.1.2-patch-validation.md b/docs/reports/v1.1.2-patch-validation.md
@@ -0,0 +1,57 @@
+# OpenMultimodalLab v1.1.2 patch validation
+
+- Status: **VALIDATED RELEASE SOURCE**
+- Source validated: `2026-08-31` (`Asia/Shanghai`)
+- Owner authorization: AlvenX workspace decision `D-032`, recorded `2026-08-31`
+- Release identity: `v1.1.2`
+- Research/evidence baseline: immutable `v1.0.0`
+
+## Patch meaning
+
+This maintenance patch addresses one observed Studio layout defect and three bounded documentation
+clarifications:
+
+1. The Studio header navigation no longer retains framework-added outer padding or button margins,
+   so its shell returns to the canonical AlvenX product-header height.
+2. The current CUDA `peak_gpu_memory_mb` field is explicitly scoped to PyTorch allocator memory,
+   not total process VRAM or whole-device peak usage. Different measurement boundaries remain
+   separate rather than being mixed into one memory ranking.
+3. Quantized or alternative runtimes enter the roadmap only after an observed memory or latency
+   problem and must use matched revisions, tasks, prompts, generation settings, and reported
+   boundaries.
+4. The project decision record now states that useful constraints from two early documentation-only
+   remnants were retained without absorbing their abandoned product scope.
+
+The patch does not add a model, adapter, dependency, task, dataset, schema, benchmark run, or
+research conclusion.
+
+## Frozen v1.0.0 surface
+
+The annotated `v1.0.0` tag, reviewed task and media identities, pinned model revisions, formal run
+configuration, raw JSONL results, manifests, report bundle, and published measurements are not
+modified. The evaluation-protocol edit is a post-release measurement-boundary clarification; it
+does not rewrite the protocol or evidence bytes reachable from `v1.0.0`, and it does not recompute
+or reinterpret the released numeric results.
+
+`CITATION.cff`, `docs/release-approvals.json`, historical release audits, dependencies, and GitHub
+Actions workflows remain unchanged.
+
+## Source verification
+
+| Gate | Result | Evidence |
+| --- | --- | --- |
+| Version identity | PASS | Package metadata, import version, Studio badge, bilingual README, changelog, maintenance policy, task status, and this audit identify software `v1.1.2`; evidence stays `v1.0.0`. |
+| Targeted package and Studio tests | PASS | Pytest collected `29` cases on Python `3.13.5`: `27` passed and `2` optional-environment cases skipped. The Header padding/margin regression and cross-surface version consistency test passed. |
+| Complete local unit suite | PASS | `unittest discover` executed `202` cases on Python `3.13.5` / Windows 11 build `26200`: `199` passed and `3` optional-environment cases skipped. A separate pytest collection covered `207` cases: `204` passed and `3` skipped. |
+| Repository and evidence checks | PASS | Repository audit: `195` text files, `278` Markdown links, and `1,056` JSON/JSONL documents. Report verification matched `4` sources and `5` outputs; runtime-license policy passed; offline contributor smoke completed `3/3` records and reported package `1.1.2`. |
+| Frozen v1.0 technical gate | PASS, bounded scope | The existing technical-strict readiness check passed `18/18`; it revalidates the immutable v1.0 evidence contract without changing the released evidence. |
+| AlvenX brand/workspace validation | PASS | Brand revision `2026-08-24.1` validated `14` hashed assets, `2` canonical masters, and `3` wordmark lockups; workspace revision `2026-08-30.1` passed. |
+| Interface regression review | PASS | Edge 152, Chrome 151, and Firefox 153 passed 900/1024/1280/1440/1600 px with the canonical 74.48 px Header, 44 px controls, and no horizontal overflow. |
+
+## Remote provenance boundary
+
+This in-repository record validates the release source without embedding a commit identity that can
+only be known after the source is committed. The final remote commit, GitHub Actions runs,
+distribution filenames and hashes, and installed-artifact checks are recorded in the workspace-level
+AlvenX post-closure audit and the GitHub Release. Those external records bind the published source
+and artifacts; this document does not manufacture that provenance in advance.
diff --git a/pyproject.toml b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "openmultimodal-lab"
-version = "1.1.1"
+version = "1.1.2"
 description = "The local-first, reproducible multimodal benchmark engine behind AlvenX."
 readme = "README.md"
 requires-python = ">=3.11"

diff --git a/src/openmultimodal_lab/__init__.py b/src/openmultimodal_lab/__init__.py
@@ -3,4 +3,4 @@
 from .models import EvaluationTask, ModelOutput, RunRecord, ScoringConfig
 
 __all__ = ["EvaluationTask", "ModelOutput", "RunRecord", "ScoringConfig"]
-__version__ = "1.1.1"
+__version__ = "1.1.2"
diff --git a/src/openmultimodal_lab/assets/studio/alvenx-studio.css b/src/openmultimodal_lab/assets/studio/alvenx-studio.css
@@ -122,13 +122,14 @@ body { margin: 0; }
   display: flex;
   gap: 5px;
   margin-left: auto;
-  padding: 4px;
+  padding: 0;
   border-radius: 999px;
   background: rgb(255 255 255 / 28%);
 }
 .header-nav button {
   min-width: 44px;
   min-height: 44px;
+  margin: 0 !important;
   padding: 9px 13px;
   border: 0;
   border-radius: 999px;

diff --git a/src/openmultimodal_lab/studio_assets.py b/src/openmultimodal_lab/studio_assets.py
@@ -103,7 +103,7 @@ def _studio_theme() -> str:
     <button type="button" data-studio-tab="reports">Reports</button>
     <button type="button" data-studio-tab="method">Method</button>
   </nav>
-  <span class="local-badge"><i aria-hidden="true"></i>Studio v1.1.1 · Evidence v1.0.0</span>
+  <span class="local-badge"><i aria-hidden="true"></i>Studio v1.1.2 · Evidence v1.0.0</span>
 </header>
 """.strip()
 

diff --git a/tests/test_package_metadata.py b/tests/test_package_metadata.py
@@ -5,24 +5,62 @@
 from pathlib import Path
 
 from openmultimodal_lab import __version__
+from openmultimodal_lab.studio_assets import BRAND_HEADER_HTML
 
 
 PROJECT_ROOT = Path(__file__).resolve().parents[1]
+CURRENT_SOFTWARE_VERSION = "1.1.2"
 
 
 class PackageMetadataTests(unittest.TestCase):
-    def test_candidate_version_matches_closure_version(self) -> None:
+    def test_current_version_matches_maintenance_patch(self) -> None:
         pyproject = tomllib.loads(
             (PROJECT_ROOT / "pyproject.toml").read_text(encoding="utf-8")
         )
 
-        self.assertEqual(__version__, "1.1.1")
+        self.assertEqual(__version__, CURRENT_SOFTWARE_VERSION)
         self.assertEqual(pyproject["project"]["version"], __version__)
         self.assertIn(
             "Development Status :: 4 - Beta",
             pyproject["project"]["classifiers"],
         )
 
+    def test_current_software_version_surfaces_agree(self) -> None:
+        tagged_version = f"v{CURRENT_SOFTWARE_VERSION}"
+        expected_markers = {
+            "README.md": (
+                f"current software is the `{tagged_version}` maintenance patch",
+                f"git clone --branch {tagged_version} --depth 1",
+            ),
+            "README.zh-CN.md": (
+                f"当前软件是 `{tagged_version}` 维护修补版",
+                f"git clone --branch {tagged_version} --depth 1",
+            ),
+            "CHANGELOG.md": (
+                f"## [{CURRENT_SOFTWARE_VERSION}] - 2026-08-31",
+                f"compare/v1.1.1...{tagged_version}",
+            ),
+            "docs/MAINTENANCE.md": (
+                f"Current public and maintained release: `{tagged_version}`",
+            ),
+            "TASKS.md": (f"当前公开维护软件是 `{tagged_version}`",),
+            "docs/reports/v1.1.2-patch-validation.md": (
+                f"# OpenMultimodalLab {tagged_version} patch validation",
+                "Status: **VALIDATED RELEASE SOURCE**",
+            ),
+        }
+
+        for relative_path, markers in expected_markers.items():
+            text = (PROJECT_ROOT / relative_path).read_text(encoding="utf-8")
+            for marker in markers:
+                with self.subTest(path=relative_path, marker=marker):
+                    self.assertIn(marker, text)
+
+        self.assertIn(
+            f"Studio {tagged_version} · Evidence v1.0.0",
+            BRAND_HEADER_HTML,
+        )
+
     def test_public_attribution_uses_the_owner_approved_identity(self) -> None:
         pyproject = tomllib.loads(
             (PROJECT_ROOT / "pyproject.toml").read_text(encoding="utf-8")

diff --git a/tests/test_studio.py b/tests/test_studio.py
@@ -324,11 +324,17 @@ def test_brand_assets_are_local_and_wordmark_is_not_interactive(self) -> None:
         self.assertIn('data-studio-tab="run"', BRAND_HEADER_HTML)
         self.assertIn('data-studio-tab="reports"', BRAND_HEADER_HTML)
         self.assertIn('data-studio-tab="method"', BRAND_HEADER_HTML)
-        self.assertIn("Studio v1.1.1 · Evidence v1.0.0", BRAND_HEADER_HTML)
+        self.assertIn("Studio v1.1.2 · Evidence v1.0.0", BRAND_HEADER_HTML)
         self.assertIn("Evidence v1.0.0", BRAND_HEADER_HTML)
         self.assertIn('document.title = "OpenMultimodalLab · AlvenX"', STUDIO_NAV_JS)
         self.assertIn("top: 14px", STUDIO_CSS)
         self.assertGreaterEqual(STUDIO_CSS.count("z-index: 100"), 2)
+        header_nav_css = STUDIO_CSS.split(".header-nav {", 1)[1].split("}", 1)[0]
+        header_nav_button_css = STUDIO_CSS.split(".header-nav button {", 1)[1].split(
+            "}", 1
+        )[0]
+        self.assertIn("padding: 0;", header_nav_css)
+        self.assertIn("margin: 0 !important;", header_nav_button_css)
         self.assertIn(
             "width: calc(min(100%, 1480px) - 2 * clamp(18px, 5vw, 74px))",
             STUDIO_CSS,