# Reproducible point-pattern teaching constructions

These files contain **synthetic data**, not observed FIRMS fire detections, actual bus-stop coordinates or an empirical regional point-event sample. They support the quadrat, NNA and K-function readings.

## Files and coordinate frame

- `synthetic_points.csv`: 96 points in each of three patterns, with `pattern`, `x_m` and `y_m` fields. Coordinates use a fictional planar metric frame, not a geographic CRS. Do not put them on a real basemap.
- `results.json`: original seed 5230, simulation seed 5231, 999 simulations, the fixed window, radius grid and computed results.
- `../../../../scripts/point_pattern_examples.py` in the repository root is the authoritative generator. Run `python scripts/point_pattern_examples.py --write` from the repository root. Its default invocation checks the calculations without regenerating images.

The observation window is the full square `[0, 2000] x [0, 2000]` metres. It is defined independently of each point pattern and remains fixed. The CSR sample contains independent uniform points. The clustered construction places 24 normally displaced points around each of four centres. The spaced construction jitters a 12 by 8 lattice. These constructions illustrate contrasting summaries; their labels do not imply that every statistic must reject a particular null.

Quadrat figures use a 4 by 4 partition with expected count six in each cell. The sensitivity figure also uses a 2 by 2 partition and a shifted grid clipped to the same window. Expected counts for clipped cells use their actual areas. All points, including those on the outer edge, enter exactly one cell.

NNA calculations use Euclidean distances to the closest other point. The expected distance and ratio use the unbounded planar Clark–Evans approximation; they are not boundary-corrected finite-window inference.

The K figures use the explicitly implemented rectangular translation correction and ordered-pair normalisation. The plotted summary is `sqrt(K/pi) - r`, with radii 25–500 m. The 2.5th and 97.5th simulated quantiles at each radius are **pointwise reference bands**, not a global envelope test or confidence intervals for a true process curve. The same 999 fixed-count uniform samples supply the quadrat reference and K bands.

The separate bus-stop-inspired construction uses 16 regularly spaced sites, each with two points 30 m apart. It is illustrative and does not appear in the empirical-record CSV. The existing bus-stop image on the page is retained only as motivation; no new empirical result is attributed to it.

## Observed Sumatra FIRMS case

The quadrat page now includes two adjoining South Sumatra windows, separately and combined. These use observed September 2026 VIIRS S-NPP detection records. See the [separate empirical data record](../firms_sumatra/README.md) for source provenance, filters, limits and reproduction. The synthetic datasets in this folder remain teaching constructions and are not substituted for observed fires.

## Verification and attribution

The generator checks boundary-point allocation, count conservation, a hand-worked chi-squared statistic, nearest-neighbour distance, the ordered-pair translation correction, monotonic K curves and agreement with the committed result record. It also rejects points outside the declared window.

Original code follows the repository's MIT code licence; original teaching prose and figures follow the repository's content licence. Data-source attribution and permissions must be recorded separately when the FIRMS case is added.