diff --git a/.github/PULL_REQUEST_TEMPLATE.md b/.github/PULL_REQUEST_TEMPLATE.md
index a5edd501..ddb1876d 100644
--- a/.github/PULL_REQUEST_TEMPLATE.md
+++ b/.github/PULL_REQUEST_TEMPLATE.md
@@ -1,3 +1,8 @@
+New public widgets and shared-helper refactors need an issue or maintainer
+discussion first. Bug fixes and features that stay inside one existing widget
+can open a pull request directly. See
+[pull request classes](https://github.com/electronmicroscopy/quantem.widget/blob/main/docs/maintainer/pull-requests.md).
+
 ### What problem does it solve?
 
 <!--
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 30c59833..af09ba71 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,12 @@ new `rcN` heading when that rc is published to TestPyPI.
 
 ## Unreleased
 
+- Maintainer docs split pull requests into discuss-first (new widgets,
+  cross-widget refactors) and incremental in-widget fixes, and add a
+  `widget-tutorials/` upload page for the public
+  [bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data)
+  dataset. A copyable agent prompt opens that Hugging Face Community
+  pull request after `python -m check_meta` prints `ok`.
 - Installation, Colab tutorial cells, and the verify snippet now install the
   TestPyPI wheel and import `load` from `quantem.gpu.io`. `quantem
   showdiffraction --demo` is documented as needing the public Fe3O4 hub folder,
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index f9bb6491..1fcdb3e1 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -16,6 +16,19 @@ Use one small branch for one issue or documentation goal. Describe the user
 problem before the implementation, and keep unrelated experiments or cleanup
 out of the branch.
 
+Adding a new widget, or a refactor that other widgets will import, starts with
+an issue or a maintainer discussion. A bug fix or a feature that stays inside
+one existing widget can go straight to a pull request. The two classes are in
+the [pull request guide](docs/maintainer/pull-requests.md).
+
+Tutorial fixtures for notebooks belong under `widget-tutorials/` in the public
+[bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data)
+dataset. Upload and download commands are on that dataset card. A coding
+agent can open the Hugging Face Community pull request with the prompt in
+[Agent-assisted development](docs/agent-prompts.md#add-tutorial-data-hugging-face-pull-request).
+The GitHub loader pull request is in
+[Contribute tutorial data](docs/tutorials/contribute_data.md).
+
 Once review begins, do not force-push. Add follow-up commits so GitHub preserves
 inline comments and the reviewer can inspect only the changes since their last
 visit. The project squash-merges completed work.
diff --git a/docs/_toc.yml b/docs/_toc.yml
index 5892aca7..2c051c44 100644
--- a/docs/_toc.yml
+++ b/docs/_toc.yml
@@ -14,6 +14,8 @@ parts:
   - caption: Tutorials
     chapters:
       - file: tutorials/download_data
+      - file: tutorials/contribute_data
+        title: Contribute tutorial data
       - file: tutorials/show4dstem
         sections:
           - file: tutorials/show4dstem_single
@@ -88,6 +90,8 @@ parts:
       - file: maintainer/index
         title: Maintainer guide
         sections:
+          - file: maintainer/pull-requests
+            title: Pull requests
           - file: maintainer/docs-design
             title: Docs design rules
           - file: changelog
diff --git a/docs/agent-prompts.md b/docs/agent-prompts.md
index 11e860b0..7215dbc7 100644
--- a/docs/agent-prompts.md
+++ b/docs/agent-prompts.md
@@ -95,12 +95,15 @@ Prepare a focused quantem.widget pull request for:
 
 <issue URL or concise goal>
 
-First read AGENTS.md, CONTRIBUTING.md, the pull-request template, and any
-instructions nearest the changed files. Inspect the worktree, branches, and
-remotes; preserve unrelated user changes. Fetch the authoritative upstream
-main branch before comparing or creating the feature branch. Follow the
-repository's branch-naming rules; if none are present, use a short creation-date
-prefix such as aug-8-show2d-contrast.
+First read CONTRIBUTING.md, docs/maintainer/pull-requests.md, the
+pull-request template, and any instructions nearest the changed files.
+New public widgets and cross-widget refactors need an issue or maintainer
+discussion first. In-widget bug fixes and features do not.
+
+Inspect the worktree, branches, and remotes; preserve unrelated user changes.
+Fetch the authoritative upstream main branch before you compare or create the
+feature branch. Follow the repository's branch-naming rules; if none are
+present, use a short creation-date prefix such as aug-8-show2d-contrast.
 
 Keep one scientific or documentation goal in the PR. Stage named files only.
 Exclude private data, machine-specific paths, notebook outputs, generated HTML,
@@ -122,3 +125,57 @@ Preserve my configured Git identity and do not add a Co-authored-by trailer.
 
 When publication is approved, push to the contributor's fork and let the human
 open the upstream PR unless they explicitly ask the agent to do that step.
+
+## Add tutorial data (Hugging Face pull request)
+
+Use this when a scientist has files that should become a public
+`widget-tutorials/` fixture. The pull request is on Hugging Face, not
+GitHub. Do not paste the Hugging Face token into this prompt.
+
+```text
+Add these files to the public Hugging Face dataset bobleesj/quantem-data
+as a Community pull request.
+
+Local folder:
+<absolute path to a folder that already contains the data file(s) and will
+get meta.json>
+
+Target hub path:
+widget-tutorials/<widget-or-shared>/<name>/<size>/
+
+Permission (required; do not invent):
+- I collected this data myself: <yes/no>
+- permission.from: <name or list>
+- permission.collectors: <name or list>
+- permission.date: <YYYY-MM-DD>
+- I have written permission from the collector: <yes, or "I collected it">
+
+Read https://huggingface.co/datasets/bobleesj/quantem-data (dataset card).
+Do not use quantem.widget.io.upload. Do not commit .npy or .emd into
+electronmicroscopy/quantem.widget.
+
+1. Confirm hf auth whoami prints a username. If it fails, stop and ask me
+   to create a Write token at huggingface.co/settings/tokens and either
+   run `hf auth login` (paste token, answer n to git credential) or
+   `export HF_TOKEN=...`. Do not run interactive hf auth login yourself.
+   Do not write the token into a file in any git repo.
+2. Copy templates/meta.json. Fill name, source, date, sample, shape,
+   dtype, and permission from the files and from the permission lines
+   above. Read shape and dtype from the array or EMD. Omit sampling,
+   units, and optics if unknown. Do not invent a number.
+3. Download check_meta/ from bobleesj/quantem-data (add --revision
+   refs/pr/N if that folder is not on main yet). Run
+   `python -m check_meta <folder>`. Stop if it is not ok.
+4. Run:
+   hf upload bobleesj/quantem-data <folder> \
+     widget-tutorials/<widget-or-shared>/<name>/<size> \
+     --repo-type dataset --create-pr \
+     --commit-message "add <widget> <name> <size> tutorial data"
+5. Give me the Community URL. Do not merge. Do not open a GitHub pull
+   request unless I also need a named loader in
+   quantem.widget.datasets (that GitHub PR waits until this HF PR
+   merges). A new widget is discuss-first.
+```
+
+See [Contribute tutorial data](tutorials/contribute_data.md) and
+[Pull requests](maintainer/pull-requests.md).
diff --git a/docs/api/datasets.md b/docs/api/datasets.md
index 90b82ed5..363645af 100644
--- a/docs/api/datasets.md
+++ b/docs/api/datasets.md
@@ -51,9 +51,13 @@ widget = Show1D.from_example("ducky", size="small")
 widget
 ```
 
-The public Hugging Face dataset is organized under `widget-tutorials/`. Reused
-sources live once under `shared/`; widget-specific monitor runs or session
-folders live under the widget name.
+The public Hugging Face dataset
+([bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data))
+is organized under `widget-tutorials/`. Reused sources live once under
+`shared/`; widget-specific monitor runs or session folders live under the
+widget name. Upload and download commands are on that dataset card. The
+GitHub loader pull request is in
+[Contribute tutorial data](../tutorials/contribute_data.md).
 
 ```text
 widget-tutorials/{widget-or-shared}/{example}/{size}/...
@@ -66,10 +70,15 @@ widget-tutorials/show1d/ducky/small/show1d_monitor.jsonl
 widget-tutorials/show1d/ducky/small/snapshots/*.npy
 widget-tutorials/shared/gold-haadf/full/data.npy
 widget-tutorials/show4dstem/gold-128-bin8/full/data.npy
-widget-tutorials/showdiffraction/fe3o4-saed/small/data.npy
+widget-tutorials/show4dstem/gold-512-bin4/full/data.npy
+widget-tutorials/show4dstem/gold-512-bin8/full/data.npy
 widget-tutorials/showfolder/gold-haadf-session/small/*.emd
 ```
 
+`showdiffraction_fe3o4()` expects
+`widget-tutorials/showdiffraction/fe3o4-saed/small/data.npy`. That folder is
+not in the current hub snapshot.
+
 This keeps widget tutorial payloads grouped together instead of placing many
 example files at the top level of the shared dataset repository. `show2d_gold`
 and `show3d_gold` intentionally share `widget-tutorials/shared/gold-haadf/full`
diff --git a/docs/api/io.md b/docs/api/io.md
index c82d056e..c3631b47 100644
--- a/docs/api/io.md
+++ b/docs/api/io.md
@@ -570,9 +570,10 @@ owns it. Shut down that kernel from JupyterLab or stop the Python process.
 
 The shared Hugging Face dataset repo
 ([bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data),
-MIT license) is the one place tutorial and reference data lives — its dataset
-card intentionally holds no instructions and points back to this page. The
-upload protocol is three steps:
+MIT license) is the one place tutorial and reference data lives. Upload and
+download commands, including Hugging Face pull requests, are on that dataset
+card. This section is the `quantem.widget.io` helper reference for
+maintainers who already have write access. The upload steps are:
 
 1. **Install the hub extra and log in once.** Uploading needs a Hugging Face
    account and a write token from
@@ -580,23 +581,29 @@ upload protocol is three steps:
 
    ```bash
    pip install "quantem.widget[hub]"
-   hf auth login   # paste the write token
+   hf auth login   # paste a Write token; answer n to git credential
    ```
 
+   Token steps for a Community pull request (no write access) are on the
+   dataset card. A Read token cannot open a PR. Ignore any hint to run
+   `git config --global credential.helper store`.
+
 2. **Upload with the bucket + sidecar convention.** The repo has two trees:
 
    - `widget-tutorials/<widget>/<dataset>/<size>/` — the baseline tutorial
      bundles behind `quantem.widget.datasets` (sizes `small`/`medium`/
-     `large`/`full`). To contribute a tutorial bundle, pass
-     `folder="widget-tutorials/<widget>"` and `name="<dataset>/full"`.
+     `large`/`full`). Contributors add these with a Hugging Face pull
+     request, not `quantem.widget.io.upload`. See
+     [Contribute tutorial data](../tutorials/contribute_data.md).
      Datasets shared by several widgets (the gold HAADF feeds both Show2D
      and Show3D) live under `widget-tutorials/shared/`.
    - `4dstem/` and `haadf/` — full-size originals for power users. A folder
      of Arina `*_master.h5` files goes under `4dstem/`, a single image file
      under `haadf/` (those are also the defaults for a directory vs a file).
 
-   Always pass `meta=` so downstream widgets get calibration — it is written
-   as a `meta.json` sidecar next to your data:
+   For `4dstem/` and `haadf/`, pass `meta=` so the helper writes a
+   `quantem_meta.json` sidecar. Tutorial loaders read `meta.json` from the
+   staged `widget-tutorials/` folder instead:
 
    ```python
    from quantem.widget.io import upload
@@ -610,10 +617,10 @@ upload protocol is three steps:
    )
    ```
 
-   No write access to the shared repo? Either open an issue on
-   [quantem.widget](https://github.com/electronmicroscopy/quantem.widget/issues) to get
-   added, or pass `repo="you/your-data"` to use your own HF dataset repo with
-   the same layout — every download helper accepts the same `repo=` override.
+   Write access to `bobleesj/quantem-data` is for maintainers. Contributors
+   open a Hugging Face pull request on the dataset card, then a GitHub pull
+   request for the named loader
+   ([Contribute tutorial data](../tutorials/contribute_data.md)).
 
 3. **Verify like a user would.** List, download to a fresh path, and open it
    in the widget before announcing the dataset:
diff --git a/docs/developer/widget-creation.md b/docs/developer/widget-creation.md
index bd9a857a..d73869ca 100644
--- a/docs/developer/widget-creation.md
+++ b/docs/developer/widget-creation.md
@@ -1,6 +1,6 @@
 # Creating a widget
 
-This guide shows how to add a new interactive viewer or tool to `quantem.widget`. Instructions for agentic-driven development are on the [Agent-assisted development](../agent-prompts) page, and the workflow for setup, test, and PRs is in [CONTRIBUTING.md](https://github.com/electronmicroscopy/quantem.widget/blob/main/CONTRIBUTING.md).
+This guide shows how to add a new interactive viewer or tool to `quantem.widget`. A new widget needs an issue or maintainer discussion first ([Pull requests](../maintainer/pull-requests.md)). Instructions for agentic-driven development are on the [Agent-assisted development](../agent-prompts) page, and the workflow for setup, test, and incremental PRs is in [CONTRIBUTING.md](https://github.com/electronmicroscopy/quantem.widget/blob/main/CONTRIBUTING.md).
 
 Let's use `Ruler2D` as an example, a widget where the user places two endpoints on a 2D image and reads back a calibrated distance.
 
@@ -31,6 +31,11 @@ Always edit the file in `quantem.gpu`, never the generated copy.
 
 ## Before you start
 
+A new public widget is a discuss-first change. Open an issue or confirm with a
+maintainer that the widget belongs in this package before choosing a base class
+or writing React code. Incremental fixes and features that stay inside one
+existing widget skip this step. See [Pull requests](../maintainer/pull-requests.md).
+
 Before choosing a base class or writing React code, define what the widget is supposed to do.
 
 For `Ruler2D`, the example would be that it accepts one 2D image, stores coordinates as `(row, col)` in the original image, allows two endpoints to be placed or dragged, and reports their calibrated distance.
diff --git a/docs/install.md b/docs/install.md
index 92640248..37d6f0de 100644
--- a/docs/install.md
+++ b/docs/install.md
@@ -23,6 +23,10 @@ pip install -i https://test.pypi.org/simple/ \
     "quantem.widget[hub]"
 ```
 
+Widget tutorial fixtures live under `widget-tutorials/` on
+[bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data).
+Upload and download commands are on that dataset card.
+
 ## Google Colab
 
 Each tutorial notebook can open directly in Colab from the badge at the top of
diff --git a/docs/intro.md b/docs/intro.md
index 453ef440..9c75a1f7 100644
--- a/docs/intro.md
+++ b/docs/intro.md
@@ -130,5 +130,12 @@ https://doi.org/10.1093/mam/ozag053.941.
 
 - **Questions or bugs:** open an issue at
   [github.com/electronmicroscopy/quantem.widget/issues](https://github.com/electronmicroscopy/quantem.widget/issues).
-- **Maintained by** the Ophus group. Contributions and feedback are welcome via
-  pull request or issue.
+- **New widgets and cross-widget refactors:** discuss first (issue or
+  maintainer). In-widget bug fixes can open a pull request directly. See
+  [Pull requests](maintainer/pull-requests.md) and
+  [CONTRIBUTING.md](https://github.com/electronmicroscopy/quantem.widget/blob/main/CONTRIBUTING.md).
+- **Tutorial data:** public
+  [bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data).
+  Upload and download commands are on that dataset card. The GitHub loader
+  pull request is in [Contribute tutorial data](tutorials/contribute_data.md).
+- **Maintained by** the Ophus group.
diff --git a/docs/maintainer/index.md b/docs/maintainer/index.md
index 8e3d5a5c..080fa0a2 100644
--- a/docs/maintainer/index.md
+++ b/docs/maintainer/index.md
@@ -1,8 +1,8 @@
 # Maintainer guide
 
-Release process, automation, signoff protocols, UI standards, storyboards, and
-performance notes for quantem.widget maintainers. Day-to-day users do not need
-this section; start with the tutorials instead.
+Release process, pull-request classes, automation, signoff protocols, UI
+standards, storyboards, and performance notes for quantem.widget maintainers.
+Day-to-day users do not need this section; start with the tutorials instead.
 
 ```{tableofcontents}
 ```
diff --git a/docs/maintainer/pull-requests.md b/docs/maintainer/pull-requests.md
new file mode 100644
index 00000000..a4eda0ac
--- /dev/null
+++ b/docs/maintainer/pull-requests.md
@@ -0,0 +1,53 @@
+# Pull requests
+
+This package has two pull-request classes. The first needs a discussion before
+code. The second can open a pull request directly.
+
+Delivery style (one theme per PR, self-review as a draft, quote replies, keep
+follow-up ideas in a new issue) follows the
+[Billinge group pull-request practices](https://scikit-package.github.io/scikit-package/programming-guides/billinge-group-standards.html#pull-request-practices).
+The class rule below is the extra gate for widgets: not every change needs an
+issue first.
+
+## Discuss first
+
+Open an issue, or confirm with a maintainer, before writing the pull request
+when the change:
+
+- adds a new public widget
+- extracts shared helpers or refactors code that other widgets import
+- changes a public API, the comm protocol, or the tutorial-data contract
+
+Those changes have side effects on other viewers, docs, release registration,
+and review load. The discussion decides whether the widget or abstraction
+belongs in this package, and what the smallest first pull request is.
+
+A new widget then follows [Creating a widget](../developer/widget-creation.md).
+
+## Incremental (no prior discussion)
+
+Open a pull request when the change stays inside one existing widget:
+
+- a bug fix
+- a control or option used only by that widget
+- docs or tests for that widget
+
+Keep the branch to one problem. Use the
+[pull-request template](https://github.com/electronmicroscopy/quantem.widget/blob/main/.github/PULL_REQUEST_TEMPLATE.md).
+
+## Shared delivery rules
+
+- One scientific or documentation goal per pull request.
+- Branch names use a short date prefix such as `aug-19-show2d-contrast`.
+- Once review starts, add follow-up commits. Do not force-push. The project
+  squash-merges completed work.
+- Put reviewer actions and verification evidence in the three visible template
+  sections. Keep the hidden checklist out of the rendered description.
+
+Tutorial fixtures for notebooks live in the public Hugging Face dataset
+[bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data)
+under `widget-tutorials/`. Upload and download commands are on that dataset
+card. A coding agent opens that Community pull request with
+[Add tutorial data](../agent-prompts.md#add-tutorial-data-hugging-face-pull-request).
+The GitHub loader pull request is in
+[Contribute tutorial data](../tutorials/contribute_data.md).
diff --git a/docs/tutorials/contribute_data.md b/docs/tutorials/contribute_data.md
new file mode 100644
index 00000000..5c65f4b7
--- /dev/null
+++ b/docs/tutorials/contribute_data.md
@@ -0,0 +1,20 @@
+# Contribute tutorial data
+
+How to add files, including the Hugging Face token
+(`hf auth login`):
+[bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data).
+A coding agent can open that Community pull request. Copy the prompt
+[Add tutorial data](../agent-prompts.md#add-tutorial-data-hugging-face-pull-request).
+The human creates the Write token and attests permission. The agent runs
+`hf upload ... --create-pr` after `python -m check_meta` prints `ok`.
+
+After that Hugging Face PR is merged, open a GitHub PR here only if a
+notebook should load the fixture by name:
+
+- helper in `src/quantem/widget/data/tutorials.py`
+- list it on [Tutorial Datasets](../api/datasets.md)
+- use it in the tutorial
+- link the Hugging Face PR
+
+A new widget is discuss-first. See [Pull requests](../maintainer/pull-requests.md).
+Do not commit `.npy` or `.emd` here. Do not call `quantem.widget.io.upload`.
diff --git a/docs/tutorials/download_data.ipynb b/docs/tutorials/download_data.ipynb
index d785f726..153ce8d4 100644
--- a/docs/tutorials/download_data.ipynb
+++ b/docs/tutorials/download_data.ipynb
@@ -206,7 +206,7 @@
     ")\n",
     "```\n",
     "\n",
-    "These helpers use the shared size language `small`, `medium`, `large`, and `full`. Tutorial payloads are grouped under `widget-tutorials/` in the public Hugging Face dataset. Reused sources are stored once, for example Show2D and Show3D both derive from `widget-tutorials/shared/gold-haadf/full` instead of keeping separate copies.\n"
+    "These helpers use the shared size language `small`, `medium`, `large`, and `full`. Tutorial payloads are grouped under `widget-tutorials/` in the public Hugging Face dataset. Reused sources are stored once, for example Show2D and Show3D both derive from `widget-tutorials/shared/gold-haadf/full` instead of keeping separate copies. To add a fixture, follow the [quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data) dataset card, then the GitHub loader steps in [Contribute tutorial data](contribute_data).\n"
    ],
    "id": "widget-datasets-11"
   },
@@ -216,22 +216,10 @@
    "source": [
     "## Upload your own data\n",
     "\n",
-    "The reference datasets above all live in one shared Hugging Face repo,\n",
+    "The examples above live in the public Hugging Face repo\n",
     "[bobleesj/quantem-data](https://huggingface.co/datasets/bobleesj/quantem-data)\n",
-    "(MIT license), and anyone can contribute with the same three-line protocol:\n",
-    "\n",
-    "```python\n",
-    "from quantem.widget.io import upload\n",
-    "\n",
-    "upload(\"/data/session/gold_512/\", name=\"gold_512\", folder=\"4dstem\",\n",
-    "       meta={\"sampling\": [0.5, 0.5], \"units\": [\"A\", \"A\"], \"voltage_kV\": 300})\n",
-    "```\n",
-    "\n",
-    "Folders of `*_master.h5` go under `4dstem/`, single images under `haadf/`,\n",
-    "and the `meta=` dict becomes a `meta.json` sidecar so widgets pick up the\n",
-    "calibration automatically. The full protocol — tokens, your-own-repo\n",
-    "uploads via `repo=`, verifying, and deleting — is in\n",
-    "[Load and I/O: how do I upload data?](../api/io.md#i-have-data-others-should-use-how-do-i-upload-it).\n"
+    "(MIT license). Upload and download commands are on that dataset card.\n",
+    "The GitHub loader pull request is in [Contribute tutorial data](contribute_data).\n"
    ]
   }
  ],