The question
tests/test_hunt_probe_discipline.py::test_no_hunt_claims_the_reserved_word is the guard that keeps hunts/ from claiming the word zeta/rigor.py owns. The attention item that opened this hunt asserted a gap:
the guard does not detect synonyms — verified, confirmed, definitively, proves are separately banned but by other checks; a hunt overclaiming in fresh vocabulary passes this guard
Two claims are bundled there, and they have different truth values. The first is about the guard's reach. The second — "by other checks" — is about whether enforcement exists somewhere. This hunt measures both, plus the question neither of them asks: whether the guard is even airtight on the one word it does own.
Scope
Writes only inside hunts/r_03a798/. Runs the real guard source unmodified against a sandbox repo root in a tempdir; plants nothing in the live tree. Nothing here is mathematics and nothing here is evidence for or against RH.
Prior art
hunts/wide_search/HANDOFF.md §"the lexical ban" already records, as a session note, that verified / confirmed / definitively / proves are "banned by documentation but not by a test". That hunt got there first and this one does not claim the observation. What is added here is the measurement: which checks exist, what the guard's own boundary is on the word it does own, and whether the documented-only ban is already being crossed in the live tree.
id: r_03a798
question: Does the hunts/ reserved-word guard detect overclaim synonyms, and do the "other checks" said to ban them exist?
frontier: one whole-tree lexical guard exists (test_no_hunt_claims_the_reserved_word); the four synonym bans are stated in hunts/README.md and CLAUDE.md with no located enforcement
proposed_attack: run the unmodified guard source in a sandbox repo root against a specimen battery spanning reserved word, its morphology, typographic evasions, the four documented bans, fresh overclaim vocabulary, and sanctioned vocabulary
dead_routes:
- grepping the guard source and reasoning about what it would match, without running it, which cannot see collection or skip rules
- planting specimens in the live hunts/ tree, which mutates a shared area other sessions may be running against
required_oracles:
- pytest executing the unmodified guard source, exit status as the verdict
- exhaustive enumeration of tests/*.py that both address hunts/ and read file text
- literal substring occurrence counts over the committed hunts/ tree
kill_conditions:
- the sandbox baseline with an empty specimen is not green, so per-specimen verdicts measure the sandbox rather than the specimen
- a test enforcing any of the four documented bans is located, making the premise's "other checks" real
- sanctioned vocabulary trips the guard, making the guard's failures noise rather than gaps
agents_may:
- read any file in the repository
- run the existing test suite
- build a sandbox repo root in a tempdir and run the real guard against it
- report a gap and name the smallest check that would close it
agents_may_not:
- modify tests/, zeta/, harness/, meta/, lean/ or any root markdown file
- modify hunts/README.md, including to add this hunt's case-log entry
- write a new guard into the suite on its own authority
- describe a documented-only rule as enforced, or a measured gap as a proven absence