Analysis¶

RunDescriptor `dataclass` ¶

Metadata describing a single experiment run.

Attributes:

Name	Type	Description
`model`	`str`	Model name used for the run.
`run_id`	`str`	Run identifier (e.g., directory name).
`label`	`str`	Label describing the run variant.
`summary_path`	`Path`	Path to the summary JSON file.
`analysis_dir`	`Path`	Directory containing analysis artifacts.
`feature_mode`	`str or None`	Optional feature mode (e.g., vision context setting).

build_summary_table(runs)

Tabular summary of discovered runs.

discover_runs(results_root)

Discover run analysis folders beneath results/<model>/run_*/.

feature_mode_for(desc)

Return the feature mode associated with a run descriptor.

feature_mode_suffix(desc)

Pretty suffix to annotate feature mode in labels.

latest_run(runs, model, label, feature_mode=None)

Select the newest run matching the provided filters.

load_evaluation_npz(desc)

Load (and cache) evaluation data for a run.

load_scenarios_npz(desc)

Load (and cache) scenario data for a run.

load_summary(summary_path)

Load a summary JSON and attach its path.