[antonsoo]

Landscape subtraction produces signed functions, but _p_norm can cancel their zero-crossing integrals or raise negative values to fractional powers. For critical points (0,0), (1,1), (2,-1), (3,1), (4,0), the L2 norm currently returns sqrt(2/3) instead of sqrt(4/3).

Integrate each segment from its endpoint magnitudes. This also avoids reconstructing heights from a potentially large coordinate intercept. Both exact and approximate landscapes use the correction. Add signed/fractional, translated-coordinate and diagram-difference regressions, plus a 0.4.1 patch release note.

Validation: 8 of the 12 new cases fail before; all 12 pass after. The complete suite on refreshed master passes 124 tests, with 19 warnings. Ruff format/lint, wheel/sdist builds and metadata checks pass.

An additional audit of 660 signed norms from recorded handwritten-digit diagrams corrects 143 errors, including 113 nonfinite results, against exact Fraction L2 integrals and adaptive quadrature for the other exponents. This reference integrates the returned piecewise-linear function; a separately observed exact-landscape construction discrepancy remains outside this norm correction.

[codecov]

Codecov Report

❌ Patch coverage is 80.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 79.28%. Comparing base (084c381) to head (d3b61b3).

Files with missing lines	Patch %	Lines
persim/_version.py	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #112      +/-   ##
==========================================
+ Coverage   78.87%   79.28%   +0.40%     
==========================================
  Files          20       20              
  Lines        1482     1477       -5     
  Branches      271      271              
==========================================
+ Hits         1169     1171       +2     
+ Misses        235      229       -6     
+ Partials       78       77       -1     

☔ 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.

[catanzaromj]

Hi @antonsoo, thanks for opening the PR and raising this! This is definitely a bug, as your example shows. Its interesting that the bug only shows when $p$ is even (so that $p+1$ is odd), since otherwise the discrepancy is cancelled away. I also appreciate you simplifying the logic around these calculations.

Can you bump the patch version and update RELEASES.txt to reflect this as well? Thanks

[antonsoo]

Bumped the patch version to 0.4.1 and added the entry to RELEASE.txt in d3b61b3, after rebasing onto current master. The full local suite passes, and all 660 recorded-diagram norm comparisons still agree with the independent integrals.