v1.7.0

Highlights

Lyman-break bias now depends on the magnitude limit, not just redshift, and the
interloper bias is an input rather than a formula. Both change numerical output.
TomographicBin construction is also ~11× faster.

Changed — affects results

• LBG bias is now b(m, z), eq. 2.7 of Wilson & White 2019,
  replacing the redshift-only power law 0.28 (1+z)^1.6. That relation gave the same
  bias to samples 2 magnitudes apart in depth. For an LSST-like ten-year dropout
  selection the effective bias moves by factors of 1.4, 0.8 and 1.2 for u, g and r.
• Interloper bias is the new b_interlopers argument, constant, default 1.5.
  It was previously the LBG power law evaluated at low redshift.
• stretch no longer moves the mean. It pivoted on the midpoint of the redshift
  grid rather than the mean of the distribution, so a 10% stretch also shifted the
  mean by ~0.005. It now scales the width and nothing else.
• mag_bias uses a central difference, with the step exposed as dm_mag_bias.
• mag_cut and m5_det must be scalars. Arrays previously failed with a numpy
  broadcast error from inside the integrals; they now raise immediately.
• A shift that pushes every interloper below zero redshift raises.
• table0 on Completeness is now the table as read. Conditioning ran in place, so
  it had been the conditioned table despite its name.

Added

• lbg_bias(mag_cut, z), usable without building a bin. Warns outside 24 < m < 25.5,
  the depths Wilson & White fit, and if the relation returns non-positive bias.
• z_lbg, nz_lbg, z_interlopers, nz_interlopers on TomographicBin. nz
  concatenates the two populations across an empty gap, so integrating it directly
  is wrong; integrate the halves.
• b_interlopers and dm_mag_bias arguments and properties.

Performance

• Completeness tables are loaded and conditioned once per band and shared. Depth
  enters completeness only as m - m5_det, so every depth and cut of a band shares
  one table. TomographicBin construction: 13.4 ms → 1.2 ms.
• mag_bias is computed once per bin instead of on every access.