This article refers to the engineering-scene simulation project record dated 2026-08-02 and reviews only the relationship between files and metadata in its gantry-crane image-level example batch. The original batch records 48 JPG images. This review combines no other packages and adds no model-training, field-performance or complete-annotation acceptance conclusion.
What the metadata records
| Fields | Recorded meaning | What they do not establish |
|---|---|---|
id, output_path |
Example identifier and output file location | Accepted image content or labels |
source_member |
Reference to a source member | Confirmed usage rights or permission to publish original images |
scenario |
Name of a generated scene | Coverage of real field distributions |
width, height |
Image dimensions | Verified measurement accuracy or model inputs |
classes_hint |
Class hints | Valid detection boxes, masks or character labels |
seed, params_json |
Random seed and transformation parameters | Reproduction in any environment or improved model performance |
Check identifiers and files rather than row order
During review, associate inventory rows and output files through example identifiers. Check duplicate identifiers, missing files, unlisted files and dimension differences. Source references and output paths provide traceability starting points. Similar filenames alone do not establish a common origin. Relationships should remain checkable after inventory rows are reordered.
Retain source relationships for derived outputs
If later work crops images, converts formats or changes generation parameters, create another version linked to source records rather than overwriting originals. Saved random seeds and transformation parameters help explain image generation. Regeneration also depends on original inputs, code, dependencies and execution conditions. A saved random seed alone is insufficient evidence that regenerated files will be identical byte for byte.
Traceability does not establish accepted performance
File-to-metadata correspondence establishes that record relationship only. It does not replace usage-rights review, task-annotation checks or independent real-data testing. This article publishes no customer source images, internal paths or complete business records. The record-keeping steps proposed here are recommendations for future work. They are not evidence of additional work completed for the original batch.
Sources: Existing engineering-scene simulation project record, the original publication record and batch metadata. Related services: Data, training and evaluation, project consultation.