---
title: "Scientific design and reproducibility"
output: rmarkdown::html_vignette
vignette: >
  %\VignetteIndexEntry{Scientific design and reproducibility}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

`limpidR` separates data validation, descriptive analysis, compositional analysis,
model fitting, validation and visualization.

## Missingness

Missing environmental context remains missing. The package does not interpret blank cells
as zero and does not automatically impute climate, water-quality, land-use or population
covariates.

## Model validation

Use lake-grouped cross-validation for cross-lake generalization questions. For temporal
forecasting within a sentinel lake, use a time-blocked validation design outside the default
leave-one-lake-out workflow.

## Hotspots

`classify_hotspots()` and `map_mp_hotspots()` provide relative point classifications.
They do not claim kriging, transport modelling or hydrodynamic interpolation.

## Depth

The bundled synthetic example is event-level. `plot_depth_profile()` therefore refuses a `limpid_db`
object and requires a genuinely depth-resolved data frame.

## Risk

Risk is assumption-sensitive. `calculate_risk()` exposes contamination, polymer-hazard and
fine-particle components separately. A composite score is produced only when the user supplies
both weights and explicit component maxima.