Detect metric columns in a data frame, dataset-agnostically

Returns the names of numeric columns in data that can sensibly be summarized as quantitative metrics in per-theme + per-subtheme tables. Package-internal columns (those pakhom engineers itself – sentiment_score, emotion_intensity, theme_membership_*, etc.) are excluded; everything else numeric is a candidate.

Usage

.detect_metric_columns(data, config = NULL, explicit = NULL)

Arguments

data

tibble with the analytical data (post-cascade)

config

Optional ThematicConfig or config list

explicit

Optional character vector of metric column names to use verbatim (intersected with the data's columns). Bypasses the config dig.

Value

Character vector of metric column names (possibly empty)

Details

Explicit override path: when config$data$column_mappings$metric_columns is non-empty, those names are used verbatim (intersected with the data's columns to avoid referencing missing fields). This matches the explicit detect_columns() mapping path in R/07_data_loading.R.

Mirrors (and consolidates) the metric-column detection used by prepare_correlation_data (R/14_correlations.R).

Caveat – sentiment_score collision

The auto-detect path excludes sentiment_score (the package- engineered column from R/10_sentiment.R) by name. If a user's corpus happens to have its own sentiment_score numeric column that they want treated AS A METRIC, the auto-detect silently drops it. Workaround: supply an explicit override via config$data$column_mappings$metric_columns (or the direct explicit= arg) – the override path returns the requested columns verbatim, bypassing the internal-column exclusion. The same collision applies to any other internal name (emotion_intensity, n_themes, etc.).