1.5: fix Size_Inner_Difference in Delboeuf/Ebbinghaus, repair the README workflow

[DominiqueMakowski]

Bugfix release. Two independent fixes, one commit each.

Size_Inner_Difference was not a difference of areas

Delboeuf and Ebbinghaus both reported a wrong value from get_parameters():

np.pi * (size_bigger / 2) ** 2 / np.pi * (size_min / 2) ** 2

Since * and / associate left to right this evaluates as (pi * a**2 / pi) * b**2 — the product
of the two squared radii, which is not an area difference and does not even have the units of one. It
returned 0.000244140625 for two circles of identical size.

It is now the signed difference between the areas of the two inner circles, positive when the left one
is larger — which is what the docstring in both classes always described. Signed rather than absolute
follows that wording and matches Angle_Difference in RodFrame, the only other difference-style key
in the package. (An area ratio was the other candidate reading, but it works out to exactly
1 + abs(difference), so it would just restate Difference.)

Ebbinghaus was affected because it reuses _delboeuf_parameters_sizeinner() and never overwrites the
key. A scan of every other *_parameters.py found no further instances of this pattern — these two
were the only ones.

Also fixes the entry's name in both class docstrings, where it was listed as Sine_Inner_Difference.

Covered by a new regression test over both illusions, checked to fail against the old expression.

⚠️ This changes user-facing output. Anything that logged Size_Inner_Difference will report
different numbers. The old value was never meaningful, so nothing correct can have depended on it, but
it may appear in existing analysis columns.

The Render README workflow could not run

It failed at Install Python with a bundled-PAT rate-limit error from remotes::install_github().
rstudio/reticulate is alive and the URL in the error resolves fine, so this was never a moved
repository — an unauthenticated install_github() just gets rate limited on shared runners. reticulate
now comes from CRAN; the development version was not needed.

Two further blockers sat behind that one, which the run never got far enough to hit:

• README.Rmd imports pyllusion in 26 chunks, but the workflow installed only Pillow and scipy. It now
  pip-installs the package itself.
• README.Rmd called reticulate::use_python() on a hard-coded Windows path belonging to one
  contributor's machine. That call is now skipped when CI is set or the path is missing, and the
  workflow points reticulate at its own interpreter via RETICULATE_PYTHON.

Plus: install_miniconda() replaced with actions/setup-python, unused dplyr dropped, job moved to
ubuntu-latest with public RSPM, contents: write granted for the commit step, an explicit push
refspec instead of a bare git push origin, and regenerated figures in docs/img committed alongside
README.md.

Not verified end to end — it only runs on a push that touches README.Rmd, and there is no R in the
environment this was written in, so the render itself is untested.

Note on releasing

1.4 is already on PyPI, so this bumps __version__ to 1.5. Since the publish workflow runs on every
push to master with skip-existing, merging this will publish 1.5 — worth holding until you want
that, or dropping the version bump and doing it separately.

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

The workflow must stop masking git push failures before approval.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Do not mask git push failures as success · New

What changed in this PR

Bugfix release correcting inner-circle area differences for Delboeuf/Ebbinghaus and repairing the README rendering workflow.

Changes:

• Fixes signed area-difference calculations and documentation.
• Adds regression coverage and bumps the version to 1.5.
• Updates CI dependency installation and Python configuration.
• Critical finding: the workflow masks git push failures with || echo.

File	Description
tests/​__init__.py	Adds regression coverage.
README.Rmd	Makes Python selection CI-safe.
pyllusion/​Ebbinghaus/​Ebbinghaus.py	Updates parameter documentation.
pyllusion/​Delboeuf/​Delboeuf.py	Updates parameter documentation.
pyllusion/​Delboeuf/​delboeuf_parameters.py	Corrects area-difference calculation.
pyllusion/​__init__.py	Bumps package version to 1.5.
NEWS.rst	Documents the fixes and release changes.
.github/​workflows/​readme.yml	Reworks the README rendering workflow.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.