Experimental Paths¶
The main branch keeps the stable public API, its direct implementation paths,
the supported Python optimizer module, and reproducible scripts under
examples/. Larger experiments are preserved in the archive tag
archive/experimental-snapshot-2026-06-13 instead of remaining importable from
the package.
What Was Archived¶
- large generated inversion and LSRTM result bundles
- CUDA profiling, sweep, and comparison scripts
- compact CPML layout and split-PML launch prototypes
- alternate snapshot, checkpoint, and direct material-gradient runtime paths
Why They Are Not In Main¶
These paths were useful for research iteration, but they were not ready to carry as supported package surface:
- their APIs were not stable enough to document as user-facing behavior;
- backend support differed across CPU, CUDA, Python fallback, and storage modes;
- the test matrix was too broad for routine release verification;
- some paths depended on large local datasets or generated artifacts;
- keeping them beside the stable solver made the codebase harder to audit.
The archive tag keeps the work recoverable without making it part of the main branch maintenance contract.