feat: add forecast_multivariate_point class

[seabbs-bot]

Summary

• Adds forecast_multivariate_point class with constructor, assertion, scoring, and metric methods following the forecast_multivariate_sample pattern
• Adds variogram_score_multivariate_point() metric that wraps scoringRules::vs_sample() treating point forecasts as single-sample ensembles
• Updates transform_forecasts() to correctly dispatch multivariate point types when recomputing .mv_group_id

Closes #1112

Test plan

• Constructor tests with renamed columns and expected structure snapshots
• is_forecast_multivariate_point() positive/negative checks
• get_metrics() returns expected metric list
• score() produces numeric variogram scores with snapshot output
• Error cases for missing joint_across and invalid columns
• Full test suite passes (720 tests, 0 failures)
• All changed files lint clean

This was opened by a bot. Please ping @seabbs for any questions.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.95%. Comparing base (42904b0) to head (fd26084).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1115      +/-   ##
==========================================
+ Coverage   97.87%   97.95%   +0.07%     
==========================================
  Files          35       37       +2     
  Lines        1881     1953      +72     
==========================================
+ Hits         1841     1913      +72     
  Misses         40       40              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[seabbs-bot]

Review of PR #1115

Overall this is a clean implementation that follows existing patterns well. A few items to consider:

Unnecessary nolint comments

1. R/class-forecast-multivariate-point.R:202 — # nolint: line_length_linter on score.forecast_multivariate_point. The line is 95 chars without the nolint comment, well under the 120-char limit in .lintr. The nolint is self-defeating (adds length). The equivalent score.forecast_multivariate_sample line has no nolint.

2. R/metrics-multivariate-point.R:18 — # nolint: object_name_linter line_length_linter on the variogram_score_multivariate_point function signature. The line is ~104 chars without the nolint (under the 120-char limit). And object_name_linter is set to NULL in .lintr, so it is disabled project-wide. The equivalent energy_score_multivariate function has no nolint at all.

Inherited docs mismatch

variogram_score_multivariate_point uses @inheritParams ae_median_sample and @inheritParams assert_input_multivariate_sample. This inherits a predicted parameter description that says "nxN matrix of predictive samples, n (number of rows) being the number of data points and N (number of columns) the number of Monte Carlo samples". For a point forecast wrapper where N is always 1, this is misleading. Consider writing a custom @param predicted that describes the actual input (a numeric vector or single-column matrix of point predictions).

transform-forecasts.R change

The generalisation from hardcoded as_forecast_multivariate_sample to dynamic dispatch via paste0("as_forecast_", forecast_type) is a good fix — it mirrors the pattern already used in the else branch and is necessary for the new class to work with transform_forecasts(). Looks correct.

Everything else looks good

• The "single-sample ensemble" approach for variogram scoring of point forecasts is sound
• Correct to omit the w (sample weights) parameter since it's meaningless with a single sample
• w_vs correctly maps to the variogram weight matrix in scoringRules::vs_sample()
• Test coverage is solid: creation, structure, type checking, metrics, scoring, error cases, snapshots
• Score method correctly simplifies the sample version by dropping the split-by-sample-length logic

[seabbs]

This looks good to me. Note a few related issues.

[seabbs]

@nickreich for awareness