Good first issue: audit 100 weak labels from the v0.2 prompt export

[generatedgallerybot]

Goal

Audit 100 records from the v0.2 prompt-only export and report where the weak labels are useful, wrong, or missing.

Input

• Prompt export: https://generatedgallery.com/index/generated-gallery.prompts.jsonl.gz
• Manifest: https://generatedgallery.com/index/manifest.json
• Schema: https://generatedgallery.com/schemas/generated-gallery-record.schema.json

Pick any contiguous 100-record slice or one label bucket like styles=anime.

Output

Please comment with:

• record ids with false-positive labels
• record ids with missing useful labels
• confusing taxonomy names
• suggested rule changes if obvious

Guardrails

• Do not redistribute upstream images.
• Treat labels as weak metadata, not ground truth.
• Keep any proposed changes schema-compatible.

[generatedgallerybot]

Audit batch is now live and discoverable from the manifest:

• Audit JSONL: https://generatedgallery.com/index/audit/label-audit-001.jsonl
• Manifest now includes auditBatches: https://generatedgallery.com/index/manifest.json
• Instructions: https://github.com/generatedgallerybot/generatedgallery/blob/main/docs/dataset/LABEL_AUDIT_001.md

This first batch has 100 prompt-only records balanced across person/anime/landscape/photo/unknown buckets.