DaoyuanLi2816 · Sep 15, 2026 · Sep 15, 2026 · Sep 15, 2026
diff --git a/.github/workflows/docs.yml b/.github/workflows/docs.yml
@@ -23,6 +23,8 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          fetch-depth: 0
       - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
         with:
           python-version: "3.12"
@@ -31,6 +33,7 @@ jobs:
         run: python -m pip install -e ".[dev]"
       - name: Verify generated documentation artifacts
         run: |
+          python scripts/apply_docs_editorial_overlay.py --check
           python scripts/refresh_visual_assets.py --check
           python scripts/publish_alignment_lab_artifacts.py --check
           python scripts/publish_verl_bridge_diagrams.py --check
@@ -45,6 +48,21 @@ jobs:
         run: playwright install --with-deps chromium
       - name: Inspect four documentation viewports
         run: python scripts/check_docs_visual.py --site site --screenshots docs-visual-screenshots
+      - name: Verify the release-bound stable editorial overlay
+        shell: bash
+        run: |
+          stable_tag="$(git tag --list 'v*' --sort=-v:refname | head -n 1)"
+          git worktree add --detach "${RUNNER_TEMP}/editorial-stable" "$stable_tag"
+          python scripts/apply_docs_editorial_overlay.py --stable-source "${RUNNER_TEMP}/editorial-stable"
+          python scripts/release_state.py --docs-overlay "${RUNNER_TEMP}/editorial-stable"
+          (
+            cd "${RUNNER_TEMP}/editorial-stable"
+            mkdocs build --strict --site-dir "${RUNNER_TEMP}/editorial-site"
+          )
+          python scripts/refresh_visual_assets.py --overlay-site "${RUNNER_TEMP}/editorial-site" \
+            --stable-source "${RUNNER_TEMP}/editorial-stable"
+          python scripts/check_docs_visual.py --site "${RUNNER_TEMP}/editorial-site" \
+            --screenshots docs-visual-screenshots/stable
       - name: Upload browser screenshots
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
@@ -81,13 +99,15 @@ jobs:
           fi
           test -n "$stable_tag"
           git worktree add --detach "${RUNNER_TEMP}/stable-checkout" "$stable_tag"
-          # Release content stays pinned; navigation names the current channels.
+          # Apply only reviewed, commit/hash-bound prose fixes; APIs stay pinned.
+          python scripts/apply_docs_editorial_overlay.py --stable-source "${RUNNER_TEMP}/stable-checkout"
+          # Navigation names the current channels.
           python scripts/release_state.py --docs-overlay "${RUNNER_TEMP}/stable-checkout"
           (
             cd "${RUNNER_TEMP}/stable-checkout"
             mkdocs build --strict --site-dir "${GITHUB_WORKSPACE}/site"
           )
-          # Keep stable prose/API content tag-pinned. Overlay only SVG layout
+          # Keep stable API content tag-pinned. Overlay SVG layout
           # corrections after proving the stable scientific inputs unchanged.
           python scripts/refresh_visual_assets.py --overlay-site site \
             --stable-source "${RUNNER_TEMP}/stable-checkout"

diff --git a/CITATION.cff b/CITATION.cff
@@ -1,5 +1,5 @@
 cff-version: 1.2.0
-title: "miniVERL: Run verl experiment semantics on one GPU"
+title: "miniVERL: verl for a single consumer GPU"
 message: "If you use miniVERL in your work, please cite it as below."
 type: software
 version: 0.16.0
@@ -8,17 +8,13 @@ license: Apache-2.0
 repository-code: "https://github.com/DaoyuanLi2816/mini-verl"
 url: "https://github.com/DaoyuanLi2816/mini-verl"
 abstract: >-
-  miniVERL compiles supported verl experiment semantics for one NVIDIA GPU.
-  Its versioned profiles consume resolved configuration and Parquet prompts,
-  run PPO with an independent critic or critic-free reward-driven algorithms,
-  and support on-policy distillation with explicit teachers. Typed trajectories,
-  role identities, transactional checkpoints and exact replay make local runs
-  inspectable and resumable. Portable PEFT, critic, Parquet and provenance
-  bundles support handoff toward pinned upstream verl. Installed PPO and GRPO
-  walkthroughs exercise the complete workflow. Runtime qualification is measured
-  on one RTX 4080 and kept separate from task-quality research; distributed
-  execution is outside the local runtime. Frozen studies preserve their original
-  negative and mixed results.
+  miniVERL brings verl workflows to a single consumer NVIDIA GPU. It runs PPO,
+  GRPO and on-policy distillation using documented verl configuration profiles,
+  with native recipes for SFT, DPO and offline knowledge distillation. Training
+  roles share the GPU in phases. Users can inspect rollouts and rewards, resume
+  model and optimizer state exactly, and export PEFT adapters, Parquet data and
+  configuration bundles. The repository includes reproducible systems measurements
+  and task studies with their original data, methods and outcomes.
 authors:
   - family-names: Li
     given-names: Daoyuan

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
@@ -71,21 +71,33 @@ required hardware or credentials are unavailable.
 * A benchmark number without the hardware and the command that produced it.
 * A comparison table cell asserting that another project lacks a feature,
   without a link to the code or docs that show it.
-* Marketing language. No "blazing", "seamless", "state of the art", or claims
-  that miniVERL is first, fastest or best at anything.
+* Unsubstantiated superlatives such as "fastest" or "state of the art".
+  Describe a useful capability or cite a measured comparison instead.
 * Silent behaviour: truncating a sequence, changing a model, reusing a stale
   cache, or swallowing an exception. Fail loudly with a hint instead.
 
 ## Documentation voice
 
 Lead with what the workflow does, who it helps and which artifact it produces.
+The short product description is **"verl for a single consumer GPU"**
+(**"单张消费级 GPU 上的 verl"**). Follow it with concrete actions: train with
+PPO/GRPO, distill a teacher, resume a run, export a model. Introduce compiler,
+semantic-lowering and provenance terminology when explaining how these work,
+after the reader has a reason to care.
+
 State a local constraint where it changes the reader's next action; link to
 the [compatibility policy](docs/compatibility.md) or
 [limitations](docs/limitations.md) for complete exclusion lists and scientific
 boundaries. Avoid repeating a catalogue of absent features after every product
 capability. Research reports and security contracts should keep their precise
 caveats close to the evidence or decision they qualify.
 
+Prefer "This workload measures timing, memory and recovery" to a list of all
+the things it does not measure. Keep a clearly labelled scope section in the
+report. Preserve negative results, failed gates, security warnings and exact
+compatibility statuses; a voice edit changes the presentation, not the evidence.
+Mark old release assessments as historical and link to the current support table.
+
 ## Adding an environment
 
 See the "New environment" issue template for the checklist. In short: seeded and