[rozyczko]

Code:

• Removed InterfaceFactory and associated logic.
  This change moved the calculator ownership into the technique specific libraries, allowing a clear separation of responsibilities.

• Removed Parameter's callback keyword and value_no_call_back

• Dropped AnalysisBase(interface=...) and made AnalysisBase arguments keyword-only.

Documentation:

• Added a detailed new user guide page, external-calculators.md, explaining how to connect an external calculator to an EasyScience model and Fitter. With diagrams, examples, and best practices for integration and fit control.

Cleanup:

• Removed a number of legacy examples.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.96%. Comparing base (bdc119f) to head (801d28d).
⚠️ Report is 2 commits behind head on release-3.0.

Additional details and impacted files

@@               Coverage Diff               @@
##           release-3.0     #333      +/-   ##
===============================================
- Coverage        94.08%   93.96%   -0.13%     
===============================================
  Files               50       48       -2     
  Lines             4094     3946     -148     
===============================================
- Hits              3852     3708     -144     
+ Misses             242      238       -4     

Flag	Coverage Δ
unittests	93.96% <100.00%> (-0.13%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/easyscience/fitting/engine_base.py	100.00% <ø> (ø)
src/easyscience/fitting/fitter.py	96.03% <ø> (ø)
src/easyscience/job/analysis.py	100.00% <100.00%> (ø)
src/easyscience/variable/descriptor_base.py	95.74% <ø> (ø)
src/easyscience/variable/parameter.py	95.22% <ø> (-0.03%)	⬇️

[rozyczko]

Thinking about Model with fit function being the only thing passed. This means Fitter would treat Model as callable (if the second argument - fit_function wasn't defined). But I assume the whole point here is to have Model be pure state only.
The calculator is in the library and it READS the model. If ModelBase got a fit_function slot, the model would hold a reference to a calculator again. We just removed this coupling!

The model would have to know its calculator and have to reach a backend. Either the model stores a calculator reference, which is exactly what we have removed, or it must find it through (hidden) global state.
This would break the "pure state" contract.

I mean... we could easily do it: Fitter(fit_object, fit_function=None). When no callable function is given, we'd use the model itself if it is callable, otherwise raise a TypeError saying a fit function is required.

So from the user POV this is easier, but internally, it introduces things we wanted to be removed in the first place.

Also, a model may have many fit functions (e.g. in reflectometry my model would have separete callables per channel and per resolution function). A single method per model won't be abnle to describe that.

Serialization - we would have to explicitly skip the calculator slot and from_dict would decode an empty, calculator-free model. You'd have to reattach it. Smells bad.

Currently, the model stays pure state. The fitter knows only get_fit_parameters() and a callable. Fitter can easily be tested with a lambda.

[damskii9992]

Just some tests that needs to be removed. But otherwise this LGTM.