[repo-improver] test-coverage: cover check_input_data validation error branches

[Mike Tokic (mitokic)]

Evidence

R/input_checks.R::check_input_data() contains a number of detailed, user-facing
validation error messages (several recently enhanced) that had no regression
coverage in tests/testthat/test-input_checks.R. A baseline reproducer confirmed
the exact current message for each branch, so the behavior is well-defined but
unguarded against accidental regressions.

Change

Add focused testthat coverage for the previously untested check_input_data
validation branches. No source behavior is changed — this is test-only:

• combo variables missing from the input data
• target variable missing from the input data
• external regressors missing from the input data
• non-numeric target variable
• missing Date column
• non-date-formatted Date column
• uneven day-of-month spacing for month data, plus a positive test that
  date_type = "week" bypasses that check
• fiscal_year_start outside 1–12
• duplicate combo-Date rows

Dependency decision

No new dependency added. Existing capabilities were sufficient: testthat
(already in Suggests) and base R. This is test-only coverage, so the dependency
ladder resolves at step 2 (existing declared dependencies).

Validation

Each category was run in its own process with
Rscript -e "devtools::test(filter = '<filter>', stop_on_failure = TRUE)",
affected category first:

Category	Filter	Result
focused reproducer	^input_checks$	FAIL 0 · PASS 39
data-input-preparation	^(input_checks|hash_data|prep_data|clamp_negative_target|undifference)$	FAIL 0 · PASS 262
forecasting-core	^(forecast_time_series|multistep|hierarchical)$	FAIL 0 · PASS 237
model-selection-results	^(best_models|final_models|summarize_models)$	FAIL 0 · SKIP 2 · PASS 700
agent-run-lifecycle	^(agent|finalize_run|load_run_results)$	FAIL 0 · PASS 155
foundation-model-integrations	^(foundation-model-credentials|chronos|timegpt|timesfm)$	FAIL 0 · SKIP 17 · PASS 281

All 18 test-*.R files map exactly to the five configured categories; no
unclassified files. No roxygen/source changes, so devtools::document() was not
required.

Rscript -e "devtools::check(args = '--no-tests')" completed all package checks
but errored only at vignette re-building ('xfun::attr()' is deprecated). This is
a pre-existing environment/toolchain limitation: the identical command on clean
main fails at the same vignette step with the same error, so it is unrelated to
this test-only change. All focused and category tests pass.

Risk and limitations

Blast radius 1 (test-only; no runtime or public-API change). These tests lock in
the current user-facing validation messages, so intentional future message edits
must update the corresponding assertions. The only unavailable gate is the
pre-existing vignette build error noted above.

Automation metadata

• Scout run: 2026-08-03 18:22 PDT (America/Los_Angeles)
• Source SHA: 53908f7d75ea2f37ccf269afada8bc62e99bc5c7
• Recommendation fingerprint: 82eda7658e3d9c96cbfdb0c1b913ca8671b273281591527a139b1ccb2bdd6c77

[Copilot]

Pull request overview

Plan (review focus):

• Verify new tests match current check_input_data() validation branches and messages.
• Check that “positive” tests isolate the intended behavior and won’t fail for unrelated future validations.
• Ensure changes are test-only and consistent with existing test conventions.

This PR adds regression coverage around R/input_checks.R::check_input_data() by asserting the current user-facing validation errors across previously untested branches, helping prevent accidental message/behavior regressions without changing runtime behavior.

Changes:

• Added testthat cases for missing required columns (combo vars, target, xregs) and invalid target type.
• Added coverage for Date column presence/type, month day-of-month consistency, fiscal year start range, and duplicate combo–Date rows.
• Added a “week” positive test intended to confirm the month day-of-month check is bypassed for date_type = "week".

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.