[sriramdevadas]

The Problem

On main, describe_grid gives the page a deck's track grid in the deck's own frame, where marks.js draws it, but takes the marks' length from get_absolute_size_y(), the depth of the deck's box in the world, and measures a seated carrier the same way. A STARlet turned 90 degrees ends its marks 1005 mm back on a deck 653.5 mm deep, and 198 mm back beside a carrier that reaches 560 mm.

PR Content

• describe_grid measures both in the deck's own frame: the deck with get_size_y(), a seated carrier from its corners with get_location_wrt.

Tests: grids_tests.py turns a STARlet 90 and 37.5 degrees, bare and with a carrier, and checks it keeps its upright grid. All four cases fail on main. Compared with main, upright grids of the STAR-family, Vantage and Nimbus decks are unchanged. make test, make lint, make format-check and make typecheck pass on Linux, and Unit Tests passes on my fork, 18 of 18 jobs (run).

Before and after, two turned STARlets from above:

🤖 Generated with Claude Code

[sriramdevadas]

Tests (usb, py3.12) failed on pylabrobot/formulatrix/mantis/driver/oem_config_tests.py::UnitBlockSourceTests::test_an_older_declared_file_takes_what_it_lacks_from_the_folder: it counted 2 warnings where it expects 1. The same test passed in the other 17 jobs of this run, and in all 18 jobs on my fork at the same commit. It imports nothing this PR changes.