Reproduce census test boundaries under declared dependencies: REPL EOF and Colors isolation #34

Owning workstream

Follow-up from 4444J99/laurea#16 source census; source snapshot 4e8a94450ce1ce12d0fb0acbb16215600f854db8. This is an execution-environment/test-isolation investigation, not a claim of a confirmed production defect.

What was executed

A local Python 3.13 / pytest 9.0.2 / click 8.1.8 environment enumerated the snapshot and ran tests without connected credentials or paid services. The snapshot declares click >=8.4.2 and pytest >=9.1.1, so this was not the supported dependency environment.

• Full suite did not finish in a bounded run. An unprivileged retry reached rege/tests/test_cli_coverage.py::TestREPLExtendedCoverage::test_repl_ctrl_c_interrupt and timed out on the REPL/EOF path.
• Root execution initially produced a chmod permission-test failure; the unprivileged run avoided that environment-specific obstacle.
• A recorded 52-file non-CLI selection completed: 1,770 passed, 2 failed. The two failures were test_formatting.py::TestColors::test_colors_defined and test_bloom_misc_final_coverage.py::TestFormattingNoColor::test_no_color_env_triggers_disable. Observed Colors class attributes were empty, consistent with shared state surviving an earlier disabling test. This needs reproduction in canonical and permuted order before attribution.
• Static enumeration found 1,998 test-function definitions. This is not reported as 1,998 passing tests.

Exact next action and acceptance

1. Reproduce the pinned snapshot in a disposable, unprivileged environment with its declared dependency versions; record resolved package versions.
2. Run the REPL EOF/interrupt test with an external bounded timeout. Verify the CLI exits or raises the intended condition rather than silently looping.
3. Reproduce the two Colors tests individually and after tests that disable colors; make fixture/module-state isolation deterministic only if the defect persists.
4. Run the complete test suite without omitting failing tests; save JUnit, skips, errors and exact source head. Distinguish environment-only resolution from a repository code fix.
5. Return the actual execution result to the LAVREA evidence record; no unearned green or new passing-test headline.

The selected-file manifest and normalized log are being retained in LAVREA's September 24 source-census evidence and downloadable verification package. No live RE:GE source was changed by the census. No paid model calls, production deployment, credentials, or coding-agent review invocation is authorized by this issue itself.