close-orchestratorlisted

# Close orchestrator A close is a long-running, interruptible, dependency-ordered process where **the ledger is the recovery map, not your memory**. This skill runs it that way. The failure mode this prevents: a close that runs for four days across many sessions, loses context, and gets re-done from a half-remembered state — producing duplicate entries, skipped reconciliations, and a completion claim nobody can verify. Load `finance-guardrails` first. Rails 1 (approval), 3 (as-of), 6 (run log), and 7 (say what you did not do) govern every step here. --- ## The ledger is the source of truth Before any close work, create or open the run directory: ``` close-runs/<period>/ LEDGER.md <- authoritative state; trust this over recollection runlog.md <- append-only, per guardrails Rail 6 tasks/<id>/ <- evidence per task: queries, extracts, tie-out blocks staged/ <- prepared-but-not-executed actions awaiting a human SUMMARY.md <- written only at the end ``` ### LEDGER.md format ```markdown # Close 2026-07 | opened 2026-08-01 | target close 2026-08-07 ## State current_phase: 2 (subledger close) blocked: none last_updated: 2026-08-19T14:22Z ## Tasks | id | task | phase | depends_on | status | evidence | owner | |----|------|-------|-----------|--------|----------|-------| | C01 | Cut off AP invoice entry | 1 | - | done | tasks/C01/ | AP | | C02 | Bank rec - operating | 2 | C01 | done | tasks/C02/ tie-out PASS | me | | C03 |