The question
tests/test_hunt_probe_discipline.py::test_no_hunt_claims_the_reserved_word is the lexical guard that scans hunts/ to ensure probe files do not claim the certainty regime reserved for zeta/rigor.py and the Lean arm.
The attention queue identified the following known miss:
guard 'tests/test_hunt_probe_discipline.py::test_no_hunt_claims_the_reserved_word' does not detect: file types outside .py/.md/.json (a .txt overclaim passes)
This hunt measures:
- The exact file extension breakdown across
hunts/in the repository today. - The empirical detection boundary of the unmodified guard against a battery of 40 file formats across 12 distinct categories.
- Whether unscanned files under
hunts/currently contain the reserved word or related morphology.
Scope
Writes only inside hunts/r_365c6c/, hunts/README.md (for Hunt #51 case-log entry), harness/departments/guard_ledger.py (closing the loop), and telemetry. Runs the real guard source unmodified against a sandbox repository root in an isolated temporary directory. Nothing here is mathematics and nothing here is evidence for or against RH.
id: r_365c6c
question: Does test_no_hunt_claims_the_reserved_word detect file types outside .py/.md/.json?
frontier: the guard filters on path.suffix.lower() in {'.py', '.md', '.json'}; 76 of 447 files under hunts/ are currently outside that set
proposed_attack: run the unmodified guard source in an isolated sandbox repository root against a battery of 40 file specimens spanning scanned formats, plain text, structured config, scripts, code, markup, extensionless files, compound extensions, dotfiles, and binary formats
dead_routes:
- inspecting the guard source without executing it against planted specimens
- mutating the live repository tree during testing
required_oracles:
- pytest executing the unmodified guard source test_no_hunt_claims_the_reserved_word in a sandbox repository root
- exhaustive enumeration of file paths and extensions across hunts/
kill_conditions:
- the sandbox baseline with an empty specimen fails pytest, indicating sandbox contamination
- positive control files (.py, .md, .json) fail to trip the guard
- clean control files trip the guard
agents_may:
- read any file in the repository
- run the test suite
- execute the guard against an isolated sandbox repository root
- measure and record detection outcomes across all file types
agents_may_not:
- touch hunts/frontier_math/, meta/, lean/, or any root markdown file
- use the reserved word in probe files or hunt artifacts
- modify the guard logic in tests/ without authorization