How it works · Wages.software
A calculation moves through explicit boundaries, never through a black box.
The sequence starts with a narrow request and ends with a result an authorized operator can inspect. Every stage either produces named evidence or refuses to continue; none of the stages files a return or releases funds.
The sequence
Scope, normalize, calculate, reconcile, then stop for a person.
First, scope the request. The published calculation routes require tenant context and a finance-role gate. The arithmetic then depends on the supplied earning lines, deductions, period and governing inputs. An unsigned exemption mask produces needs_sme; an unsigned federal table returns zero with needs_sme_unsigned_table. These distinct outcomes should not be collapsed into one generic failure.
Second, normalize without erasing meaning. Dollar inputs become integer cents, dates become explicit boundaries, and deductions retain their authority-specific treatment. The engine does not collapse several bases into one convenient total. That is how it can later explain why federal, Social Security, Medicare, state, or local figures differ.
Third, run small engines. Gross-to-net, taxable-basis construction, federal withholding, FICA, deductions, and reconciliation remain distinct decisions. Each returns its own reason and failure state. The separation makes a disputed number traceable and prevents a successful calculation in one area from concealing an unavailable table in another.
Fourth, compare totals. Quarterly and annual artifacts can be placed beside one another to expose a mismatch. The software does not invent a balancing adjustment. A discrepancy stays visible until an authorized person resolves the source data or the rule interpretation.
Finally, stop. A reviewed result can be handed to a system that actually owns payroll operations. Wages.software itself does not submit a file, create an ACH transfer, mark wages paid, or tell an employee that a deposit is on its way. That final stop is part of the design.
A useful first review
Prepare one synthetic case and identify the governing source.
A productive evaluation does not begin with every employee and every jurisdiction. It begins with one de-identified or synthetic case whose expected outcome is already understood: perhaps an invented deduction that reduces one basis but not another, or a reconciliation that should fail to agree. The manual review maps each input to a decision boundary, identifies the supplied table version, and compares the expected cents and reasons with the described result.
That review reveals whether the product fits before anyone discusses integration. It also keeps sensitive staff records out of an early conversation. If the case cannot be represented without inventing a rule or expanding the product into payroll operations, the correct outcome is to say so. Early access is a fit check, not a disguised implementation contract.
A small paper example keeps classification separate from subtraction.
Consider a fictional adult employer reviewing a statement with gross wages of 100,000 cents and one deduction of 3,000 cents. On paper, the reviewer labels two hypothetical authorities A and B. The invented, signed example mask reduces the basis for A and leaves B unchanged. This is an arithmetic exercise, not a statement about the legal treatment of any deduction. The expected bases are therefore 97,000 cents and 100,000 cents. A worksheet that shows 97,000 for both has already lost an important distinction before anyone discusses withholding.
The reviewer writes the classification assumption beside the numbers. That note matters more than naming a familiar benefit: a label cannot supply the governing treatment. If the example mask lacks the required sign-off, the taxable-basis decision is needs_sme. The missing authority is not repaired by the fact that subtraction is easy. The useful result of this stage is a checked relationship between gross, reduction and basis for each authority, together with the reason a result is held. None of those values establishes that a worker has been paid.
A returned zero needs its reason before it can be interpreted.
Now imagine the paper case moves to federal withholding with an unsigned bracket table. The published engine behavior is specific: it returns zero with needs_sme_unsigned_table. A reader who copies only the number into a summary can make that held calculation look like a legitimate zero liability. Keep the result and reason together. The calculation has not established what should be withheld. This distinction is especially useful in a demonstration, because a tidy numeric output can otherwise conceal the very missing evidence the exercise was supposed to expose.
A second reviewer should be able to identify the supplied period and table version without guessing from a filename. If the registry has no applicable entry, or selection is ambiguous, record that condition as the outcome of the review. Do not choose a nearby period merely to obtain a number. The registry is published as empty; this site supplies no statutory table to complete the example. A public explanation can show where a decision stops while leaving the actual rate and statutory interpretation with the responsible source and reviewer.
A disagreement stays visible through the handoff.
For a separate fictional reconciliation, put one annual total at 250,000 cents and its comparison total at 249,999 cents. The difference is one cent. Naming that difference does not explain its origin. A reviewer can check which artifact supplied each amount, whether the compared boxes mean the same thing, and whether an explicit adjustment belongs in the source material. The published reconciliation includes a named fractions-of-cents adjustment line; it does not justify inserting an unexplained balancing amount because the reviewer prefers a green result.
Write two separate conclusions. The first says what agrees or differs in the supplied artifacts. The second says what remains to be investigated. A manual note might read: annual comparison differs by one cent; source-line review remains open. It should not read: filing corrected. This exercise has submitted nothing and changed no source record. The distinction also prevents an otherwise useful comparison from taking responsibility for work owned by an external payroll operator, bank or filing process.
Finish with a reviewable explanation, not a paid status.
A compact evaluation packet can contain the synthetic inputs, the expected relationship, the exact returned reason and the question left for a person. Treat that packet as a document you author outside this website. There is no upload or saved evaluation workspace here. For the two-basis example, the question is whether the supplied classification was correct; for the unsigned-table example, it is which signed source actually governs. Combining those into a generic calculation failed message would make the next conversation less useful.
An evaluator may reasonably ask why a calculation product ends with so many boundaries. The answer is that a number and permission to act on it answer different questions. The published routes compute and return; the marketing page does not persist a pay run, release a transfer or file a return. A useful evaluation establishes the narrow decision the software can explain and the evidence it refuses to invent. It can end with an unresolved case. That is a concrete outcome, even when no operational rollout follows.
This walkthrough is a human review method. It offers no AI-generated calculation, no checkout and no payment action. The published disbursement status remains honest_off_no_partner in the wired routes. Reading a sequence or preparing a synthetic packet does not override that result, provision an account or authorize an external operator to move money.
Where to go next
The rest of this site.
Every page here is its own argument rather than a restatement of the home page. If you would rather ask a person, the address below reaches one.