Build a good table

/extract never stores a template: it reads the table you select, each time. How you lay the table out decides what Moby matches, what it fills and what it leaves alone. This page gives the reading rules, the habits that make a run reliable and the mistakes seen most often.

How /extract reads a table

Element	Rule
Header rows	The first rows of the selection, top to bottom: category (matching only), column titles, optional description row. The data starts right under the last one
Description row	A row directly under the titles with hints ("in euros", "DD/MM/YYYY", "bottom right of the invoice") becomes part of the meaning of each column. A row of numbers or dates is data, never a description
Filled column	A matching key: Moby uses its values to find the document of each row. It is never written, even when partly blank
Empty column with a title	Filled from the document, one value per cell
Formula column	Skipped entirely
Column with no title and no data	Dropped
Merged label above columns	One document-type block (Invoice, Delivery note, Bank statement). A block holding only filled columns is the ledger block: an empty column inside it stays a key and is never filled
Same document type in two blocks	Read as one group when both blocks carry the same label. Two different labels for one document type mean two groups and two extractions
Mode	Keys and columns to fill: matching. No filled column: extraction, one new row per document. No column left to fill: Moby only matches and writes the Document column

The rules

For every table

#	Rule	Avoid	Do
R1	One value, not a presence	"Is there an invoice in this document?"	"Invoice number"
R2	A clear title, and a description under it only when the title alone is unclear	"Amount" alone: excl. VAT, incl. VAT or VAT?	"Amount incl. VAT", or "Amount" with the description "incl. VAT, in euros, bottom right of the document"
R3	One reading per title	"Reference": invoice, order or customer?	"Order number, format PO-XXXX, top right"
R4	The workflow extracts, it does not compute	Asking for the sum of the lines	Extract the values, then =SUM() in Excel. The agent can compute
R5	Paste only the useful rows and columns	The whole general ledger or trial balance	The rows and columns the check needs
R6	One document per PDF	A PDF bundling several documents, or one document spread over several files	/split a bundle first, /merge the pieces first

For a matching table

#	Rule	Why
K1	Every key gets an empty twin in the document block	Matching uses the filled keys and never writes them. You can only compare the ledger and the document when the document's value is extracted next to the key: "Amount" filled on the ledger side, "Amount (invoice)" empty on the invoice side
K2	One labelled block per document type	Without a label, the columns of a second document type are read as if they came from the first
K3	Never split one document type under two labels	Two groups mean every document is read twice
K4	Keys that identify: a reference first, then a distinctive amount, then the counterparty and a date	Moby tries identifiers first, then a distinctive amount, then the entity and its context

For an extraction table

#	Rule	Why
E1	Select every header row and keep the rows under the header free	Moby writes one row per document there without inserting rows. Only cells inside the selection are protected, so a description row left out of the selection is overwritten

Common mistakes

What you see	Why	Fix
One new row per document instead of your rows being filled	The table has no filled column, so Moby reads it as an extraction	Paste the rows to test with their ledger columns (reference, amount, date)
Rows under the header are overwritten	Extraction writes from the first row under the header, without inserting	Select the description row too, and clear or move what sits under the header
Nothing is filled, only the Document column	No empty column with a title	Add the columns to fill, empty, in the document block
The value Moby found is not on the sheet, so no check is possible	A key has no empty twin	Add an empty "Amount (invoice)" column in the document block
A column you wanted filled stays empty	It sits under the ledger label, or it has no title, or it holds formulas	Move it into a document block, give it a title, move the formula elsewhere
A column you wanted filled is treated as a key	It was already full	Clear it, or add a new empty column
The wrong amount or date comes back	The title can be read several ways	A precise title, or a description row
Values of a second document type look wrong	Several document types share one block, or one type sits under two labels	One merged label per document type, the same label for the same type
The run is refused	More than 30 columns or 400 rows	Split the table or the run
Columns or rows are ignored	The selection stops before the edge of the table	Select the whole table

Two layouts that work

Titles only, for extraction. Select the header rows.

	A	B	C	D
1	Invoice no.	Invoice date	Supplier	Amount incl. VAT
2	empty: one row per document from here

A ledger block and a document block, for matching. Select the whole table.

	A	B	C	D	E	F	G	H
1	Accounting (merged A1:D1)				Invoice (merged E1:H1)
2	Entry ref	Entry date	Customer	Amount excl. VAT	Invoice date	Customer	Amount excl. VAT	Invoice no.
3					on the invoice, DD/MM/YYYY	name printed on the invoice	excl. VAT, number only	top right
4	5351200093	02/12/2025	Spécialités Pharma	46,348

Columns A to D are filled: the keys. Columns E to H are empty: filled from the invoice. F and G are the twins of C and D, and H adds the identifier.

With several document types, add one block per type: for a bank reconciliation, the account columns as keys, then a block for the reconciliation statement, one for the bank statement and one for the bank confirmation. The tie-out checks are formulas, not extracted values.

Shortcuts

You want	Do this
A sheet already built for an audit test	/workpaper, tab Moby: ready-made sheets by cycle. See Workpaper library
A second opinion on your table before a run	Ask the agent: "check my table before extracting". It reports what the run should do and proposes edits; it changes the sheet only when you say which ones
To practise on sample documents	Your first dive