Start from the latest usable account snapshot
Shelter begins with connected balances and posted transactions. It does not invent a balance when the source is missing, stale, internally inconsistent, or in a different currency.
Methodology
How Shelter forecasts your next 30 days
Shelter combines dated commitments with a range of plausible everyday-spending paths. The goal is not to pretend the future is certain. It is to show the risky day, the money already spoken for, and how much uncertainty remains.
Method last reviewed August 23, 2026
Direct answer
What goes into a Shelter forecast?
The opening balance, expected income, recurring bills, debt payments, subscriptions, and observed discretionary spending. Known events keep their dates and exact amounts; uncertain spending becomes a range of complete 30-day paths.
Five stages
How is the forecast built?
Each stage has a narrow job and a fail-closed boundary.
Place known money on the calendar
Expected income, recurring bills, debt payments, and subscriptions are represented on the date they are expected to post. Exact stored amounts stay in minor currency units so repeated calculations do not introduce rounding drift.
Model uncertain everyday spending as paths
Everyday spending is not a single fixed number. Shelter resamples contiguous blocks of observed spending to create 200 reproducible paths through the forecast window. Keeping spending in blocks preserves the way real spending clusters across days.
Calculate the balance across every path
Shelter accumulates the opening balance, dated income, commitments, and discretionary spending day by day. It then reads lower, middle, and upper outcomes from complete balance paths rather than adding unrelated daily percentiles together.
Publish one versioned forecast
The result is stored with its source versions, calculation policy, currency, cutoff time, and checksum. Product surfaces read that published record instead of quietly rebuilding different numbers for each screen.
Fail closed
What keeps the forecast honest?
Unavailable is safer than a confident answer built from broken money data.
- A forecast covers 30 days; it does not imply certainty beyond that window.
- Missing or incompatible source data suppresses the answer instead of being converted to zero.
- Known bills are kept separate from estimated discretionary spending.
- A changed calibration invalidates the old published forecast until a rebuild lands.
- The same source snapshot and policy produce the same reproducible spending paths.
- Shelter is read-only. A recommendation cannot transfer money or execute a payment.
Limits
What can make a forecast wrong?
Unexpected purchases and transfers can change the path immediately.
Income can arrive late or for a different amount. A biller can move a posting date.
A disconnected institution can make the account picture incomplete.
Thirty days is a short-term planning window, not a promise about long-term solvency.
Common questions
Forecast methodology FAQ
Is the 30-day forecast a guarantee?
No. It is an estimate based on the connected data available at the forecast cutoff. New purchases, delayed income, changed bill dates, missing accounts, and bank-sync interruptions can change the result. Confirm the current balance with your bank before making a payment.
Why does Shelter show a range instead of one perfect number?
Bills and posted balances can be exact, but everyday spending is uncertain. Multiple whole-window spending paths make that uncertainty visible instead of hiding it inside one overconfident line.
What happens when the data is not trustworthy enough?
Shelter suppresses forecast-dependent money guidance and explains that the forecast is unavailable. Missing amounts, incompatible currencies, stale source versions, and failed reconciliation are not treated as zero.
Does Shelter learn from forecast accuracy?
Published forecasts are scored against later balance snapshots. A calibration can move only when a windowed, directionally consistent signal is strong enough, and changing it requires a new canonical forecast before guidance resumes.
See the method in context
Read the security boundaries, try the manual safe-before-payday worksheet, or see how the connected forecasting product applies the same timing logic automatically.