[smidl]

Closes #26.

The problem

Geometry2/Computation2 are named for being the second machine defined, not for the machine they describe. CONTRIBUTING.md uses the word "anchor" three times and never defines it — the defining section survives on the contributing branch and was dropped in #23, leaving line 63 reading "add an motors" mid-substitution. Detail in #26.

What changed

Pure rename plus docs — no behaviour change.

before	after
motors/motor1.py → Geometry, Computation	motors/synrm_3f_36s.py → Geometry, Computation
motors/motor2.py → Geometry2, Computation2	motors/synrm_5f_40s.py → Geometry, Computation

One module per anchor, named <class>_<phases>f_<slots>s. Each exports plain Geometry/Computation, so the module name carries the identity and the class names stay uniform:

from motors.synrm_5f_40s import Computation, Geometry

The numbering existed only because both machines shared one namespace. Per-anchor modules remove the need for it, rather than trading Geometry2 for a longer awkward class name.

Turns are deliberately not in the name. Nc is routinely swept, so a name containing it goes stale; the name carries only what the anchor fixes for good — class, phases, slots. Same reasoning for currents and bus voltage.

Old paths kept as deprecation shims. motors/motor1.py and motors/motor2.py re-export the old names, so branches predating this keep importing and migrate on their own schedule. Verified: motors.motor2.Geometry2 is motors.synrm_5f_40s.Geometry → True.

Also: each anchor module gets a one-line docstring naming its machine, the two test modules follow the anchor names, and CONTRIBUTING regains the anchor definition with the motors/ layout row and the "add an motors" sentence fixed.

Verification

Both anchor modules were diffed against the motor1.py/motor2.py they replace, with the class renames normalised away: identical apart from the rename and the docstring.

• pytest → 26 passed, 1 deselected — same as main
• ruff check . / ruff format . --check → clean

One thing reviewers should know about renames

#28 and #29 both landed in motor2.py while this branch was open, and this PR produced no merge conflict for either — because moving a file makes the new path an unrelated add, so git replays the branch's copy verbatim and silently drops what landed meanwhile. Both fixes (np.array unit conversion + its import numpy as np, and the set_derived_params stub removal) had to be carried across by hand and re-verified by diff.

That is an argument for landing a rename promptly rather than letting it age: every subsequent fix to the moved file has to be hand-carried, with nothing to warn you.

[sadda]

Please use git mv instead of a hard-copy so that I can see the actual changes.

[smidl]

Good catch — I did use git mv, but the deprecation shims I left at motors/motor1.py and motors/motor2.py meant both the old and new paths existed at the branch tip, so git paired base↔head on the old path and reported the anchor module as a new file. It rendered exactly like a hard copy. Splitting into commits wouldn't have helped, since the PR view diffs base against head.

Dropped the shims and force-pushed. It now reports as renames:

 CONTRIBUTING.md                                 | 22 +++++++++++++++++++---
 motors/{motor1.py => synrm_3f_36s.py}           |  2 ++
 motors/{motor2.py => synrm_5f_40s.py}           | 12 +++++++++---
 notebooks/run.py                                |  2 +-
 notebooks/run_optimization.py                   |  2 +-
 tests/test_ansys_regression.py                  |  2 +-
 tests/{test_design.py => test_synrm_3f_36s.py}  |  2 +-
 tests/{test_design2.py => test_synrm_5f_40s.py} | 10 +++++-----
 8 files changed, 39 insertions(+), 15 deletions(-)

39/15 instead of the previous 741/699. The only content changes are a one-line docstring per anchor module, the class renames in the 5-phase one, and import paths.

I also reverted the test-fixture renames I'd made (geometry2 → geometry etc.). They were churn that dropped the test file below git's rename-similarity threshold without serving the point of the PR; the fixture bodies change, the names don't. Happy to tidy those separately if you want them consistent.

Verified independently of the diff: both anchor modules were compared against the motor1.py/motor2.py they replace with the class renames normalised away — identical apart from the docstring, including the #28 and #29 fixes that landed while this branch was open. pytest 26 passed / 1 deselected, same as main; ruff clean.

One consequence of dropping the shims: callers must update the import path, and there is no alias, so a stale import fails loudly instead of silently resolving. That affects pl_mag2.0, the newmodel_haitao branches and my own vs_5f — a one-line change each. I think loud is the right trade here, but say the word if you'd rather keep shims for a release and accept the noisier diff.