harness/departments/guard_ledger.py carries the guard scripts/make_context.py --check with fired=None, known_misses=() and scope="undetermined until demonstrated". The ledger's own rule is that an outcome costs a demonstration, so the entry is currently a worklist item, not a claim — tests/test_guard_ledger.py::test_the_worklist_surfaces_the_undemonstrated_entries asserts exactly that it is still open.
The ledger names the smallest mutant: one added public symbol with CONTEXT.md left unregenerated. This hunt builds it, runs it past the guard, and records the outcome — plus enough siblings to say where the guard's sensitivity stops.
What this hunt may touch
hunts/r_414eed/ only. In particular it does not edit the guard ledger, tests/test_guard_ledger.py, or hunts/README.md; the mutants are applied to a throwaway copy of the tree in /tmp, never to the repository. Consequences for the ledger are reported, not applied — see HANDBACK.json.
Method in one paragraph
scripts/make_context.py is a pure-ast, stdlib-only script whose ROOT is derived from __file__, so it can be pointed at a copy of the tree. probe.py copies the seven directories build() reads plus CONTEXT.md into a temporary sandbox, checks that the copy reproduces the committed CONTEXT.md byte for byte (control 1), applies one mutant, runs the guard, undoes the mutant, and checks the guard is quiet again before the next one (control 2). Beyond the exit code it also regenerates CONTEXT.md and diffs it, so a catch can be attributed to the mutant's symbol reaching the index rather than merely to the per-module line count moving.
Nothing here bears on RH
This hunt measures a repository hygiene script. It touches no mathematics and is not evidence for or against anything about ζ (docs/08).
id: r_414eed
question: Does scripts/make_context.py --check detect its declared smallest mutant, and where does its sensitivity stop?
frontier: guard ledger entry records fired=None, known_misses=(), scope "undetermined until demonstrated" — zero mutants ever run
proposed_attack: apply 24 smallest-edit mutants to a throwaway copy of the tree and read the guard's exit code, with a CONTEXT.md diff to attribute each catch to a mechanism
dead_routes:
- editing the live tree to test the guard, which would make the mutation indistinguishable from ordinary work in progress
- running the guard without a byte-exact baseline control, which would make every catch an artefact of the copy
required_oracles:
- the guard's own process exit code under subprocess
- byte comparison of the regenerated CONTEXT.md against the committed one
- a same-sandbox undo check after every mutant
kill_conditions:
- the unmutated sandbox fails the guard, so no catch can be attributed to a mutant
- an undo leaves the sandbox dirty, so rows are contaminated by their predecessors
- the declared smallest mutant passes the guard, in which case the ledger entry becomes fired=False and the guard is dead
agents_may:
- copy the tree into a sandbox and mutate the copy
- run the guard and record exit codes and diffs
- report a proposed ledger amendment
agents_may_not:
- edit harness/departments/guard_ledger.py or tests/test_guard_ledger.py
- record an outcome in the guard ledger on their own authority
- promote a boundary miss to a defect without stating the edit that reaches it