You asked whether we actually have processes that notice "this account hit its limit — time to move money." We didn't — the targets existed on paper, scattered and partly contradictory. Now we do: a single canonical rulebook, an automated first-of-the-month sweep, and a tested pipeline that hands you the exact transfers to make.
The full spec lives in docs/money-map.md in the finances repo — the single source of truth. Every other document now points there instead of restating numbers.
| Account | Role | Target | Overflow goes to… |
|---|---|---|---|
| Business checkingChase PERFBUS | S-corp operating float — ~2 months of payroll + overhead | cap $40k | Cash Plus, monthly, booked as an owner distribution |
| Personal checkingChase | Paycheck landing pad, kept deliberately thin | ~$4k | Cash Plus, monthly |
| Cash PlusVanguard — the hub | All card autopays pull from here. $115k liquidity floor (~5 months of burn) + $30k tax reserve | $145k | Brokerage — invested, not parked |
| BrokerageVanguard, core index | Absorbs everything above the targets: 80% VTI / 20% VXUS, dividends reinvested | no cap | — (buy and hold) |
Untouched by the sweep: the 529 (auto glide path), SEP IRA, BTC IRA, Caroline's brokerage, and the HSA. The ~$15k UI position rides along as the play bucket under the 5% rule. One open variable is flagged in the doc: if the Gusto withholding bump replaces quarterly estimates this fall, the $30k reserve shrinks and the Cash Plus target drops toward $115k.
The math, in order:
The same run also verifies every card's autopay actually executed, scans for anomalies and dead account feeds, logs a net-worth snapshot to a running ledger, and — on quarterly runs — nags about the manual balances that go stale (home value, BTC IRA) plus the upcoming estimated-tax date.
I ran the full pipeline end-to-end in dry-run mode before trusting it. It worked, and it caught a genuinely dangerous edge case:
Tiller's feed hadn't yet absorbed yesterday's restructure — the $18.5k Cash Plus → brokerage transfer and the VTI/VXUS buys. The raw algorithm therefore said "invest another $20,100," which would have moved the same money twice.
The agent cross-checked recent activity, spotted the in-flight transfer, adjusted the balance, and correctly skipped the invest leg. That safeguard is now a codified rule in the sweep prompt — the in-flight guard — not a judgment call.
One move only: skim ~$1,599 from personal checking to Cash Plus. The business account is under its cap, and Cash Plus — after the restructure — sits almost exactly on its $145k target. The plan landed right on the number.
Your instinct was right that the rules were smeared across files with contradictions. Fixed:
money-map.md owns every target, threshold, and the algorithm. Repo instructions forbid restating its numbers elsewhere.f87b90b).