Tier-2 test evidence: budgets, timings and reference checks (WP45 item 25)
Tier-2 test evidence: budgets, timings and reference checks (WP45 item 25)¶
- [tested: tests/tier2/test_density_2wcd.py::test_ver49_2wcd_budget] The WP19/WP20 plan predicted stages 2 and 3 on the 2WCD structure at 0.05 nm to take at most 60 s and under 1.5 GB peak RSS (plan, Design §5). The budget test records the wall clock and
ru_maxrss(kB on Linux) in a fresh interpreter and never gates on them. Source:tests/tier2/test_density_2wcd.py(test_ver49_2wcd_budget). - [verified]1 Example 05's frozen solve on the ClyA reference mesh took 2,512 s on one core (WP16 D11 Outcome); its Tier-2 wrapper is
slowand records rather than gates, and the budget assumes about twice that [tested: tests/tier2/test_examples_05_clya_reference.py::test_ver46_the_frozen_case_solves_on_the_reference_mesh]. Source:tests/tier2/test_examples_05_clya_reference.py. - [tested: tests/tier2/test_examples_06_pdb_to_mesh.py::test_ver46_06_the_exported_map_overlays_the_aligned_structure] A PDB coordinate column is written to 0.001 Å, so a written atom lies at most half of that from the true one; the map bound of example 06 therefore allows
√3times that rounding. For a single Gaussian frame the map at distancedfrom an atom is at leastexp(−d²/(σR)²); the grid node nearest the written atom lies within√3h/2of it.Ris the smallest shipped radius of any heavy atom, which gives 0.928 ath = 0.05 nm(WP25 plan, Design §2), so a map in nm or with transposed axes fails. Source:tests/tier2/test_examples_06_pdb_to_mesh.py(test_ver46_06_the_exported_map_overlays_the_aligned_structure). - [tested: tests/tier2/test_examples_06_pdb_to_mesh.py::test_ver46_06_the_exported_map_overlays_the_aligned_structure] A residue table names some hydrogens without a leading
H(NAD'sNH2T), so a radius looked up by name filtered onHtakes a hydrogen's radius and makes a bound vacuous; the lookup goes through the set's own table. Source:test_examples_06_pdb_to_mesh.py(comment at the bound). - [verified]2 Example 07's ePNP-NS solve at
size_scale4 (chosen for runtime) takes a few minutes on one core (WP32 D9) [tested: tests/tier2/test_examples_07_pdb_to_charged_run.py::test_ver46_07_the_prepared_entry_walks_every_stage_to_a_charged_solve]. Source:tests/tier2/test_examples_07_pdb_to_charged_run.py.