chore: refactor and cleanup

.claude/CLAUDE.md

@@ -9,29 +9,30 @@ Template repository for analysing single-cell data.
 ## Project structure
 
 - `data/`
-    - Directory containing data relevant to project
-    - Contains one subdirectory for each dataset
-    - Each dataset-specific subdirectory uses the following structure
-        - `raw/`: original, unaltered data that always exists unless the data has been loaded from an external data directory
-        - `processed/`: processed data from `raw/`
-        - `results`: analysis results
+  - Directory containing data relevant to project
+  - Contains one subdirectory for each dataset
+  - Each dataset-specific subdirectory uses the following structure
+    - `raw/`: original, unaltered data that always exists unless the data has been loaded from an external data directory
+    - `processed/`: processed data from `raw/`
+    - `results`: analysis results
 - `figures/`
-    - Directory to collect generated figures
-    - Contains one subdirectory for each dataset
+  - Directory to collect generated figures
+  - Contains one subdirectory for each dataset
 - `jobs/`
-    - Directory to collect scripts for submitting HPC jobs, e.g., with sbatch, (in `jobs/scripts/`) and their generated output files (in `jobs/logs/`)
+  - Directory to collect scripts for submitting HPC jobs, e.g., with sbatch, (in `jobs/scripts/`) and their generated output files (in `jobs/logs/`)
 - `notebooks/`
-    - Directory containing Jupyter notebooks
-    - Contains one subdirectory for each dataset
+  - Directory containing Jupyter notebooks
+  - Contains one subdirectory for each dataset
 - `scripts/`
-    - Directory containing Python or R scripts
-    - Contains one subdirectory for each dataset
+  - Directory containing Python or R scripts
+  - Contains one subdirectory for each dataset
 
 **Note:**
+
 - The dataset-specific subdirectories in `data/`, `figures/`, `notebooks/`, `scripts` all have the same name
 - Before committing a notebook with git, it is converted into a corresponding script in `scripts/` via pre-commit hooks (jupytext); jupytext also keeps them in sync in subsequent commits
-    - Scripts use `# %%` cell delimiters (`py:percent` format).
-    - See `notebooks/template.ipynb` ↔ `scripts/template.py` as an example
+  - Scripts use `# %%` cell delimiters (`py:percent` format).
+  - See `notebooks/template.ipynb` ↔ `scripts/template.py` as an example
 
 ## Setup
 

.editorconfig

@@ -11,7 +11,7 @@ charset = utf-8
 trim_trailing_whitespace = true
 insert_final_newline = true
 
-[*.{yml,yaml}]
+[*.{yml,yaml,md}]
 indent_style = space
 indent_size = 2
 end_of_line = lf

{{ cookiecutter.project_name }}/.claude/CLAUDE.md

@@ -3,75 +3,87 @@
 This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
 
 ## Rule 0
+
 When anything fails: STOP. Explain to Q (the user). Wait for confirmation before proceeding.
 
 ## Before consequential actions
+
 For actions that are hard to reverse or change shared state (not trivial reads):
+
 ```
 ASSUMPTION: [assumption that warrant action]
 DOING: [action]
 EXPECT: [predicted outcome]
 IF WRONG: [what that means]
 ```
+
 Then the tool call. Then compare. Mismatch = stop and surface to Q.
 
 ## Checkpoints
+
 Regardless of the above, verify reality matches your model at least every 3 actions. Thinking is NOT verification-observable output is.
 
 ## Epistemic hygiene
+
 - "I believe X" ≠ "I verified X"
 - "I don't know" beats confident guessing
 - One example is anecdote, three is MAYBE a pattern
 - Data and world model disagree ⇒ challenge, critique and adapt your world model. Do NOT soften your rationale to explain away disagreeing data — that's backwards.
 - Do NOT simply agree with Q → critically assess Q's assumptions and challenge them if needed
 
 ## Autonomy check
+
 Before significant decisions: Am I the right entity to decide this?
 Uncertain + consequential → ask Q first. Cheap to ask, expensive to guess wrong.
 
 ## Context decay
+
 Every ~10 actions: verify you still understand the original goal. Say "losing the thread" when degraded.
 
 ## Chesterton's fence
+
 Can't explain why something exists? Don't touch it until you can.
 
 ## Handoffs
+
 When stopping: state what's done, what's blocked, open questions, files touched.
 
 ## Communication
+
 When confused: stop, think, present theories, get signoff. Never silently retry failures.
 
 ## Project overview
 
-{{ cookiecutter.project_name }}: {{ cookiecutter.package_description }}
+fibrysis: Analysis of cancer cell-fibroblast interactions within the CNS
 
 ## Project structure
 
 - `data/`
-    - Directory containing data relevant to project
-    - Contains one subdirectory for each dataset
-    - Each dataset-specific subdirectory uses the following structure
-        - `raw/`: original, unaltered data that always exists unless the data has been loaded from an external data directory
-        - `processed/`: processed data from `raw/`
-        - `results`: analysis results
+  - Directory containing data relevant to project
+  - Contains one subdirectory for each dataset
+  - Each dataset-specific subdirectory uses the following structure
+    - `raw/`: original, unaltered data that always exists unless the data has been loaded from an external data directory
+    - `processed/`: processed data from `raw/`
+    - `results`: analysis results
 - `figures/`
-    - Directory to collect generated figures
-    - Contains one subdirectory for each dataset
+  - Directory to collect generated figures
+  - Contains one subdirectory for each dataset
 - `jobs/`
-    - Directory to collect scripts for submitting HPC jobs, e.g., with sbatch, (in `jobs/scripts/`) and their generated output files (in `jobs/logs/`)
+  - Directory to collect scripts for submitting HPC jobs, e.g., with sbatch, (in `jobs/scripts/`) and their generated output files (in `jobs/logs/`)
 - `notebooks/`
-    - Directory containing Jupyter notebooks
-    - Contains one subdirectory for each dataset
+  - Directory containing Jupyter notebooks
+  - Contains one subdirectory for each dataset
 - `scripts/`
-    - Directory containing Python or R scripts
-    - Contains one subdirectory for each dataset
+  - Directory containing Python or R scripts
+  - Contains one subdirectory for each dataset
 
 **Note:**
+
 - The dataset-specific subdirectories in `data/`, `figures/`, `notebooks/`, `scripts` all have the same name
 - Before committing a notebook with git, it is converted into a corresponding script in `scripts/` via pre-commit hooks (jupytext); jupytext also keeps them in sync in subsequent commits
-    - Scripts use `# %%` cell delimiters (`py:percent` format).
-    - See `notebooks/template.ipynb` ↔ `scripts/template.py` as an example
-    - See `README.md`'s "Workflow" section for the full hook-modifies/re-stage/re-commit dance
+  - Scripts use `# %%` cell delimiters (`py:percent` format).
+  - See `notebooks/template.ipynb` ↔ `scripts/template.py` as an example
+  - See `README.md`'s "Workflow" section for the full hook-modifies/re-stage/re-commit dance
 - `notebooks/index.md` is generated from `README.md` by a pre-commit hook (3-way merge via `git merge-file`) — edit `README.md`, not `index.md`
 
 ## Setup
@@ -107,4 +119,4 @@ There are no tests. Development is notebook-driven.
 ## Data conventions
 
 - Store datasets as Zarr zip archives unless told otherwise; use the `write_zarr` helper function from `{{ cookiecutter.package_name }}.io`
-- Read Zarr zip archives with the `read_zarr` helper function from `{{ cookiecutter.package_name }}.io`
+- Read Zarr zip archives with the `read_as_dask` helper function from `{{ cookiecutter.package_name }}.io` and only load data into memory if needed

{{ cookiecutter.project_name }}/.editorconfig

@@ -11,7 +11,7 @@ charset = utf-8
 trim_trailing_whitespace = true
 insert_final_newline = true
 
-[*.{yml,yaml}]
+[*.{yml,yaml,md}]
 indent_style = space
 indent_size = 2
 end_of_line = lf

{{ cookiecutter.project_name }}/pyproject.toml

@@ -65,6 +65,7 @@ envs.hatch-test.overrides.matrix.deps.env-vars = [
 [tool.pixi]
 workspace.channels = [ "conda-forge" ]
 workspace.platforms = [ "linux-64", "osx-64", "osx-arm64", "win-64" ]
+pypi-dependencies."{{ cookiecutter.package_description }}" = { path = ".", editable = true }
 tasks.run-pre-commit = "pre-commit run --all-files"
 tasks.setup-pre-commit = "pre-commit install"
 environments.default = { features = [ "dev", "jupyter" ], solve-group = "default" }

{{ cookiecutter.project_name }}/src/{{ cookiecutter.package_name }}/io/_dask.py

@@ -81,6 +81,7 @@ def read_as_dask(
     """Read AnnData with dask.
 
     ```python
+    store = zarr.storage.ZipStore(path / to / store.zarr.zip)
     adata = read_as_dask(file_path)
 
     obs_mask = adata.obs[column] == group