Skip to content

Month-End Close

Finance month-end-close Source

Install: claude /plugin install finance-skills

Portability: 3 stdlib scripts, no network, no ERP connection. Export CSV or JSON; the scripts plan, prove and gate. Not accounting or audit advice. Thresholds and aging limits are your policy; the scripts make you state them.

A slow close is rarely slow everywhere. One chain of dependent tasks sets the close date, reconciliations "agree" without a proof, and flux review accepts "timing" as a reason. Each script attacks one of those.

Workflow

# 1 - Calendar: dependency order, critical path, owners, reviewers, load.
python scripts/close_calendar_planner.py --plan close_plan.json
# 2 - Reconciliations: match, prove, age. One run per account.
python scripts/reconciliation_matcher.py --gl gl_cash.csv --source bank.csv \
  --gl-balance 52708.60 --source-balance 47144.72 --as-of 2026-10-31 --preparer Sam --reviewer Lee
# 3 - Flux: every material change needs a quantified, evidenced explanation.
python scripts/flux_analyzer.py --input tb_compare.csv --pct 10 --abs 10000 --reviewer Lee

Step 1 schedules each task as soon as its predecessors finish, then checks the plan against people: one person booked for 1.5 days on WD1 is a plan that will slip. Speed up the critical path, nothing else.

Step 2 matches by reference, then by amount within a date window, then pairs digit transpositions (7,980 vs 7,890) into one correcting entry. What is left is a reconciling item: in the GL only (timing) or in the source only (book a journal entry). Opening-balance rows are balances, not transactions: when the two sides open at different amounts, that difference came from last month and blocks sign-off until it is traced there; it is never booked this period. The proof must close to zero.

Step 3 marks an account material when it passes your threshold rule, flips sign, or appears or disappears. Its explanation fails if it is missing, stock ("timing", "in line with expectations"), or has no number in it.

The gates block on (exit 4)

  1. A task with no owner, or a key control with no reviewer or reviewed by its preparer.
  2. A plan whose critical path ends after the target workday.
  3. A reconciliation that does not prove, or an item older than the stale limit (default 90 days, policy).
  4. Preparer and reviewer being the same person, or a flux reviewer who owns the line.
  5. A material fluctuation with a missing, boilerplate or unquantified explanation.

Hard rules

  • A sign-off is not evidence of review. Record what was checked and where the support is.
  • Never plug a reconciliation. An unexplained difference is a finding until someone explains it.
  • Thresholds are policy, not standards. No auditing standard prescribes "10% and 10,000". State yours.
  • Explanations name the driver and the amount: what moved, by how much, why, and the document behind it.
  • Do not speed up off-critical-path tasks to shorten the close; they have slack.

Scripts

Script Role Exit codes
scripts/close_calendar_planner.py Topological order, critical path, slack, WD deadlines, calendar dates, owner load, reviewer segregation 0 · 2 at risk · 3 bad input or cycle · 4 blocked
scripts/reconciliation_matcher.py Two-pass matching, proof, aging buckets, journal entries needed, review check 0 · 2 items · 3 · 4 unreconciled
scripts/flux_analyzer.py Materiality by your rule, sign flips, explanation lint, evidence check, self-review 0 · 2 evidence · 3 · 4 incomplete

All support --sample and --output json.

References and assets