fix(twin): read a bus's GeoJSON coordinates as [longitude, latitude]

[nilsonfh]

What

bd syntgrid-06l. Two readers parsed a bus's GeoJSON geo value as [latitude, longitude], so every freshly exported twin transposed every bus.

GeoJSON Point coordinates are [longitude, latitude] (RFC 7946), and that is what pandapower writes into net.bus["geo"] from the builder's x = longitude / y = latitude node attributes (PandapowerGridBuilder.create_bus_geodata, and gridalyn/twin/core/graph.py documents the convention).

• gridalyn/twin/adapters/network.py:_coords_from_geo — the base export.
• gridalyn/projects/workflows/scripts/run_digital_twin_ev_powerflow.py:_coords_from_geo — the per-scenario node artifacts the dashboard draws.

How it was found, and why nothing caught it

Rebuilding the Trois-Rivières twin to look at it in the dashboard: gridalyn twin base wrote grid_buses.parquet with lat in [-72.6204, -72.5878] and lon in [46.3315, 46.3544]. The dashboard catalog then reported extent.bbox [-72.62, -72.62, 46.35, 46.35] and centre {lat: -13.133, lon: -13.133}, and viewState.js frames the map from that centre — so the map could not show the network at all.

The committed instances/default/digital_twin/base/grid_buses.parquet (2026-08-12, same adapter_id) has the columns the right way round, so this is a regression against the repo's own artifact, not the original behaviour. It stayed invisible because the committed base predates the pandapower geo column, both fields are floats, and nothing under tests/ referenced either helper.

Verification

• tests/test_network_adapters.py::GeoJsonCoordinateOrderTest, four tests: what pandapower actually writes (the premise, independent of our readers), the exported bus table, a Québec-range bound that a transposition cannot satisfy whatever the columns are named, and the second reader.
• Failing-first, measured: re-introducing the exact defect fails 3 of the 4 (the premise test still passes, as it must).
• 819 passed, 1 skipped across the twin/adapter/catalog/repository/publication/hygiene/public-API/layer-direction files.
• Full suite: 2211 passed, 76 skipped (the operator-verified heavy studies, whose outputs are gitignored).
• mypy: gridalyn 116/116, twin 12/12. black, isort, flake8 clean.

Not in this PR

Artifacts already written with the columns transposed are not migrated: outside this repo nothing tracked carries them, and the committed base predates the defect. The issue records that decision.

🤖 Generated with Claude Code