artifact_granularity

Back to L8 | Browse all axes | Browse all options

Axis artifact_granularity on sub-layer L8_D_artifact_granularity (layer l8).

Sub-layer

L8_D_artifact_granularity

Axis metadata

  • Default: 'per_cell'

  • Sweepable: False

  • Status: operational

Operational status summary

  • Operational: 5 option(s)

  • Future: 0 option(s)

Options

flat – operational

Single flat directory; cells distinguished by filename suffix.

All cells write into one directory; cell IDs become filename suffixes. Useful for sweeps with thousands of small artifacts where per-cell directory creation is wasteful.

When to use

Sweeps with thousands of small artifacts.

References

  • macroforecast design Part 3, L8: ‘reproducibility = manifest + provenance + bit-exact replicate.’

Related options: per_cell, per_target, per_horizon, per_target_horizon

Last reviewed 2026-05-05 by macroforecast author.

per_cell – operational

One sub-directory per sweep cell (default).

Default. Each cell gets its own cell_NNN/ directory containing the cell’s full set of artifacts. Isolates every cell’s output for clean per-cell inspection.

When to use

Default; isolates each cell’s artifacts.

References

  • macroforecast design Part 3, L8: ‘reproducibility = manifest + provenance + bit-exact replicate.’

Related options: per_target, per_horizon, per_target_horizon, flat

Last reviewed 2026-05-05 by macroforecast author.

per_horizon – operational

Group cells by forecast horizon (one directory per horizon).

When the sweep varies over multiple horizons, groups by horizon. Useful for horizon-by-horizon analysis.

When to use

Multi-horizon sweeps. Selecting per_horizon on l8.artifact_granularity activates this branch of the layer’s runtime.

References

  • macroforecast design Part 3, L8: ‘reproducibility = manifest + provenance + bit-exact replicate.’

Related options: per_cell, per_target, per_target_horizon, flat

Last reviewed 2026-05-05 by macroforecast author.

per_target – operational

Group cells by target variable (one directory per target).

When the sweep varies over multiple targets, this groups the artifacts by target rather than by cell. Useful when downstream analysis is per-target.

When to use

Multi-target studies where target-grouping aids browsing.

References

  • macroforecast design Part 3, L8: ‘reproducibility = manifest + provenance + bit-exact replicate.’

Related options: per_cell, per_horizon, per_target_horizon, flat

Last reviewed 2026-05-05 by macroforecast author.

per_target_horizon – operational

Group cells by (target, horizon) pair.

Combines per_target and per_horizon for sweeps that vary across both axes.

When to use

Studies sweeping across both target and horizon.

References

  • macroforecast design Part 3, L8: ‘reproducibility = manifest + provenance + bit-exact replicate.’

Related options: per_cell, per_target, per_horizon, flat

Last reviewed 2026-05-05 by macroforecast author.