1. get_model_residual overwrites whatever file data_path names.

What happened: I passed the profile’s own path and it silently replaced parent-uio66.gr with a list of residual values. The tool name sounds read-only, and the description doesn’t say it writes a file.
Fix: call the argument output_path, refuse to overwrite existing files unless told to, and refuse when the path matches a loaded profile.

2. finite defaults to False, so .xyz clusters are silently treated as periodic.

What happened: an .xyz file has lattice 1 Å, so the periodic calculator gives garbage: oscillating G(r) and Rw of 6–12 at the correct parameters. Nothing warns about it. With finite=True the same parameters give Rw 0.2742.
Fix: use the finite (Debye) calculator by default for .xyz input, or warn when the lattice is 1 × 1 × 1.

3. There’s no way to evaluate the model without refining.

What happened: solve with an empty variable list fails with “Variables or tags cannot be found (all)”. My workaround was to “refine” delta1 with lb = ub = 0.
Fix: add an evaluate/residual call that returns Rw and χ² at the current values, or let variable_names=[] do that.

4. get_model_evaluation crashes before any solve.

Error: TypeError: Profile.set_observed_profile() got an unexpected keyword argument 'xname'.
When: it works only after a solve has built a recipe.

5. The solver barely moves from the starting values.

What happened: in all 25 grid points, s2 changed by less than 0.005 and s1 by about 0.015.
Likely cause: it stops once a parameter reaches a restrained bound. g22 Zr3 Uiso hit 1e-4 with σ = 1e-7 every time.
Comparison: your least_squares(x_scale="jac") gets much further from generic starts (Rw ≈ 0.285, versus 0.43–0.48 here).
Fix: expose the optimiser settings (method, x_scale, max_nfev, ftol/xtol) and report why it stopped and how many function evaluations it used.

6. Long solves time out with no progress feedback.

What happened: the slow periodic solves took more than 60 s and the call timed out on my side. I then had to poll list_recipe_parameters to see whether it had finished.
Fix: run the solve as a background job that returns an ID, plus a status call.

8. Constraints carry over between solves in the same session.

Workaround: clear and rebuild everything before each fit.