[estevesjh]

Summary

HodMor.buzzard() now carries parameters fitted back to the delivered mock catalogue (mock_lob_sigma_catalog.fits: 1.2M halos, 129 log-mass × 8 redshift bins), instead of the values transcribed from the generator's ini.

• Every fitted parameter lands within 0.5σ of the generator's input, so this is a re-measurement of what the catalogue actually realises, not a correction.
• log10_Mmin is held at the generator's value: it sits a factor 40 below the catalogue's mass floor, where it is a 2.4% curvature at 10¹³ and 0.02% at 10¹⁵.
• The fit has to carry the catalogue's λ_tr ≥ 5 threshold. The delivered file only holds halos the mock's target loop walked, so a binned mean of λ_tr is a tail mean; fitting 1 + <λ_sat> straight to it returns α = 0.482, 44% too shallow.
• Documented in the docstring: pdf is a continuous shifted-Poisson interpolation, not the mock's unclipped draw, so it is not the mock's p(λ_tr | M).

Tests

• The des_y1 / buzzard comparison relaxes from bit-identical to percent-level agreement (log10_Mmin still exactly equal; epsilon and z_pivot remain the differences that matter).
• New: test_the_fitted_buzzard_set_recovers_the_generator_inputs checks all four fitted parameters are within 0.5σ of the generator's values.
• tests/test_selection.py: 71 passed. Full suite on this branch: 877 passed, 22 skipped, 33 xfailed.

Notes

• Rebased onto current main (no conflicts).
• Independent of Covariance: credit Wu et al., fix chi_h^2, match cluster-lensing-cov, add validation ladder #8 (covariance); the two touch different files.

🤖 Generated with Claude Code