Re-run a result against the engines installed now

A result records which version of its engine produced it (see glance()'s engine_version). cpt_verify() answers the question that record exists for: does the answer still hold? It re-runs the same method on the same series with the settings the result recorded, using whatever engines are installed today, and reports whether the changepoints moved and which versions changed in between.

Usage

cpt_verify(object, ..., tolerance = 0)

# S3 method for class 'ggcpt_verification'
print(x, ...)

Arguments

object

A ggcpt object from cpt_detect().

...

Further arguments for the re-run, overriding what was recovered from the result (for instance an engine argument the recorded call held as a variable, which cannot be recovered from the object).

tolerance

Positions a changepoint may move and still count as the same. Defaults to 0: an exact match.

x

A ggcpt_verification object.

Value

A ggcpt_verification object: a list with verified (TRUE when every changepoint reproduces within tolerance), then and now (the two sets of locations), added and removed (locations in one set with no partner in the other), versions (a tibble with component, then, now and changed), not_recovered (call arguments that could not be replayed) and refit, the new ggcpt. With a print() method.

Details

Upstream engines change: a default is revised, a bug is fixed, a tie-breaking rule moves. None of that raises an error, so a result produced last year and re-run today can differ without anything saying so. This is the check to run when re-opening an analysis, and before quoting an old result as current.

The re-run uses the series stored on the result, its method, the change type it can be asked for, its penalty, and every engine argument of the original call that was written as a literal (seed = 42, n_intervals = 500). An argument written as a variable cannot be recovered from the object, and is listed in not_recovered; pass it again through ... if it mattered.

See also

glance.ggcpt() for the recorded engine_version.

Other result class: annotate_segments(), as_cpt_series(), as_ggcpt(), as_json(), cpt_annotations(), cpt_export(), ggchangepoint-conditions, is_ggcpt(), new_ggcpt(), print.ggcpt()

Examples

set.seed(2026)
fit <- cpt_detect(c(rnorm(80), rnorm(80, 4)), method = "pelt")
cpt_verify(fit)
#> ggcpt_verification (method: pelt)
#>   Verdict:            reproduces with the engines installed now
#>   Changepoints then:  80
#>   Changepoints now:   80
#> 
#> Versions:
#> # A tibble: 3 × 4
#>   component            then  now   changed
#>   <chr>                <chr> <chr> <lgl>  
#> 1 changepoint (engine) 2.3   2.3   FALSE  
#> 2 ggchangepoint        0.6.0 0.6.0 FALSE  
#> 3 R                    4.6.1 4.6.1 FALSE