Curated example reflectometry datasets — together with the plans and models generated from them — used to develop and test reflectometry analysis tools such as nr-analyzer and AuRE.

Each example is a self-contained folder holding a reduced dataset, the scientist's description of the sample, and the artifacts a modelling pipeline produces from them. Having real datasets paired with their expected outputs lets us exercise the tools end to end and catch regressions against known-good results.

Examples live under refl1d_models/. Every example follows the same convention:

refl1d_models/<example-name>/
├── context.md            # the scientist's free-form sample description
├── <data files>          # reduced reflectivity data (see "Examples" below)
├── plan/
│   └── job_<id>.yaml      # plan-data output: a create-model-ready config
├── models/
│   └── <name>.py          # create-model output: a refl1d-ready model script
└── reference/             # (optional) hand-written expert model, for comparison
    └── <name>.py

The pipeline each example captures:

context.md is the input a human provides; plan/ and models/ are outputs of the tools. A reference/ model, where present, is a hand-written model from a domain expert — useful as a ground truth to compare a generated model against.

A copper film (CuOₓ / ~50 nm Cu / ~3 nm Ti on Si, measured in D₂O with neutrons entering from the silicon side). The dataset is a set of partial files for one measurement sequence that are co-refined as a single state, plus the combined auto-reduced file:

• REFL_226642_{1,2,3}_*_partial.txt — the three angle segments (co-refined)
• REFL_226642_combined_data_auto.txt — the auto-stitched combined dataset
• plan/job_226642.yaml, models/sample5_ocv_226642.py

A 1200 Å Ni[58] magnetic layer on Si, measured with polarized neutrons (PNR). The data are header-less three-column (Q R dR) spin-state files:

• Ni58_stitched_u.dat / Ni58_stitched_d.dat — spin-up / spin-down
• plan/job_Ni58_stitched.yaml, models/Ni58_basic_model.py
• reference/Ni58_basic_model.py — the hand-written expert model this example was originally built from

The plan-data, create-model, and run-fit commands come from nr-analyzer:

git clone https://github.com/neutrons-ai/nr-analyzer.git
cd nr-analyzer
python3 -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -e ".[dev]"

Generating a plan and model from a context.md (plan-data, and create-model's LLM mode) additionally needs AuRE installed in the same environment and a configured LLM endpoint — see the nr-analyzer installation notes. Running a fit on an existing model script (run-fit) does not need an LLM.

Clone the repo and point a tool at an example folder. Each plan references its data files by bare name, and the analyzer locates them through the ANALYZER_* data-directory environment variables. Since the data sits one level up from plan/, point those at the example folder before running create-model:

export ANALYZER_PARTIAL_DATA_DIR="$PWD/refl1d_models/LR-cu-film-corefined"
export ANALYZER_COMBINED_DATA_DIR="$PWD/refl1d_models/LR-cu-film-corefined"
create-model --config refl1d_models/LR-cu-film-corefined/plan/job_226642.yaml

A repo- or sample-level .env can set these instead of exporting them by hand; see the nr-analyzer configuration docs for the full ANALYZER_* set and the .env cascade.

The generated models/*.py scripts reference the data by absolute path. If you cloned the repo somewhere other than ~/git/refl1d_models, update that path before running a fit.

1. Create refl1d_models/<example-name>/ and drop in the reduced data files.
2. Write a context.md describing the sample (substrate, layers, ambient, measurement conditions, and any modelling hypotheses).
3. Generate the plan and model with plan-data / create-model (outputs land in plan/ and models/); optionally add a hand-written reference/ model.

MIT — see LICENSE.