Notes

How the thing is built, and why.

Working notes on retail software: what we decided, what we measured, and what we got wrong first.

One record, and why it matters
The usual retail stack is a POS, an inventory system, a payroll tool and a reporting layer, joined by nightly syncs. Every join is a place where two systems can disagree, and by the time you notice, the disagreement is a month old.
Architecture
How we test that a number is right
Every metric is checked against a value computed independently from raw transaction data — not against the product’s own aggregates, which would only prove it agrees with itself.
Correctness
What we will not claim
A model that reads your question and writes a number is a model that can be confidently wrong. Ours does not get to produce figures — only to choose which validated query to run.
Assist

One record, and why it matters

Revire writes a sale once. Stock, ledger, commission, loyalty and the customer record all move in the same transaction.

There is nothing to reconcile because there is no gap. It also means a question has exactly one place to look, which is what makes plain-language answers possible at all.

How we test that a number is right

Comparing a system against itself proves only that it is consistent, not that it is correct.

That discipline has caught real defects. Comparing totals rather than per-group values hid a refund-handling bug for a while: the totals matched because two errors cancelled. Comparing 4,348 product groups one at a time found it.

What we will not claim

Refusing is a feature.

A number you cannot check, on a page you cannot audit, is worse than no number. If Assist cannot map your question to something it can answer, it says so.

More notes as we write them. If there is something you want us to explain, ask — hello@revire.one.

Questions we have not answered?

Ask and we will write it up.

Email usBook a demo