Skip to content

OperatorMark 1.0.0-rc5 — source and evidence

Pre-release
Pre-release

Choose a tag to compare

@Hc1012 Hc1012 released this 25 Oct 15:14
· 3 commits to main since this release
2f4a753

Initial public release candidate for OperatorMark: Ownership Evidence in Useful Copies of Neural PDE Operators, by Hadi Chamas.

Download

Download OperatorMark_1.0.0-rc5.zip below. It contains the complete source directory tree, compact evidence, analysis scripts and scientific documentation: 444 files, approximately 4.2 MB compressed. Extract its OperatorMark folder and follow the README inside.

The automatically generated Source code archives contain only this repository's overview and attribution files. Use the named artifact ZIP for the full research package. SHA256SUMS.txt supplies the artifact's SHA-256 digest.

Scientific scope

This is a ten-owner study of active and passive ownership evidence on one discrete Darcy-flow benchmark. Seven useful Jacobi-corrected copies defeat both evaluated signals. Larger derivative-labeled attacks preserve utility and remove the active mark while retaining passive evidence; their advantage over matched-access scratch training is unmeasured. The frozen primary, intact O05 failure, historical negatives and conditional scope limits remain explicit.

The archive supports saved-result reproducibility, not independent raw-model or end-to-end campaign replication. Full corpus, checkpoints, teacher caches and private keys are not distributed. Physical qualification and formal custody assurance are not claimed.

Publication status

The preprint Watermark Removal in Neural PDE Operators: Limits of Passive Ownership Evidence (v0.7) is now public on Zenodo: doi:10.5281/zenodo.23646972. That record also supplies the manuscript's LaTeX source and OperatorMark_Manuscript_Reproduction.zip, containing the Figure 1 and Table 2 rendering scripts and per-owner records. The manuscript and its LaTeX source are licensed under CC BY 4.0; the existing research artifact and reproduction-supplement code retain their MIT licenses. The rc5 archive, source pins and scientific results are unchanged. No experiments, numerical solvers or statistical analysis were rerun for this publication metadata update.