Datasets:

garn-garn
/

smartgridbench-review-artifact

The source snapshot SHA identifies the reviewed repository state used to prepare this package. The evidence capture floor SHA identifies the clean source state used by the main paper-grade experiment captures. The artifact import SHA identifies the final repository state used for the reviewer-facing evidence refresh.

Included

36 canonical scenario JSON files: 31 hand-authored scenarios plus 5 generated scenarios promoted after validation and manual review.
5 negative schema fixtures.
Paper-grade evidence registry and summary CSVs.
Compact per-trial judge scores for paper-grade rows, with private URLs and absolute operator paths removed.
Current failure-taxonomy surface and 46-row stratified manual-audit sidecars with reviewer identities and machine paths redacted.
Current post-PR175 mitigation before/after export, with private run URLs and local generated-config paths redacted.
Compact profiling/W&B inventory, with private run URLs, run IDs, and private artifact names redacted.
12-row manual judge audit sanity table.

Excluded

Raw upstream utility or telemetry datasets that are not redistributed in this artifact.
VM-local logs, credentials, private run URLs, and full trajectory logs with environment-specific paths.
Generated scenario candidates that were not promoted into the canonical 36-scenario package.
Invalid or superseded cohorts except as explicit exclusion rows in data/evidence_registry.csv.

Anonymization Notes

This package is intended for double-blind review. It avoids author names, institutional names, private run URLs, local usernames, and machine-specific absolute paths. Relative benchmark paths are preserved when useful for joining evidence tables.