Acceleration Mirror Authority¶
SC-NeuroCore contains multiple acceleration trees under src/sc_neurocore/accel/.
They do not all carry the same authority.
Rule¶
Treat an acceleration file as authoritative only if at least one maintained Python execution path loads it directly and tests cover that path.
The polyglot acceleration tree is research-only unless it satisfies that rule.
It is not shipped as the production runtime, and it is not required by
pip install sc-neurocore.
Everything else is one of:
- research-only source
- partial port
- transcript mirror
- historical scaffold
Current authoritative Julia entrypoints¶
The maintained Julia surface is currently limited to files explicitly loaded by Python code or referenced by exercised parity paths:
src/sc_neurocore/accel/julia/world_model/predictive_model.jlsrc/sc_neurocore/accel/julia/chiplet/kl_refine.jlsrc/sc_neurocore/accel/julia/_native/learning_bridge.jlsrc/sc_neurocore/accel/julia/fault_injection/fault_injection.jl
Package-level reference:
sc_neurocore.accel.julia.AUTHORITATIVE_JULIA_ENTRYPOINTS
Current authoritative Mojo entrypoints¶
The maintained Mojo surface is currently limited to Python loaders and compiled library paths that are actually consumed from Python:
src/sc_neurocore/accel/mojo/runner.pysrc/sc_neurocore/accel/mojo/world_model/lgssm.mojosrc/sc_neurocore/accel/mojo/fault_injection/fault.mojosrc/sc_neurocore/accel/mojo/wong_wang/__init__.pysrc/sc_neurocore/accel/mojo/wilson_cowan/__init__.py
Package-level reference:
sc_neurocore.accel.mojo.AUTHORITATIVE_MOJO_ENTRYPOINTS
Non-authoritative mirror zones¶
These areas must not be treated as the source of truth unless a maintained Python loader and tests are added later:
src/sc_neurocore/accel/julia/studio/*.jlsrc/sc_neurocore/accel/julia/analysis/*.jlsrc/sc_neurocore/accel/julia/analysis_spike_stats/*.jl- large parts of
src/sc_neurocore/accel/mojo/kernels/*.mojo
Known transcript-style examples:
src/sc_neurocore/accel/julia/studio/compiler.jlsrc/sc_neurocore/accel/mojo/kernels/app.mojo
Practical workflow¶
When fixing behaviour:
- change the authoritative Python source or maintained compiled backend first
- update tests
- update docs
- only then refresh any mirrors if they are still worth keeping
Do not reverse that order. Updating a mirror first creates drift and false confidence.