corporate-sweep

# Corporate Sweep — the shrink half of the mechanism Spec SSoT: `Corporate.md` §7.4 (eviction procedure) and §7.5 (why a dynamic organization must automate this). This skill is the sweep; **disposal is the human's** (§12). **Why this exists.** Creation fires on a signal that occurs often — work arrives that no current seat fits. Dissolution fires on a signal that requires *looking back* — nothing happened here for a while. An unautomated backward-looking check does not run, so an organization with automated creation and manual eviction grows monotonically, and every surviving seat taxes every future decomposition. That is framework gravity applied to an org chart. ## 1. Gather ground truth - The roster `.agent/corporate.json` — seats, `createdFor`, `seatCeiling`, residency classes. - Per-seat evidence: delegations received, locker deliverables produced, practice entries attributable to the seat, budget consumed. **Counts come from executing a command, never from estimation** — a miscounted sweep argues for the wrong eviction. - The resource axis as it stands *now* (§4), which may have changed since the roster was written. ## 2. Evaluate the candidate tests (§7.4) Any one is sufficient to **raise** a candidate. None is sufficient to act. 1. **No practice** — zero practice entries attributable to the seat over N cycles (§8.3: one signal read in both directions — practice implies a seat, absence of practice implies its removal). **Two exclusions, or this test manufactures...