Add a vignette, and fix the row reordering it exposed

[chross22]

The README explains but never demonstrates — every example in it is eval = FALSE, because they need Copernicus credentials. The vignette runs.

Its covariates are synthetic (a Gulf of Maine grid with a seasonal cycle and winter cloud gaps), so every number and figure is computed at build time, while the calls are the ones you would make against a real fetch.

That is the part the README cannot do:

• coverage dropping to 0.17 in January and back to 0.92 in June is a sentence in the README and a figure here
• min_coverage reporting 1 cell rather than 39 is an argument there and an output here
• downscale_grid(method = "nearest") inventing no values is a claim there and TRUE here

Writing it found a bug

matchData() processes rows a period at a time, so a table whose periods were interleaved came back grouped by period rather than in the order it arrived:

input  id: 1 2 3 4
output id: 1 3 2 4

Anyone aligning the result against the input by position — cbind(), or assigning a column straight across — would have got silently mismatched rows.

Worse: I had documented the opposite this afternoon, in both the roxygen and the vignette draft, having assumed it rather than checked. The order is now restored before returning, which makes the claim true rather than retracting it.

A test pins it, and checks the covariate follows its own row rather than merely the row count — the failure mode where row counts match but values are shuffled.

Also

• Vignette infrastructure in DESCRIPTION (knitr, rmarkdown, VignetteBuilder).
• figure/ — knitr scratch output from running knit() directly rather than through R CMD build — is gitignored after briefly being committed.

Verification

R CMD check — Status: OK, including checking re-building of vignette outputs ... OK. Full suite green. Installed into both R 4.6.1 and 4.3.2.

Note: pandoc is not on this machine's PATH; the build used the copy RStudio bundles. CI has its own.

🤖 Generated with Claude Code