# Tier 2: gates that need the byte-pinned authority assets.
#
# Selector: -m assets
#
# About 6.9 GiB of files that gpuwm-hex ships with no fetch path.  They are
# pinned by byte count and SHA-256 in
# tools/run_cuda_v841_full_physics_x4.py::AUTHORITY_PINS -- masked over the
# netCDF file_id nonce for the three history files, since that nonce is random
# per write and would otherwise make every re-mint look like corruption.
#
# WHERE THEY LIVE (relative to the checkout root, wherever the assets are):
#   work/v841-vr-static/run-static-v841-conus-official-full-a/
#       x4.163842.grid.nc                 224,139,172 bytes
#       x4.163842.static.nc               298,860,376 bytes
#   work/v841-vr-static/run-real-init-v841-conus-official-full-a/
#       x4.163842.init.nc               1,489,665,020 bytes
#   work/v841-full-physics-gf-gwdo-native-authority-20260820a/native-run-a/
#       history.2026-08-10_12.00.00.nc  1,584,808,024 bytes   (F000)
#       history.2026-08-10_12.30.00.nc  1,584,808,024 bytes   (F030)
#       history.2026-08-10_13.00.00.nc  1,584,808,024 bytes   (F001)
#       native-gwdo-authority-receipt.json
#       native-gwdo-launch-receipt.json
#       run-closure.status
#
# The four members are the tests that turn a pin into a path and open it.
# tests/conftest.py detects them on `default_authority_paths` rather than
# trusting a marker: reading AUTHORITY_PINS itself is deliberately NOT
# detected, because the tests that assert on pin metadata are exactly the ones
# that should keep running on a box with no assets -- they are what proves the
# pins are well formed before anyone spends 6.9 GiB fetching files.

tests/test_cuda_v841_full_physics_x4.py
