Changes in version 0.1.0 (2026-09-30)                  

  - Add an offline HTML report with source coverage, analysis scope,
    diagnostics, separate uncertainty sections and an installed
    Markdown/Rmd template.

  - Preserve submission months across valid UTC-to-London calendar
    crossings and keep wholly missing force counts unavailable in ratio
    summaries.

  - Bound national spatial-assignment memory by reusing exact coordinate
    work without removing repeated events.

  - Establish an independent R package, source register, and milestone
    checks.

  - Bundle the public force list and Census 2021 / PACE 2023
    classification tables.

  - Add bulk archive discovery, verified caching, immutable snapshots
    and whole-file revision selection with provenance contracts.

  - Bundle the May-July 2026 sample, retaining missing Dyfed-Powys
    submissions.

  - Add versioned ONS boundaries, projected point assignment and
    boundary-distance diagnostics; preserve records without coordinates.

  - Add verified Census TS021 and RM032 exposure tables with stable
    ethnicity code matching, age/sex cross-tabs and explicit user
    exposures.

  - Add full-grid coverage, financial-year benchmarks, midnight and
    snap-point diagnostics. Date-only UTC strings cannot evade the
    summer midnight screen.

  - Add event rates with submitted population-time, exact Poisson ratio
    intervals, missing-ethnicity bounds, comparable exposure scenarios,
    age-sex direct standardisation and count-model bootstrap or
    posterior rank uncertainty.

  - Render an offline assumptions vignette with separately labelled
    sampling intervals and allocation ranges, and verified
    hand-calculated examples.

  - Add Poisson/negative-binomial count models, optional lme4 random
    effects and exploratory residual Moran diagnostics.

  - Add multivariate CAR disparity with separate ethnicity fields, full
    joint posterior ratios, exceedance probabilities, convergence
    diagnostics and maps.

  - Add a synthetic lattice generator and replicated spatial recovery
    evidence.

  - Add three separate hit-rate diagnostics and a time-quality-gated
    darkness design with explicit UK daylight-saving transitions and
    exclusion audits.

  - Correct the published Other Black archive label, audit every raw
    sample ethnicity label, and warn on unrecognised future source
    classifications.