Skip to content

Tier-1 test evidence: stores, paths, OCC names, NaN and reduction tolerances (WP45 item 23)

Tier-1 test evidence: stores, paths, OCC names, NaN and reduction tolerances (WP45 item 23)

  • [tested: tests/tier1/test_run.py::test_ver32_a_run_removes_its_scratch_once_the_store_holds_the_payloads] Store.put copies every payload, so a run's scratch workspace would hold a second copy of each. The driver removes the scratch it made once the store holds the payloads and leaves a caller-named workspace alone. Source: tests/tier1/test_run.py (test_ver32_a_run_removes_its_scratch_once_the_store_holds_the_payloads).
  • repr doubles every backslash, so a message that quotes a Windows path with repr names C:\\Users\\... where the file is C:\Users\.... A POSIX backslash is a legal file-name character, so a diagnostic check must drive a separator on either platform. Source: tests/tier1/test_reproduce_diagnostics.py.
  • [tested: tests/tier1/test_primitives.py::test_name_edges_leaves_an_already_named_edge_alone] A boolean result of OCC is a compound and refuses a name query, so an edge's name is read off its one face. Source: tests/tier1/test_primitives.py (test_name_edges_leaves_an_already_named_edge_alone).
  • [tested: tests/tier1/test_nan_gates.py::test_ver70_a_nan_permissive_gate_is_found_at_its_line] Every ordering comparison with NaN is false, so if value > tol: raise is skipped by a NaN; the house form if not value <= tol: raise fails it. Source: tests/tier1/test_nan_gates.py.
  • The sub-agent frontmatter keys that code.claude.com documented on 12 October 2026 are the set that the files of .claude/agents/ were written against; the test that checked them, test_plan_shape.py, was deleted by the test scope (WP50, §8.2.13 N3).
  • [tested: tests/tier1/test_reduction.py::test_ver50_off_axis_gaussian_closed_forms] The VER-50 tolerances rest on values the WP19 plan recorded when it was written: the off-axis Gaussian's Cn variance is 1.2 × 10⁻⁵ (limit 3 × 10⁻⁵) and its raw variance 3.2 × 10⁻⁴ (limit 10⁻³), with the Cn variance 0.8 % of its peak; the cos(mθ) mean is 2.5 × 10⁻⁴ of its peak [tested: tests/tier1/test_reduction.py::test_ver50_cos_modulation]. Source: tests/tier1/test_reduction.py (test_ver50_off_axis_gaussian_closed_forms, test_ver50_cos_modulation).