From the PR #6 silent-failure review — three compounding problems with the statscan profile:
config/statscan.yml is loaded into cfg$local_config, which nothing reads — every key in it (RDC paths, cycles) is silently ignored.- The profile never sets
data_source: master, so survey_var() falls back to pumf resolution at the RDC: PUMF variable names, PUMF bounds (age max 85 not 110), PUMF rows of the variables sheet.
- On a machine where the default paths exist, the statscan profile runs PUMF data while claiming to be a Master configuration, with zero complaint.
Fix: set data_source: master in the profile; merge statscan.yml keys to the top level (or read local_config explicitly); add a startup assertion that profile name and data_source agree. Also from the review: add a failing imputation_health target that consumes analysis_data$logged_events (a warning shown once under {targets} caching is effectively silent — a red target is not), and consider the universe-nested smoking predictor refinement for the imputation predictor matrix (structural-NA variables currently excluded as predictors entirely; nesting would recover information, e.g. quit-time predicting initiation age among former smokers only).
From the PR #6 silent-failure review — three compounding problems with the
statscanprofile:config/statscan.ymlis loaded intocfg$local_config, which nothing reads — every key in it (RDC paths, cycles) is silently ignored.data_source: master, sosurvey_var()falls back topumfresolution at the RDC: PUMF variable names, PUMF bounds (age max 85 not 110), PUMF rows of the variables sheet.Fix: set
data_source: masterin the profile; merge statscan.yml keys to the top level (or readlocal_configexplicitly); add a startup assertion that profile name anddata_sourceagree. Also from the review: add a failingimputation_healthtarget that consumesanalysis_data$logged_events(a warning shown once under {targets} caching is effectively silent — a red target is not), and consider the universe-nested smoking predictor refinement for the imputation predictor matrix (structural-NA variables currently excluded as predictors entirely; nesting would recover information, e.g. quit-time predicting initiation age among former smokers only).