Light contract-style data test template (conditional on observed drift)

[larnsce]

Milestone: v1.3.0 (attach once the milestone exists). Depends on #37. Conditional: only start if v1.1/v1.2 experience shows post-review drift in published packages (data updates silently breaking floor guarantees). Part of the v1.1 roadmap: docs/roadmap-v1.1.md.

Tasks

• Template tests/testthat/test-data.R asserting only stable contracts: data loads, non-empty, every column present in data-raw/dictionary.csv and vice versa. The tests read the dictionary at runtime, never interpolate literals.
• Explicitly banned assertion styles, from premortem finding F2 (snapshot tests become red-CI landmines on the first legitimate data update, and students respond with skip()): no dim() literals, no frozen date ceilings, no level-set equality.
• Data-update playbook (edit the dictionary first, rerun the tests) shipped in the package-resident standards file so future sessions in reviewed packages see it.

Acceptance criteria

A simulated data update (new rows, new category value) passes the template tests unchanged; a removed or renamed column fails them.