Rutba Office 1.51 reads receipts and invoices, and trusts a figure only when the sums agree

Rutba Office learnt to read printed words on your own computer in 1.47. In 1.50 its engine learnt to read a page as a document: a bill’s supplier, number, dates, subtotal, tax and total, and a form’s tick boxes. Version 1.51, released on 15 October, puts that to work in three apps, in English, Urdu, Arabic, Hindi, Chinese and Japanese, with nothing uploaded.

Worksheets: a row from each receipt

Data, From Picture, Receipt or Invoice takes one picture or many, or the pages of a PDF, and shows each as a row to check: date, supplier, number, subtotal, tax, total and currency, beside the page with the box each value was read from. Insert puts the rows in at the selected cell, dates as dates and amounts as numbers, as one step Undo takes back. A shoebox of receipts becomes an expense sheet without retyping it.

The sums decide what to trust

A receipt checks itself: the items add up to the subtotal, and the subtotal and the tax add up to the total. The suite’s documentation measured this on 90 pages drawn in the six languages, clean, scanned and photographed. Of the 66 pages with figures, 34 had sums that agreed, and on every one of them the subtotal, the tax and the total were all right.

So a page whose sums agree goes in as read, and the rest are marked for a person to look at rather than dropped: of the 32 pages whose sums did not agree, 17 still had the right total. The usual faults are the ones a person makes too, a lost decimal point or a rupee sign read as a 7, and the sums catch both.

Mail and PDF

• Mail shows an attached invoice’s details. Open a message with a PDF or picture attached that reads as an invoice or a receipt, and a note says who it is from, its total, its due date and its number, each pressed to show where on the attachment it was read. It is read only when the message is opened, on threads of its own, and kept with the message so it is not read twice.
• PDF finds fields on a scan. Recognise Text can list a scanned invoice’s details in a Fields pane, each marking its place on the page, with Copy as a Row for a sheet. On a scanned form it finds the tick boxes by the words beside them, and Make Check Box Fields turns them into real check boxes, ticked as they were on paper.

What it cannot read yet

The engine does not read Arabic-Indic or Urdu digits: ٧ comes back as a V. A receipt printed in them gives no figure as sure, and says so rather than guess. Urdu set in Nastaliq stays a beta, lines of items are not yet offered as a table (one page in two was read right), and the measurements are on pages the suite drew, not on real till rolls. Those are the next to measure, and the documentation will publish them the same way.

Also in 1.51

Calendar’s month shows as many of a day’s events as fit again, with a list of the rest a click away, after 1.49’s visual pass had left one a day. Documents keeps every button’s name on its Home tab at 1,300 pixels wide. And office.rutba.io has a new Tour: 34 pictures of the suite’s features, taken by the suite itself on invented samples.

Before it went out, the release ran 2,576 engine tests with none failing and passed all 1,783 checks that drive the real windows, and the file fuzzer damaged 5,400 files and 1,800 PDFs without a fault. The next version is numbered 2.0.