[timholy]

noise_cover(V; shrink=0) returns scales a and b such that the data divided by a*b' has noise of unit average variance in every row and column, given a matrix V of noise-variance estimates. For Poisson counts, V = Y gives the biwhitening of Landa, Zhang, and Kluger (2022). The targets count the zero entries of V; zero rows and columns receive zero scale.

With shrink = 0, a max-flow test on the zero pattern of V certifies before solving that the scaling exists, and a SolverFailure names the rows or entries that prevent it. shrink = δ > 0 instead scales (V + δ*r*c'/s) / (1 + δ), which has the same total as V and always admits a scaling; the rank-one term is never formed as a dense matrix.

The products are computed by the PowerMean solver, which now accepts optional per-group targets and, for p = 1, an additive rank-one term. Without them, soft_cover and soft_symcover are unchanged. A new guide page describes how to build V for counts, camera data, and differences such as ΔF, and how to use the normalized matrix.

Assisted-by: Claude Opus 5.5 noreply@anthropic.com

[codecov]

Codecov Report

❌ Patch coverage is 98.82353% with 5 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.33%. Comparing base (dcaf2eb) to head (ca39345).

Files with missing lines	Patch %	Lines
src/powermean_newton.jl	95.65%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #107      +/-   ##
==========================================
+ Coverage   98.31%   98.33%   +0.02%     
==========================================
  Files          25       26       +1     
  Lines        7824     8222     +398     
==========================================
+ Hits         7692     8085     +393     
- Misses        132      137       +5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.