Dictionary

The Dictionary tab holds the diagram's naming standard in three dictionaries, each on a sheet you edit like a spreadsheet:

• Domains: named data types, such as Name for VARCHAR(100). A column linked to a domain holds its data type.
• Words: the standard words logical names are made of, each with the abbreviation a physical name uses for it, such as Customer and CUST.
• Terms: whole logical names, each with the physical name and the domain it gives, such as Customer Name with CUST_NM.

A table's or a column's logical name derives its physical name from these dictionaries, and a column's term can also link it to a domain — see Deriving Physical Names.
A fourth sheet, Checks, lists what the dictionaries find wrong in the diagram's names and types and in the dictionaries themselves, such as a forbidden word in a logical name — see The Checks Sheet.

Opening the Tab

Switch to the Dictionary tab in the toolbar, the button with a book icon between Code Generator and Settings, or choose Dictionary under Tab in Quick Search.
It can be the tab the diagram opens on — see Canvas Type in Lock.

The tab opens on the Domains sheet until you pick another.
Each time it opens, the sheet comes back with the selection and the scroll you left, or with its first cell selected on a first visit. Its grid takes the keyboard, so typing or the arrow keys act at once.

The sheet shown, and each sheet's search, selection, scroll, and column widths, are yours alone.
None of it is saved in the file, recorded in the undo history, or shared with collaborators. It is kept while the editor stays open, across tab switches and across documents loaded into the same editor.

The Sheets

The bar across the top holds the four sheet buttons, Domains, Words, Terms, and Checks, and the button of the sheet shown looks pressed. Each shows a count beside its name: the rows its sheet holds, whatever the search shows, or on Checks the number of checks, colored by their level — see The Checks Sheet.
The search field comes next — see Searching. At the bar's end, on Domains, Words, and Terms and not in a read-only editor, come four tools: Add Row, Delete Rows, Import CSV, and Export CSV — see Adding and Deleting Rows and Importing and Exporting CSV.
In a narrow pane or a long display language, the bar wraps onto more lines, so every button stays in reach.

Under the grid, a status line counts the rows the search leaves, as in 4 rows, and, while the selection covers any row, how many it covers, as in 1 selected. A row two ranges share counts once, so that number is how many rows Delete Rows would remove.
Row numbers down the left count the rows as the sheet shows them, from 1. A read-only cell, which is every cell on Checks and every cell in a read-only editor, is drawn on a gray fill, except where a check marks it.
An empty Domains, Words, or Terms sheet reads Import a CSV file or add a row, except in a read-only editor.

Each sheet's columns, in order:

Sheet	Columns
Domains	Domain, Category, Data Type, Description
Words	Word, Abbreviation, English Name, Description, Synonyms, Forbidden Words
Terms	Term, Physical Name, Domain, Description, Synonyms
Checks	Type, Target, Detail

What the cells do:

• A domain's Data Type is written as a column type, such as VARCHAR(100) or NUMERIC(15,2), and every column linked to the domain holds it — see Domains and Data Types.
• A word's Abbreviation is what a physical name uses for the word — see From the Words.
• A word's Synonyms and Forbidden Words list other words for the same thing, separated by commas. Each item is trimmed, an empty item is dropped, and an item equal to the word itself is ignored. A logical name that uses one composes no physical name, and the Checks sheet flags it and names the word to use instead — see What Each Check Finds.
• A term's Term is a whole logical name, and its Physical Name is the name that logical name gives — see From a Term.
• A term's Domain shows the name of the domain it links, and is empty while it links none or links a domain since removed — see A Term's Domain.
• Category, English Name, each Description, and a term's Synonyms are notes. They change nothing, and no check reads them.

In a right-to-left display language, the bar, the grid's menu, and the status line are mirrored, while the sheets and a term's domain list stay left to right.

Deriving Physical Names

The dictionaries can give a table or a column its physical name, the name Schema SQL and the Code Generator write, from its logical name. A term whose Term is the logical name is looked for first; otherwise the name is composed from the words' abbreviations.
For example, with the words Customer (CUST), Name (NM), Number (NO), Order (ORD), and Date (DT), and the term Customer Number with the physical name CUST_NO and the domain Number, whose data type is INT:

• The logical name Customer Number names a column CUST_NO, links it to Number, and gives it the type INT.
• Order Date composes ORD_DT.
• A table with the logical name Customer is named CUST.

With an empty Words sheet and an empty Terms sheet, nothing is derived.

When a Name Is Derived

A physical name is derived as a logical name is written:

• On the canvas, when the logical name cell's editor closes on a logical name other than the one it opened with, by Enter, Tab, Escape, or a click elsewhere. Each key you type writes only the logical name — see Writing a Logical Name.
• At once, when you pick a term from the list under the cell — see Completing Term Names.
• When a paste writes a logical name cell — see Copying/Pasting Columns.
• When Find and Replace replaces a logical name, one match or all of them.
• When a coding agent sets a logical name.

It is derived only while the physical name is empty or was itself derived — see Names of Your Own.
A term picked, a paste, or a replacement writes the logical name and what it derives as one undo step. On the canvas, the name derived as the editor closes is a step of its own after the typing: the first undo takes back the derived name, with the domain and type a term brought, and leaves the logical name you typed.

Nothing is derived again later on its own. A word's new Abbreviation, a term's new Physical Name or Domain, and new words or terms leave the names already derived as they are; use Derive Physical Names Again for that. A domain's Data Type is the exception: it reaches its columns at once.

From a Term

The Terms sheet is searched first for a term whose Term equals the logical name. Spaces at either end and the Unicode composition are ignored, but letter case counts, so customer number does not match Customer Number. Of two terms with one name, the first on the sheet counts.

• The term gives its Physical Name, or, while that is empty, the name its words compose, as below, if they compose one.
• For a column, the term also links its Domain while the Domains sheet lists it, which gives the column the domain's data type — see Linking a Column. It does so even when the term gives no name. A term with no domain unlinks nothing.
• A table takes only the term's physical name.

From the Words

With no matching term, the words compose the name:

1. The logical name is cut at spaces, _, -, ·, and /, and no word reaches across a cut.
2. Each piece is read from the left, taking at each point the longest text the Words sheet lists, as a Word, a synonym, or a forbidden word, and never going back. With the words 가입, 가입일, and 일자, 가입일자 reads as 가입일 and 자, leaving 자 unregistered, while 가입 일자 reads as 가입 and 일자. Letter case counts: member is not the word Member.
3. The physical name is each word's Abbreviation as typed, joined with _, such as ORD_DT. Of two words with one name, the first on the sheet gives the abbreviation. No length limit applies, and the Code Generator's Table Name Case and Column Name Case play no part.

It is all or nothing. If any stretch is a synonym, a forbidden word, or text no word lists, or a word has an empty Abbreviation, no name is composed, and the physical name stays as it is.

Names of Your Own

The dictionaries fill a physical name only while it is empty or was itself filled by them. The editor remembers this for each table and column, though nothing on the canvas shows it.

• A name you write becomes your own: typing in a name cell, replacing a table or column name with Find and Replace, pasting over a column's name cell, and a coding agent's rename. A logical name written later leaves it as it is, writes only the logical name, and links no domain.
• Clearing the name gives it back to the dictionaries, since an empty name is always derived.
• A logical name that derives nothing leaves the physical name and its state as they were: an empty name stays empty, and a derived name keeps its name and stays the dictionaries'.
• A new table or column starts with an empty name, so its first logical name derives it. A foreign key column a relationship adds is named after its key and counts as your own — see Foreign Key Column Names.
• The names a Schema SQL, GraphQL, DBML, or AML Import or Import and Add brings in, and the names of a diagram made by a release from before the Dictionary tab, count as your own, so a logical name written on them later derives nothing.
• A duplicated table, a column moved into another table by its drag handle, a paste of copied tables, and a JSON file's Import and Add keep each name's state. Columns pasted into selected tables as new columns count as your own, unless the pasted name is empty and their logical name derives one.
• Derive Physical Names Again gives a name of your own back to the dictionaries, an imported one included.

Two Columns, One Name

A column derives nothing when another column of its table already has the name it would derive. It keeps its name, its state, and its domain link, and only its logical name is written. The names are compared ignoring letter case and spaces at either end, as foreign key column names are.
When one change would give two columns the same name, as Replace All or Derive Physical Names Again over a table can, the first in the table takes it.
For example, a self-reference on members keyed by MBR_NO adds MBR_NO_2, which copies the logical name Member Number. Derive Physical Names Again would name it MBR_NO, which the key holds, so it keeps MBR_NO_2.
Tables have no such rule: two tables can derive the same name.

Derive Physical Names Again

Use Derive Physical Names Again after changing a word's Abbreviation or a term's Physical Name or Domain, since those changes rename nothing on their own. It gives the names it covers back to the dictionaries and derives them again by the rules above, names you wrote yourself included. It is offered on the ERD tab only, and not in a read-only editor:

• The table context menu, right after Focus on this table or Focus on selected tables. While the table is one of two or more selected tables, Derive Physical Names Again covers all of them, wherever on the table you right-click. Otherwise, a right-click on a column's row gives Derive Physical Name Again, for that column alone, and one on the header gives Derive Physical Names Again, for the table's name and every column.
• The canvas context menu, right after Name View: Derive Physical Names Again, for every table and column in the diagram. It is left out while the diagram has no table.
• Quick Search, as rows for the focused column, the focused table, the selected tables, and the whole diagram.

Each covered table and column that has a logical name is derived again. A column also takes the domain its term gives, and with it the domain's type, which spreads along the relationships while Relationship DataType Sync is on.
It leaves alone a table or column with no logical name, and a column whose new name another column of its table holds — see Two Columns, One Name. Each of these keeps its name and its state, your own or the dictionaries', and a column also keeps its domain link.
A logical name that derives nothing now leaves the physical name as it is, but gives it back to the dictionaries, so the next logical name you write derives it.
It is one undo step, however many names, links, and types it changes. It records no step only when it changes nothing, and giving a name of your own back to the dictionaries counts as a change, even when the name already holds what they give.

A term's own Physical Name is not composed again from changed words, so edit the term before deriving again.

Domains and Data Types

A column linked to a domain holds the domain's Data Type. Linking a column gives it that type at once.
When a domain's Data Type changes, whether edited, pasted, or imported on the Domains sheet or changed by a collaborator or a coding agent, every column linked to it takes the new type in the same step. While Relationship DataType Sync is on, each of those changes goes on along the relationships as a typed change does; while it is off, only the linked columns change.
Renaming a domain changes only the name the columns' Domain cells show.

A type change made on the Domains sheet is undone there, together with the column types it gave — see Dictionary Sheets.
An undo or redo on the ERD tab that links a column to a domain again, or brings back a column linked to one, gives the column the domain's type as it stands now. One that puts back another type unlinks the column.

Linking a Column

A column is linked to a domain by:

• Picking the domain in the column's Domain cell, which View Option → Domain shows, or, in Logical Names, in the type cell that stands for it — see The Domain Cell.
• A logical name whose term gives a domain — see From a Term. This happens only while the name is derived, so not for a name of your own or for a column whose derived name another column holds.
• Derive Physical Name Again or Derive Physical Names Again, which links the domain of each covered column's term.
• A relationship: a foreign key column copies its key's domain, along with its type and logical name — see Identifying Relationships. A key whose copy gets another type, as a serial key's does, hands on no domain.
• Duplicating its table, or moving it into another table by its drag handle, both of which keep the link. A paste of copied tables or columns keeps it only while this diagram lists the domain and the domain still has the copied column's type, so a copy from another diagram lands unlinked — see Copying/Pasting Tables and Memos. Import and Add links no column.
• A domain's name pasted into a Domain cell — see Copying/Pasting Columns.

Unlinking

• Typing a type into a linked column's DataType cell, or accepting a suggestion, unlinks the column as soon as its type differs from its domain's, and the column keeps the type you typed — see DataType Autocomplete. Typing the domain's type back does not link it again; pick the domain in its Domain cell instead. A paste over the DataType cell and a coding agent's type change unlink the same way.
• While Relationship DataType Sync is on, every column a type change reaches whose new type differs from its own domain's is unlinked in the same step, such as a foreign key linked to a domain of its own. Map Columns unlinks the columns it retypes the same way.
• None in the Domain cell unlinks the column and keeps its type.
• Changing or clearing a term's Domain, or removing a term, changes no column.

Removing a Domain

Removing a domain's row leaves its columns at their types, and their links stay. The columns show no domain, and the Checks sheet lists Unknown domain for each of them while the Domains sheet still holds a domain. Undoing the removal on the Domains sheet shows the links again.
While the domain is gone, typing a type other than the domain's into such a column unlinks it for good, and so does None in its Domain cell.
The file leaves those links out when it is saved, so the diagram reopens with those columns unlinked.

A term linked to a removed domain shows an empty Domain cell, marked as an Unknown domain. Delete or Backspace on the cell unlinks it — see A Term's Domain.

The Checks Sheet

The Checks sheet lists what the dictionaries' checks find in the dictionaries and in the diagram's tables and columns. The checks only read: they change nothing.
Each check runs only while the sheet it reads holds a row, and the word checks only while the Words sheet holds a word with a name, so a diagram with empty dictionaries lists nothing.

Its three columns are read-only:

• Type names the check — see What Each Check Finds.
• Target names a word, term, or domain by its name, a table by its name, and a column as table.column, the table and column named as the Name View names them, with unnamed where the view shows no name.
• Detail shows what the check found and, after →, what the dictionaries offer instead, several joined by commas, such as 멤버 → 회원. An empty side reads —, as in — → NUMERIC(15).

The words' checks come first, then the terms' and the domains', then table by table, each table followed by its columns.
The list follows every edit as it lands, a collaborator's and a coding agent's included, and every Name View switch.
The Checks button's count is red while any check is an error, amber while every one is a warning, and muted at none.

The rows cannot be edited, added, deleted, sorted, imported, or exported: the bar shows no tools, a right-click opens no grid menu, and Undo and Redo do nothing. Searching, selecting, and copying work as on the other sheets.

Enter or a double-click on a row goes to what it names:

• A word, term, or domain: the tab shows its sheet with the cell the check reads selected, the name or a term's Domain, and its row first in view. The sheet's search is cleared if it hides the row. A row since removed goes nowhere.
• A table or a column: the ERD tab comes up with the table selected alone and the cell the check reads focused. That is the name, the logical name, or the Domain cell, or, where the row does not draw that cell, the cell drawn in its place. The canvas takes the keyboard, and the view moves by the rule in Where a Jump Lands.

A double-click on a header, a column edge, a scrollbar, or below the rows goes nowhere.
A coding agent lists the same checks, in the same order, with erd_list_checks — see Dictionary Checks.

The other three sheets mark the cells the checks read, a name or a term's Domain, in red or amber text on a tinted fill:

• Red: a name another row of its sheet has, a term's Domain linking a removed domain, and a Term that uses a forbidden word.
• Amber: a Term that uses a synonym or an unregistered word.

A cell takes the louder of its levels. The marks show in a read-only editor too, and the Checks sheet itself is not colored. For the marks on the ERD canvas, see Marks on the Canvas.

What Each Check Finds

Type	Level	Finds	Detail, for example
Duplicate name	Error	A Word, Term, or Domain that another row of the same sheet also has. Each of the rows is listed.	Member
Unknown domain	Error	A column or a term still linked to a removed domain, while the Domains sheet holds a domain	The removed domain's name
Forbidden word	Error	A stretch of a logical name or of a Term that a word lists in its Forbidden Words	커스터머 → 고객
Unregistered word	Warning	A stretch of a table's or a column's logical name, or of a Term, that no word lists	자
Synonym	Warning	A stretch that a word lists in its Synonyms	멤버 → 회원
Physical name mismatch	Warning	A table or a column whose logical name matches a term but whose physical name is not the one the term gives	mbr_no → MBR_NO
Domain type mismatch	Warning	A column whose DataType is not the Data Type of the domain it links	INT → NUMERIC(15)

• Names are compared trimmed, with letter case counting, so Member and member are two names, and an empty name is never a duplicate. The sheets refuse an edit that would give two rows one name, so Duplicate name shows up only where two edits crossed, such as two collaborators naming rows at once, or in a file edited elsewhere.
• The word checks read a name as From the Words does, and list each different stretch once per name. Forbidden word and Synonym offer the word that lists the stretch.
• Physical name mismatch compares the name, trimmed, with the term's Physical Name, or with what its words compose while it has none, and says nothing while the term gives no name. A foreign key column that copied its key's logical name shows it whenever its own name differs from the term's, such as members_id → MBR_ID. Derive Physical Name Again fixes it, unless another column holds that name.
• Domain type mismatch never results from edits in one editor, since a linked column always ends at its domain's type. It can come when a collaborator links a column while another changes that domain's type, or in a file written elsewhere. Picking the domain again in the column's Domain cell gives the column its type, and None unlinks it. A link to a removed domain shows only as Unknown domain.

Marks on the Canvas

The ERD canvas writes a name the checks flag in its check's color:

• A logical name cell, in the table header or in a column's row, is red while its logical name uses a forbidden word, and otherwise amber while it uses a synonym or a word the Words sheet lacks.
• A physical name cell, in the header or in a row, is amber while the term its logical name matches gives another physical name.

Only those two kinds of cell are marked, never the Domain or DataType cell, so a column's Unknown domain and Domain type mismatch show on the Checks sheet alone.
A cell is marked only for what it draws: Physical Names can mark only physical names, Logical Names only logical names, and Both Names each cell by its own checks — see Name View. A blank cell keeps its placeholder color, and empty dictionaries mark nothing.
On the header of a table in a colored table group, a mark keeps its color only while it stays readable on the group's color; otherwise the name takes the header's own black or white.

The marks show on the ERD canvas, in Time Travel, and in the Auto Layout preview, in a read-only editor too. The Visualization tab, the two sides of the Diff Viewer, and an exported image draw none.
They follow every change as it happens, whether to a name or to a word, term, or domain.

Searching

The search field reads Search and shows on all four sheets, in a read-only editor too. Each sheet has its own search.
A row matches when one of the cells searched holds the text, ignoring letter case, the Unicode composition, and spaces around the search. A match never runs across two cells. The cells searched:

• Domains: Domain, Data Type, and Description.
• Words: Word, Abbreviation, English Name, and Description.
• Terms: Term, Physical Name, and Description.
• Checks: all three, as they read in the display language.

A new search puts the selection on the first cell and scrolls to the top.
While the field holds text, an x button with the tooltip Clear search empties it and leaves the caret in the field. Escape in the field empties it too.
In the field, Ctrl + Z (Windows/Linux) or ⌘ + Z (Mac) undoes your typing, not the sheet, and the row keys of Adding and Deleting Rows are text keys.
Add Row and the inserts clear the search first, since an empty row matches no search. An import keeps it.

Selecting and Moving

The grid selects as a spreadsheet does:

• A click selects a cell, and a drag selects a range, scrolling at the sheet's edge. Shift + click extends the selection from the active cell, and Ctrl + click (Windows/Linux) or ⌘ + click (Mac) adds another range.
• A column header selects its column, a row number its row, and the top-left corner every cell, as Ctrl + A (Windows/Linux) or ⌘ + A (Mac) does.
• On a Mac, Ctrl + click is a right-click — see The Grid's Menu.

With the keyboard:

• The arrow keys move one cell, and with Ctrl (Windows/Linux) or ⌘ (Mac) held go to the sheet's edge that way.
• Home and End go to the row's first and last cell, and with Ctrl or ⌘ held to the sheet's first and last cell.
• Page Up and Page Down move a screen of rows.
• Holding Shift with any of the keys above extends the selection instead.
• Tab and Shift + Tab move right and left, and Shift + Enter moves up. Enter alone edits the cell — see Editing Cells.

When rows go away under the selection, by an undo or a collaborator's removal, it is pulled back inside the rows left.

Drag a column header's right edge to resize the column, or double-click the edge to fit the column to the widest text in the rows in view. No key resizes a column.

Editing Cells

Typing on the selected cell replaces its value. Enter, F2, or a double-click edits the value as it stands, with the caret at its end.
A cell holds one line, so a tab or a line break pasted into it becomes a space.

• Enter writes the edit and moves down, and Tab writes it and moves right. With Shift, they move up or left.
• A click elsewhere, or the keyboard leaving the grid, writes it too.
• Escape cancels it.

An edit that leaves the value as it was writes nothing, and a written edit is one undo step of its sheet, however much it typed.
Delete or Backspace, with no edit open, empties every selected cell, as one step.
While an edit is open, the undo and redo keys undo and redo your typing in the cell, and the row keys type as text — see Keys on the Tab.
Every cell on Checks, and every cell in a read-only editor, is read-only and opens no edit.

When any row of an edit, a paste, or a clear is refused, nothing of it is saved, and an error toast says why, counting the rows and naming the first three:

• A name that another row of the sheet has, or that another row of the same paste writes: Nothing saved: 1 row shares a name with another row (Member), or Nothing saved: 2 rows share a name with another row (…). Names are compared trimmed, with letter case counting, so Member and member differ. An empty name is never refused.
• A term's Domain given as text that no domain on the Domains sheet has: Nothing saved: 1 row names an unknown domain (…).
• A row a collaborator removed while you edited it: Nothing saved: the row was removed.

A write can reach beyond its cell:

• A term given a Term while its Physical Name is empty gets the name its words compose, when they compose one — see From the Words. This happens once: the name is not composed again when the words change later.
• A domain's new Data Type retypes every column linked to the domain, in the same undo step — see Domains and Data Types.

A Term's Domain

Enter, F2, or a double-click on a term's Domain cell opens a list under the cell, or above it where there is more room, with a search field reading Domain.
The list holds None, then every named domain in the Domains sheet's order, and the highlight starts on the domain the term links, or on None. Typing a character on the cell opens the list already searched by it.

• Typing narrows the list to the domains whose names hold the text, ignoring case, drops None, and highlights the first match.
• ↑ and ↓ move the highlight, wrapping round at both ends.
• Enter or Tab picks the highlighted row, and a click the row clicked, as one undo step. The term links the very domain picked, even one of two domains with one name, and None unlinks it.
• Escape or a click outside closes the list without writing anything typed.

With no named domain on the Domains sheet, the field opens with no list.
Pasted text links the first domain listed with that name, empty text unlinks, and text equal to what the cell shows keeps the link as it is. Any other name is refused — see Editing Cells. Delete or Backspace on the cell unlinks the term, even from a removed domain.

Copying and Pasting

Ctrl + C (Windows/Linux) or ⌘ + C (Mac) copies the selection as tab-separated text, a line per row, the way spreadsheets write it, so it pastes into a spreadsheet as cells. A tab or a line break inside a value becomes a space.
Several ranges copy together only when they cover the same rows, side by side, or the same columns, one above another. Otherwise nothing is copied, and the error toast Nothing copied: the ranges share no rows or columns appears.

Ctrl + V (Windows/Linux) or ⌘ + V (Mac) pastes tab-separated text, such as cells copied from a spreadsheet, from the active cell down and to the right:

• Quotes are pasted as text, so a spreadsheet cell holding a line break does not paste as one cell.
• A single value fills every selected range.
• Rows past the last row are added as new rows, in order, so pasting into an empty sheet fills it.
• Columns past the last column are left out, and once the rest is saved the toast Columns past the last one weren't pasted says so.
• A paste of several cells then selects the block it wrote.

A paste is one undo step, rows it adds included, and is refused as a whole over a duplicate name or an unknown domain — see Editing Cells.
A copy, a paste, or a clear over more than a million cells does nothing, and the error toast Nothing done: the selection covers too many cells appears.
Cut, Ctrl + X (Windows/Linux) or ⌘ + X (Mac), does nothing on a sheet. Copying and pasting on this tab never reach the diagram, and while a cell edit is open they act on the text in the cell.

Adding and Deleting Rows

Add Row adds one empty row at the end of the sheet. It clears the search, selects the new row's first cell, and gives the grid the keyboard, so typing fills that cell.
Delete Rows removes every row the selection covers, in any column, and one undo puts them all back in their places. Its tooltip names the key that does the same.

With the keyboard in the grid and no cell being edited:

• Alt + Enter (Windows/Linux) or ⌥ + Enter (Mac) inserts as many empty rows below the selection as it covers.
• Alt + Shift + Enter (Windows/Linux) or ⌥ + Shift + Enter (Mac) inserts them above.
• Alt + Backspace or Alt + Delete (Windows/Linux), or ⌥ + Backspace or ⌥ + Delete (Mac), deletes the rows the selection covers.

They do what the grid menu's insert and delete rows do — see The Grid's Menu. An insert clears the search and puts the new rows right above the first row the selection covers, or right below its last. It selects them in the first column, the first of them active, and is one undo step.

These keys are the Add Column and Remove Column bindings of the Shortcuts tab, so a host that rebinds those with setKeyBindingMap moves them too. Alt + Shift + Enter is the Add Column key with Shift added, and has no row of its own there.
In an open edit and in the search field they are text keys, so on a Mac ⌥ + Backspace deletes a word. On Checks and in a read-only editor they do nothing. On this tab they never add or remove a table's column.
In Obsidian they reach the tab even where an Obsidian hotkey uses the same key.

A row never filled in, or one emptied of every value, shows on its sheet and reaches collaborators, but the file leaves it out, so it is gone when the diagram is opened again.

The Grid's Menu

Right-click a cell on Domains, Words, or Terms to open the grid's menu. A cell outside the selection becomes the selection first.
A right-click below the last row or right of the last column opens it on the selection as it stands, so an empty sheet opens it anywhere under the headers. The menu key or Shift + F10 opens it for the active cell.

• Insert 1 row above, Insert 1 row below, and Delete 1 row, each counting the rows the selection covers, as in Insert 3 rows above, with its key on the right — see Adding and Deleting Rows.
• Sort by <column>, ascending and Sort by <column>, descending, naming the column you right-clicked, as in Sort by Word, ascending — see Sorting. They are left out right of the last column.
• Import CSV and Export CSV, which do what the bar's buttons do — see Importing and Exporting CSV.

With no row showing, on an empty sheet or under a search that matches nothing, only Insert 1 row below, Import CSV, and Export CSV show.
Checks and a read-only editor open the browser's own menu instead, and so does a right-click on a header or a scrollbar.
The menu closes on a pick, on Escape, on a press elsewhere in the tab, when the selection or the sheet changes, and when the sheet's rows change, by your edit or a collaborator's.

Sorting

Sort by <column>, ascending and Sort by <column>, descending reorder the sheet's rows themselves, not only the view. The new order is saved in the file, seen by collaborators, and followed by Export CSV, and one undo puts the old order back.
Rows compare by the text the cell shows, a term's Domain by the domain's name, in the display language's alphabetical order, with numbers compared by value, so 2 comes before 10. Empty cells go last either way, and rows that compare equal keep their order. Sorting a sheet already in that order records nothing.
The search stays as it was, and so does the selection, whose cells now hold other rows.

Importing and Exporting CSV

Each of Domains, Words, and Terms comes in from a CSV file and goes out to one. These files are the Dictionary tab's own: the canvas context menu's Import and Export neither read nor write them — see Importing or Exporting Files.

Importing

Import CSV, on the bar or in the grid's menu, opens the browser's own file picker for .csv files, not the host's import dialog. Picking the same file again imports it again.
The file goes into the dictionary its header names, whichever sheet shows, and the tab then shows that sheet and gives its grid the keyboard. The sheet keeps its search, selection, and scroll, so rows the search hides, or rows added below the view, stay out of sight.
The file is read as UTF-8, with or without a byte order mark, or as EUC-KR when it is not valid UTF-8, as older public data standard files are.

How the rows land:

• Each row with a name updates the sheet's row of that name, in the fields the file has columns for. Columns the file lacks leave those fields as they are. Names are compared trimmed, with letter case counting.
• Names the sheet lacks are added after its last row, in the file's order.
• An import never removes a row.
• A row with an empty name is skipped. When the file holds a name twice, the later row wins and the earlier counts as skipped. Blank lines are not rows.
• A term's domain is found by its name, the first domain of that name, and a name no domain has leaves the term unlinked, so import the domains first.

One import is one undo step of that sheet.
The toast then reads, for example, 1 added · 1 updated · 2 skipped. A row the same as the sheet's counts as neither added nor updated, so importing the same file again changes nothing and records no step.
While the file's terms name domains the diagram lacks, a second line names the first three, as in 4 unknown domains: 연월일, 금액, 번호.
A file whose header matches no layout, or two, brings nothing in, and the error toast Couldn't read this file as words, terms or domains appears.

File Layouts

Each dictionary reads two layouts. The first is the editor's own, which Export CSV writes, with the field names as its header:

Dictionary	Header	A file needs
Words	name,abbreviation,englishName,description,synonyms,forbiddenWords	name, abbreviation
Terms	name,physicalName,domain,description,synonyms	name, physicalName
Domains	name,category,dataType,description	name, dataType

The columns may come in any order, a term's domain holds the domain's name, and any other column is ignored. Letter case counts in the header, and the cells are read exactly as written, spaces included.

The second is the layout of the files of Korea's public data standard (공통표준):

Dictionary	A file needs	Also read
Words	공통표준단어명	공통표준단어영문약어명 (abbreviation), 공통표준단어 영문명 (English name), 공통표준단어 설명 (description), 이음동의어 목록 (synonyms), 금칙어 목록 (forbidden words)
Terms	공통표준용어명	공통표준용어영문약어명 (physical name), 공통표준도메인명 (domain, by name), 공통표준용어설명 (description), 용어 이음동의어 목록 (synonyms)
Domains	공통표준도메인명, 데이터타입	공통표준도메인분류명 (category), 공통표준도메인설명 (description), 데이터길이, 데이터소수점길이

A domain's 데이터타입, 데이터길이, and 데이터소수점길이 join into one data type: NUMERIC(15,2), CHAR(8), the type alone where there is no length, such as DATE, and a scale of 0 left out, as in NUMERIC(15).
In these files the cells are trimmed, and other columns, such as the revision history, are ignored.
In either layout, spaces in a header name do not matter.

Exporting

Export CSV, on the bar or in the grid's menu, saves the sheet shown in the editor's own layout: every row in sheet order, whatever the search shows, with a term's domain written by its name, or left empty for a removed domain.
The file is UTF-8 with a byte order mark and CRLF line ends, and only a field holding a quote, a comma, or a line break is quoted. It opens in a spreadsheet and imports back to the same rows.
It is named <database name>-<timestamp>-words.csv, -terms.csv, or -domains.csv, and saved the way the editor's other exports are — see Exporting.

Checks has no export, and a read-only editor offers no import or export.

Undo and Redo

Domains, Words, and Terms each keep an undo history of their own, apart from the diagram's. On this tab, Ctrl + Z and Ctrl + Shift + Z (Windows/Linux) or ⌘ + Z and ⌘ + Shift + Z (Mac), and the toolbar's Undo and Redo, act on the sheet shown, and the toolbar shows no Time Travel. On Checks they do nothing.
The ERD tab's Undo never takes back a sheet's edit — see Dictionary Sheets.

Keys on the Tab

While the keyboard is in the tab, these editor shortcuts act — see Shortcuts:

• Search, Ctrl + K (Windows/Linux) or ⌘ + K (Mac), opens Quick Search.
• Toggle Explorer, Ctrl + F (Windows/Linux) or ⌘ + F (Mac), switches to the ERD tab and opens the Explorer on its search.
• Undo and Redo act on the sheet shown — see Undo and Redo.
• Add Column and Remove Column insert and delete rows — see Adding and Deleting Rows.

While a cell edit is open, a term's domain list included, Search and Toggle Explorer do nothing, so the text typed is not lost.

Every other editor shortcut stays in the tab and never reaches the canvas: Alt + N (Windows/Linux) or ⌥ + N (Mac) adds no table, and Delete and Backspace delete nothing on the canvas.
The grid's own keys act on the sheet instead: Enter edits the cell, Delete and Backspace clear the selected cells, Ctrl + A (Windows/Linux) or ⌘ + A (Mac) selects every cell, and Escape cancels an edit, closes the grid's menu, or empties the search field. Copy, cut, and paste stay on the tab as well — see Copying and Pasting.

Read-only Editors

In a readonly editor the tab opens and shows every sheet.
Switching sheets, searching, selecting, scrolling, resizing columns, and copying work as usual, and so does going to a check from Checks with Enter or a double-click. The checks' marks show too.

No cell takes an edit, a paste, or a clear, and every cell no check marks is drawn on the read-only fill.
Add Row, Delete Rows, Import CSV, and Export CSV are hidden, the row keys do nothing, and a right-click opens the browser's menu. An empty sheet shows no hint, and the toolbar shows no Undo or Redo.

Collaborators and the File

The diagram's file holds the words, terms, and domains in their sheet order, a sort's included, and leaves out a row that holds no value.
It also holds each table's and column's logical name, whether each physical name is the dictionaries', and each column's domain link while the domain is listed. The checks and the marks are worked out from these and never saved.

In a collaborative session, editors see each other's dictionary edits as they land, sorts and inserts included. Each keeps its own sheet, search, and selection, and its sheet histories hold only its own edits.
An edit or a paste onto a row a collaborator removed meanwhile saves nothing, and the toast Nothing saved: the row was removed appears.
An editor from a release before the Dictionary tab saves the diagram without its logical names, domain links, and dictionaries — see Mixing Releases.

A Schema SQL, GraphQL, DBML, or AML Import keeps the dictionaries, a JSON Import replaces them with the file's, and Import and Add brings in none — see Importing External Files.
A coding agent edits the words, terms, and domains, sets logical names and domains, derives physical names again, and lists the checks through the MCP server — see Dictionaries and Dictionary Checks. It has no tool for CSV files.