scrollprize

/

ink_9um

Like 1

Follow

Vesuvius Challenge 33

Hello scrollprize ink_9um maintainers,
I am reviewing the lineage of hybrid_3d2d-seed42/step-075000.pth from revision 7109667e2607db1b90c37c8b09cb876ea7fe7bb1.
Checkpoint SHA-256: e635558ae6a1a807a7e5ec1e83adfd45bc3c0ac53883ea43f1d4e085d62a9cab.
The saved sampler contract matches the embedded manifest SHA-256 0e20230f4d2517987f31baed4c6fc2fa8110d230ce2d14a60dfe5d6a838c68d1. PHercParis2Fr47/Frag1 is absent from the declared representations, but that alone does not establish independence.
Could you point to existing records for:

1. The generating run/commit and effective prepared-data/sampling manifest, identifying label/mask revisions and whether Frag1, aliases or overlapping renders were included in training or validation?
2. Any pretrained initialization, resume/warm-start, teacher or pseudo-label ancestry, with checkpoint/data identifiers—or an explicit record that these were not used?
3. Whether Frag1 influenced architecture, seed, checkpoint, offset or release selection, and which validation/selection records support this?
   A public or redacted manifest/short statement is sufficient. “Unknown” or “unavailable” would also help. I am not requesting new training/evaluation or asserting contamination. Please redirect me to the appropriate maintainer if needed.
   Thank you.