Consider recipe's `skip` with calibration

If we set skip = FALSE, it breaks.

If skip = TRUE, we might want to do something a little less subtle than the NA in the print output.

library(workflows)
library(recipes)
#> Loading required package: dplyr
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
#> 
#> Attaching package: 'recipes'
#> The following object is masked from 'package:stats':
#> 
#>     step
library(parsnip)
library(tailor)

dat <- modeldata::sim_classification(1000)

rec_skip_false <- recipe(class ~ ., data = dat) |>
  step_filter(non_linear_1 > 0, skip = FALSE)

tlr_cal <- tailor() |>
  adjust_probability_calibration(method = "logistic")

workflow(rec_skip_false, logistic_reg(), tlr_cal) |>
  fit(data = dat, data_calibration = dat)
#> Error in `vctrs::vec_cbind()`:
#> ! Can't recycle `..1` (size 502) to match `..2` (size 1000).


rec_skip_true <- recipe(class ~ ., data = dat) |>
  step_filter(non_linear_1 > 0, skip = TRUE)

workflow(rec_skip_true, logistic_reg(), tlr_cal) |> 
  fit(data = dat, data_calibration = dat)
#> Registered S3 method overwritten by 'butcher':
#>   method                 from    
#>   as.character.dev_topic generics
#> ══ Workflow [trained] ══════════════════════════════════════════════════════════
#> Preprocessor: Recipe
#> Model: logistic_reg()
#> Postprocessor: tailor
#> 
#> ── Preprocessor ────────────────────────────────────────────────────────────────
#> 1 Recipe Step
#> 
#> • step_filter()
#> 
#> ── Model ───────────────────────────────────────────────────────────────────────
#> 
#> Call:  stats::glm(formula = ..y ~ ., family = stats::binomial, data = data)
#> 
#> Coefficients:
#>  (Intercept)  two_factor_1  two_factor_2  non_linear_1  non_linear_2  
#>      0.27458       2.48406      -2.42843       0.69884      -1.48342  
#> non_linear_3     linear_01     linear_02     linear_03     linear_04  
#>      0.39114      -0.21790      -0.09806      -0.09972       0.09155  
#>    linear_05     linear_06     linear_07     linear_08     linear_09  
#>      0.02849       0.11098      -0.14325       0.01839      -0.24682  
#>    linear_10  
#>     -0.19332  
#> 
#> Degrees of Freedom: 501 Total (i.e. Null);  486 Residual
#> Null Deviance:       695 
#> Residual Deviance: 366.9     AIC: 398.9
#> 
#> ── Postprocessor ───────────────────────────────────────────────────────────────
#> 
#> ── tailor ──────────────────────────────────────────────────────────────────────
#> A binary postprocessor with 1 adjustment:
#> 
#> • Re-calibrate classification probabilities using logistic method.
#> NA
#> NA
#> NA

^{Created on 2026-03-20 with reprex v2.1.1}

Start by reproducing the two examples using workflows, recipes, parsnip, and tailor, focusing on workflow() with fit(), step_filter(), and adjust_probability_calibration(). Trace how calibration handles the rows removed when skip is FALSE and how the trained workflow is printed when skip is TRUE. Done means the reported behavior is resolved without the size mismatch or unexplained NA output.