Add Claude docs (#5)

* Bump to 1.1.2 for new CRAN submission.

* Add Claude documentation in .claude/, which is ignored by the package build.

Author: njhenry

.Rbuildignore

@@ -4,4 +4,5 @@
 ^docs$
 ^pkgdown$
 ^\.github$
+^\.claude$
 Makefile

.claude/CLAUDE.md

@@ -0,0 +1,58 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Commands
+
+```bash
+# Install package locally
+make install        # runs devtools::install()
+
+# Rebuild roxygen2 documentation and NAMESPACE
+make build-docs     # runs devtools::document()
+
+# R CMD CHECK equivalent (standard R package check)
+Rscript -e "devtools::check()"
+
+# Run a single function's tests interactively
+Rscript -e "devtools::load_all(); <function_call>"
+
+# Rebuild pkgdown site
+make deploy-docs    # builds and deploys to GitHub Pages
+```
+
+There are no formal unit tests (no `tests/` directory). Validation is done through vignettes in `vignettes/`.
+
+## Architecture
+
+**mbg** is an R package for Model-Based Geostatistics — estimating continuous spatial surfaces from point-referenced observations combined with raster covariates.
+
+### Core Components
+
+**`MbgModelRunner` R6 class** ([R/MbgModelRunner.R](R/MbgModelRunner.R)) is the main orchestrator. It holds all inputs as fields and calls the modular functions below in sequence. All functions can also be called standalone.
+
+**Pipeline stages:**
+
+1. **Input loading** — `load_covariates()`, `build_id_raster()`, `build_aggregation_table()` prepare rasters and pixel-to-polygon mappings.
+
+2. **ML stacking (optional)** — `run_regression_submodels()` ([R/run_ml_models.R](R/run_ml_models.R)) fits caret models (Random Forest, GAM, etc.) to produce predictive raster surfaces used as additional covariates in the INLA model.
+
+3. **INLA preparation** — `prepare_inla_data_stack()` ([R/prepare_inla_data_stack.R](R/prepare_inla_data_stack.R)) builds the INLA data stack with fixed effects (covariates), SPDE spatial mesh (Gaussian process), optional admin-level IID random effects, and optional nugget term.
+
+4. **Model fitting** — `fit_inla_model()` ([R/fit_inla_model.R](R/fit_inla_model.R)) wraps `INLA::inla()` with PC priors and sensible defaults.
+
+5. **Prediction** — `generate_cell_draws_and_summarize()` ([R/generate_cell_draws_and_summarize.R](R/generate_cell_draws_and_summarize.R)) draws 250 posterior predictive samples per pixel, applies inverse link function, and produces mean/median/UI rasters.
+
+6. **Aggregation** — `aggregate_draws_to_polygons()` ([R/aggregate_grid_to_polygons.R](R/aggregate_grid_to_polygons.R)) propagates uncertainty through aggregation from raster cells to polygon-level summaries.
+
+### Key Design Decisions
+
+- **INLA is a Suggested dependency** (not required), checked at runtime in [R/zzz.R](R/zzz.R). The package can be installed without INLA.
+- **terra >= 1.9.1 required** for efficient aggregation table support.
+- Raster operations use **terra**, vector/polygon operations use **sf**, tabular data uses **data.table**.
+- The `id_raster` (created by `build_id_raster()`) defines the spatial template — all rasters and predictions align to it.
+- Aggregation uses fractional pixel allocation, preserving uncertainty through the `cell_draws` matrix (pixels × samples).
+
+### Example Data
+
+`inst/extdata/` contains Benin communes boundaries and child stunting observations, used in the vignettes as a worked example.

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: mbg
 Title: Model-Based Geostatistics
-Version: 1.1.1
+Version: 1.1.2
 Authors@R: c(
     person("Nathaniel", "Henry", email = "nat@henryspatialanalysis.com",
         role = c("aut", "cre"), comment = c(ORCID = "0000-0001-8150-4988")),