Collect prompt-grid checkpoints into one reusable results table

Read every valid prompt-grid checkpoint in one or more directories and combine it with optional prior results. This is useful after cost-gated phases because checkpoints for models that are inactive in the current configuration remain available for later analysis and reuse.

Usage

collect_prompt_grid_results(output_dir, existing_results = NULL)

Arguments

output_dir

Character vector of checkpoint directories created by run_prompt_grid().

existing_results

Optional results data frame, run bundle, RDS path, or CSV path accepted by the existing_results argument of run_prompt_grid().

Value

A tibble of raw prompt-grid results. It can be passed directly to run_prompt_grid(existing_results = ...).

Details

Checkpoint rows are identified by the strong task_hash stored with each checkpoint and by input_row. Exact duplicate rows are removed. Conflicting rows with the same task and input identities cause an error rather than being silently resolved.