Understanding validation status

Inspect the method before using a result

validation_status("t_means_two_independent")
#> <ig_validation_status> t_means_two_independent
#>   label:                 Means: Difference between two independent means (two groups)
#>   implementation status: implemented
#>   validation status:     independently_checked
#>   G*Power reference:     reference_pending
#>   independent check:     passed
#>   documented examples:   DOC-001, DOC-002, DOC-003, DOC-005, DOC-006, DOC-007, DOC-008, DOC-009, DOC-010, DOC-011, DOC-012, DOC-013, DOC-014, DOC-015, DOC-016, DOC-017, DOC-098, DOC-099
#>   - G*Power reference comparison is not fully certified for this method. Available captures and passing regression cases are limited evidence; complete agreement with G*Power is not claimed.
#>   - Documented manual examples are secondary evidence, not application captures.

independently_checked records the existence of independent numerical checks. It does not mean every option combination is certified. reference_pending means complete GPower comparison is not certified; captured cases may already exist. The Boschloo extension is outside the GPower inventory and therefore has G*Power reference status not_applicable.

validation_status("boschloo_proportions_two_independent")
#> <ig_validation_status> boschloo_proportions_two_independent
#>   label:                 Proportions: Inequality, two independent groups (Boschloo's unconditional exact test)
#>   implementation status: implemented
#>   validation status:     independently_checked
#>   G*Power reference:     not_applicable
#>   independent check:     passed
#>   documented examples:   none
#>   - This procedure is not part of the G*Power inventory; no G*Power reference exists.

Standalone Boschloo power and sample-size functions have independent checks. The explicit advisor_power_spec() bridge preserves the selected central, minlike or prespecified directional variant for completed, mapped advisor recommendations. Supply event probabilities, direction and allocation explicitly. Unmapped or incomplete recommendations remain unavailable for automatic planning; the mapping does not certify every numerical branch.

Inspect the actual outcome

spec <- analysis_spec(
  "t_means_two_independent", "a_priori",
  inputs = list(alpha = 0.05, power = 0.8, d = 0.5),
  options = list(tails = "two")
)
result <- solve_design(spec)
result$status
#> <ig_status> ok
result$warnings
#> character(0)
result$validation
#> $documented_example_ids
#>  [1] "DOC-001" "DOC-002" "DOC-003" "DOC-005" "DOC-006" "DOC-007" "DOC-008"
#>  [8] "DOC-009" "DOC-010" "DOC-011" "DOC-012" "DOC-013" "DOC-014" "DOC-015"
#> [15] "DOC-016" "DOC-017" "DOC-098" "DOC-099"
#> 
#> $fixture_ids
#> character(0)
#> 
#> $gpower_reference_status
#> [1] "reference_pending"
#> 
#> $independent_checker
#> [1] "validation/independent/t/check_t_means_two_independent.R"
#> 
#> $method_validation_status
#> [1] "independently_checked"

Use a design estimate only after checking its typed status, warnings and assumptions. In particular, exact unconditional compromise and tetrachoric approximation confidence intervals still have pending branches. Classifier bootstrap validation remains at development status.

Reproduce the public checks

The source repository includes an installed-package checker with synthetic inputs and separate reference calculations. It emits per-case comparisons and a run summary, and exits unsuccessfully on a mismatch. Its scope and tolerances are documented in the public validation instructions.

Separate bounded numerical campaigns exercise exact table enumeration, inverse targets, rank and regression models, multivariate approximations and utility calculations. Their method/mode/option inventory distinguishes an independently checked sample from an uncovered branch or an intentional refusal. A finite sample for every registered method is not certification of all its inputs. The numerical contracts state the oracle, tolerance, supported scope and remaining boundary for each work package. The installed release preflight separately checks API/help availability, startup effects and report round-trips under full and minimal dependency profiles.

The package regression suite and this independent checker answer different questions. Neither replaces full reference-corpus comparison. Some provenance checks skip because development inventories, captured fixtures and ledgers are not shipped. The validation guide lists those skips and the roadmap defines the remaining qualification gates.