/**
 * Estate-sensor tests (mmnto-ai/totem#2580 slice-1). One `describe` per
 * invariant in the design's "Invariants to lock in via tests" list — the
 * numbering is load-bearing: these are the claims the sensor is allowed to
 * make, and each one is pinned by the test that names it.
 *
 * Git is never invoked: every fixture injects an exec spy that answers canned
 * porcelain, so the tests assert the CLASSIFIER, not the local git install.
 * The filesystem side (husk sweep) uses real temp trees, because the evidence
 * classes are literally fs shapes.
 *
 * Known untested arm: the mid-sweep TOCTOU path in `classifyHusk` (a directory
 * that readdir named and that vanishes before the `.git` probe, recorded as
 * `vanished or turned unreadable mid-sweep`). Reaching it needs a race between
 * two fs syscalls, which is not portably schedulable without an fs seam the
 * scan does not have. Recorded as a gap rather than covered by a test that
 * would assert nothing.
 */
export {};
//# sourceMappingURL=estate-scan.test.d.ts.map