Every receipt app in this category asks a self-employed person to review a year of expenses one at a time, which is why the shoebox stays in the truck until April. ReceiptInbox is a compressor: it groups near-identical records — the same normalised vendor, the same suggested Schedule C line, the same business-use band — and turns 412 expenses into 19 decisions, one of which can settle forty-seven receipts at once. The hero figure is not a balance, it is a countdown of work remaining, computed as decisions left × this person's own trailing median seconds per decision, so the two-minute promise is measured rather than claimed. One colour rule runs through every screen: mint means a person confirmed it, and nothing the app guessed is ever mint. Every suggestion is a dashed slate chip carrying the word Suggested and the evidence behind it — you filed Home Depot as Supplies 12 times out of 12 — and is excluded from every total until confirmed, which is how a records tool suggests a Schedule C line without ever appearing to rule on one. It runs no category model: the only evidence is this person's own prior decisions for this exact vendor. Receipt photographs are read into text and discarded in the same request, because the words are what the published guidance on adequate records actually asks for. Mileage is typed, in three numbers, and valued at the rate published for the period the trip falls in — 2026 changed mid-year, so a January drive and an August drive are not worth the same. The readiness meter counts what the person confirmed, never whether it is right, and says so on screen. No deduction-value figure, no tax-owed estimate, no audit-risk score. The plain expense export is free forever, and an expense recorded past the free cap is saved and held, never refused.