Synthetic Asset Metadata Normalization Brief

Role: Lead Computer Vision Data Quality Engineer specializing in synthetic visual catalog curations and asset lineage management.

Context

• Asset Repository: {{synthetic_asset_repo}}
• Target Metadata Standard: {{metadata_schema_standard}}
• Corrupted Fields Inventory: {{corrupted_field_inventory}}
• Geometric Boundaries: {{resolution_aspect_bounds}}
• Target Storage & Export Format: {{export_format}}
• Downstream Fine-Tuning Target: {{downstream_model_target}}

Task

Draft an operational data cleaning and schema normalization brief to rectify corrupt generation parameters, sanitize prompt logs, and validate aspect-ratio metadata across our synthetic media catalog.

Method

1. Ingest asset logs from {{synthetic_asset_repo}} and isolate records missing core generation provenance.
2. Parse unstructured generation strings to extract explicit positive prompts, negative prompts, guidance scales, and sampler names.
3. Map missing or malformed attributes identified in {{corrupted_field_inventory}} to fallback structures dictated by {{metadata_schema_standard}}.
4. Cleanse prompt strings by stripping trailing escape characters, redundant weighting brackets, and unparsed seed flags.
5. Audit image dimension metadata against {{resolution_aspect_bounds}} to correct discrepancies between header values and physical pixel counts.
6. Standardize color space tags (sRGB vs Display P3) and embed cleaned JSON provenance sidecars in {{export_format}}.
7. Validate that cleaned metadata tokens strictly conform to the intake format needed for {{downstream_model_target}} fine-tuning runs.
8. Generate a metadata integrity score detailing schema conformance rate across the entire library.

Constraints

• MUST preserve historical generation seeds and model checkpoint hashes during string sanitization.
• MUST NOT overwrite original raw logs; all transformations must write to immutable versioned schemas.
• String sanitization regexes must handle nested weighting syntax without truncating token concepts.
• Schema output must guarantee zero null values for primary retrieval keys.

Output format

Deliver an engineering brief with the following distinct sections:

1. Metadata Health Diagnostic (bulleted summary with failure rates per field)
2. Field-by-Field Sanitization Rules (structured table: Raw Field, Parsing Logic, Default Rule, Output Type)
3. Prompt String Normalization Logic (ordered steps and regex criteria)
4. Validation Protocol (concrete test queries verifying schema conformance in {{export_format}})

Self-review

• Confirm all 6 variables ({{synthetic_asset_repo}}, {{metadata_schema_standard}}, {{corrupted_field_inventory}}, {{resolution_aspect_bounds}}, {{export_format}}, {{downstream_model_target}}) are integrated.
• Verify prompt cleaning preserves semantic weighting structures.
• Ensure export format requirements align with downstream fine-tuning specs.